Evidence First

What Is a Life Well Spent?

Evidence First — Thu, 10 Sep 2026 16:18:44 GMT

There is a particular unease that often arrives not in crisis, but in ordinary moments. You are waiting for coffee to brew. You are scrolling in bed, half-aware that 40 minutes have passed. You are answering emails you will not remember next week. Then a thought breaks through: This is my life. It is passing while I am doing this.

The feeling is not exactly panic. It is closer to grief in advance, a sense that time is moving faster than attention can catch. Childhood summers once seemed endless. Now whole seasons disappear between bills, errands, meetings, notifications and sleep. The question that follows is ancient, but it feels newly urgent in an age of constant distraction: What should a person spend a life on?

No study can answer that completely. Science cannot tell you whom to love, what calling to pursue or which memories will matter most when you are old. Human lives are too varied for that. But research can help separate better bets from worse ones. It can show which investments tend to deepen well-being and which tend to disappoint. It can reveal, with appropriate humility, what people usually need not merely to survive, but to feel that their lives have been meaningfully lived.

The evidence points less toward a formula than toward a pattern. A life well spent is usually not built around constant pleasure, relentless achievement or private comfort. It is built around close relationships, a body and mind cared for well enough to participate in life, meaningful contribution and experiences that command our presence. Just as important, it requires resisting the temptations that make life busier without making it deeper.

The clearest finding may also be the least surprising: relationships matter enormously.

One of the best-known bodies of evidence on the subject comes from the Harvard Study of Adult Development, which has followed participants and their families for decades. Its recurring conclusion is simple enough to sound like folk wisdom: good relationships are strongly tied to happiness and health. Not fame. Not wealth. Not unbroken professional ascent. Relationships.

That finding does not mean extroverts win at life, or that everyone needs a packed social calendar. The point is not constant sociability. It is dependable connection. People tend to do better when they have others they can rely on, speak honestly with, care for and be cared for by. A weekly dinner with an old friend, a marriage repaired through attention, a sibling called regularly, a neighborhood where people notice one another: these are not sentimental luxuries. They are part of the architecture of a good life.

This matters because modern life often treats relationships as whatever remains after the “important” obligations are done. Work gets the best hours. Screens fill the gaps. Loved ones get the leftovers: tired attention, delayed replies, distracted evenings. Yet the evidence suggests this order is often backward. If close relationships are among the strongest predictors of long-term well-being, then protecting time for them is not indulgent. It is rational.

Health belongs in the same category. It is easy to discuss health in the language of optimization: steps, macros, sleep scores, body-fat percentages, supplements, longevity hacks. But the deeper reason to care for the body is more basic. Health is what allows a person to keep entering the life they value.

The evidence for physical activity is broad and unusually consistent. Regular movement is associated with lower risk of premature death, cardiovascular disease, diabetes and some cancers. It is also linked to better sleep, mood, cognitive function and daily capacity. But a meaningful life does not require becoming a fitness obsessive. The humane version of the evidence is gentler: move enough, sleep enough and care for your body enough that your future self is not unnecessarily robbed of participation.

A walk with a friend, a bike ride to work, dancing in the kitchen, lifting weights, gardening, hiking, playing with children: these are not merely ways to avoid disease. They are ways of staying available to life. A body is not just an object to improve. It is the instrument through which every conversation, embrace, meal, journey and act of service is experienced.

The same is true of mental health. Chronic stress, loneliness and exhaustion narrow a life. They make the world smaller. They make people less patient, less generous, less curious and less able to enjoy what they already have. Rest is therefore not the enemy of a meaningful life. It is one of its conditions. A person who is always depleted may be technically alive, but much of life becomes inaccessible.

Then there is the question of meaning. Pleasure matters. Comfort matters. But research on well-being has repeatedly suggested that people tend to need something more than feeling good. They need to feel that their lives count for something.

This is where contribution enters. Studies of prosocial behavior, a term researchers use for actions intended to benefit others, generally find a positive relationship between helping and well-being. The evidence is not a commandment. It does not prove that every person must volunteer in a formal organization, become a parent, enter a helping profession or devote themselves to a grand cause. But the broader pattern is persuasive: human beings often flourish when they are useful to someone or something beyond themselves.

That usefulness can be quiet. It might look like caring for an aging parent, mentoring a younger colleague, raising children, doing honest work, showing up for a friend in grief, building something that serves a community or simply being the person others can count on. In a culture that often equates meaning with scale, it is worth remembering that significance is not the same as visibility. Some of the most meaningful lives are not impressive from a distance. They are deeply felt up close.

Experiences also appear to matter more than we often assume. Research comparing experiential and material purchases has generally found that people derive more lasting satisfaction from experiences than from possessions, especially when those experiences are shared or become part of a person’s identity. A useful object can improve life, of course. A safe car, a comfortable home, a good pair of shoes: these things matter. But the evidence gives us reason to be skeptical of the endless pursuit of upgrades whose main purpose is status.

The vacation remembered for years, the concert with a friend, the ordinary Sunday meal that becomes a family ritual, the class that opens a new world, the afternoon spent outdoors: these often become part of the story people tell about who they are. Possessions can fade into the background. Experiences, especially shared ones, can become memory, identity and connection.

This may explain why time can feel so strangely lost when it is consumed by passive entertainment. There is nothing wrong with rest, television, games or scrolling in moderation. A good life need not be stern. But many people recognize the difference between leisure that restores and distraction that merely anesthetizes. One leaves you more yourself. The other leaves you wondering where the evening went.

That distinction brings us to attention, perhaps the least visible but most important currency of all. We often speak as if time is spent only by the hour: eight hours working, one hour commuting, two hours with family. But attention determines whether we actually inhabit those hours. A parent can be physically present and mentally absent. A couple can sit together while each disappears into a phone. A person can walk through a beautiful place and barely see it.

The research on social media and well-being is complicated. It does not support the simplistic claim that all digital life is harmful. Online connection can be real connection. Communities formed through screens can sustain people who might otherwise feel alone. But evidence increasingly suggests that heavy, passive or comparison-driven use can be associated with lower well-being, particularly among young people. The mechanism is not mysterious. A tool that constantly redirects attention toward comparison, outrage, performance and novelty can make ordinary life feel insufficient.

The problem is not technology alone. It is the broader human temptation to be absent from one’s own life. We can be absent through ambition, worry, fantasy, resentment, entertainment or fear. We can trade our attention for things we do not even value. And because attention feels weightless moment by moment, we may not notice the cost until months or years have passed.

This is why the common advice to “live every day like it is your last” is both powerful and flawed. Taken literally, it can become frantic or irresponsible. Most people should still save money, answer emails, do laundry, attend appointments and plan for a future they may well get to live. The better lesson is not that every day should be extraordinary. It is that ordinary days are the substance of a life.

A life is not mostly made of mountaintop revelations. It is made of Tuesdays. It is made of how you speak to the person making breakfast beside you. It is made of whether you go outside when the weather is good. It is made of whether you call back, repair the rupture, notice your child asking for attention, do the work with care and let the small pleasures register.

The evidence also warns against a common substitute for meaning: status. Ambition can be healthy. Work can provide mastery, purpose, income, friendship and identity. There is nothing noble about avoidable financial stress, and there is nothing shallow about wanting to use one’s talents well. But when status becomes the organizing principle of life, the returns are unreliable.

The problem with status is that it is comparative. There is always someone richer, more admired, more productive, more beautiful, more free. Status also tends to move the goalpost. What once seemed like success quickly becomes normal, and a new level appears necessary. A person can spend decades climbing toward a feeling that keeps receding.

The evidence does not say achievement is meaningless. It says achievement is a poor substitute for love, health, contribution and presence. A career triumph is real. But if it costs every dinner, every friendship, every quiet morning, every unmeasured joy, the bargain may be worse than it looked at the time.

There are important qualifications. Well-being research often deals in averages, and averages do not dictate individual lives. Some people find deep meaning in solitude. Some flourish in demanding careers. Some experience family not as refuge but as pain. Some forms of service become exploitative when they erase the self. Poverty, discrimination, illness, grief and unsafe environments can constrain choices in ways that lifestyle advice often ignores. The evidence should not be used to scold people for failing to thrive under burdens they did not choose.

Nor should it be reduced to a moral checklist. The point is not to optimize every hour, convert friendship into a longevity strategy or turn meaning into another achievement project. The point is to ask what kinds of time tend to repay human beings at the deepest level.

That question becomes especially important because many of the worst uses of time do not feel bad immediately. They feel easy. Comparison feels stimulating. Overwork feels responsible. Avoidance feels relieving. Scrolling feels harmless. Buying feels rewarding. Saying yes to everything feels virtuous. But a life can be drained by activities that are pleasant, respectable or convenient in the short run.

The better uses of time are often modest and sometimes effortful. Relationships require patience. Health requires maintenance. Contribution requires inconvenience. Presence requires resisting distraction. Meaning often asks something of us before it gives something back.

Still, the picture that emerges from the evidence is hopeful. It suggests that a good life is not hidden behind rare genius, perfect discipline or dramatic transformation. Much of it is available in ordinary choices repeated over time: to be close rather than merely connected, useful rather than merely busy, active rather than numb, present rather than entertained, rested rather than constantly depleted, guided by values rather than appearances.

No study can tell a person exactly what to do with a Saturday afternoon. But the evidence can help sharpen the question. Not “How do I fit more into my life?” but “What makes my life more worth inhabiting?” Not “How do I become impressive?” but “What would I be grateful to have given myself to?” Not “How do I avoid wasting time forever?” but “What deserves my attention now?”

Time will pass either way. That is the hard fact underneath the question. The days do not wait until we have clarified our values. They move through whatever is already on the calendar, whatever habit is already installed, whatever screen is already open, whatever fear is already making the decision.

The good news is that a life does not have to be rebuilt all at once to become more honest. Attention can return. A friendship can be revived. A body can be walked around the block. A conversation can be entered fully. A person can do one useful thing for someone else. A meal can be tasted. A sunset can be noticed. A phone can be put down.

The evidence does not promise that these things will make life painless. They will not. But it strongly suggests that they are among the things least likely to be regretted. In the end, the best-supported answer to how we should spend our short time on earth may also be the most human one: with people we love, in bodies we care for, doing things that matter, while we are still here to notice.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Relationships and long-term well-being

Claim or topic:

Close relationships are strongly associated with happiness, health, and long-term well-being.

Source:

Harvard Gazette, “Things money can’t buy, like happiness and better health”

Source type:

Reputable journalism summarizing long-running academic research.

What it supports:

This source summarizes findings from the Harvard Study of Adult Development, a decades-long study often cited for the conclusion that good relationships are strongly linked with happiness and health.

Important caveat:

The article is a public-facing summary, not the full underlying dataset or a systematic review. The Harvard Study is influential, but no single cohort study can fully answer what makes every individual life meaningful.

2. Social connection and happiness

Claim or topic:

Social connection and perceived social support are associated with higher life satisfaction and well-being.

Source:

World Happiness Report 2025, “Connecting with Others: How Social Connections Improve the Happiness of Young Adults”

Source type:

Expert organization / research report.

What it supports:

This source supports the article’s claim that social connection is not merely sentimental, but is consistently linked with well-being in population-level research.

Important caveat:

Much of this evidence is observational. It can show strong associations, but it does not always prove that social connection alone causes higher well-being in every case.

3. Physical activity, health, and daily functioning

Claim or topic:

Regular physical activity is associated with lower risk of major diseases and premature death, and with better sleep, mood, cognitive function, and daily capacity.

Source:

Physical Activity Guidelines for Americans

Source type:

Government evidence-based guidance.

What it supports:

This source supports the article’s claim that movement is not just an “optimization” project. It is one of the better-supported ways people protect long-term health and preserve their ability to participate in life.

Important caveat:

The article uses this evidence at a broad level. It does not offer individualized medical advice, and physical activity needs vary by age, disability, health status, and circumstance.

4. Helping others and well-being

Claim or topic:

Prosocial behavior, meaning actions intended to benefit others, is generally associated with higher well-being.

Source:

American Psychological Association summary of research on prosocial behavior and well-being

Source type:

Academic research.

What it supports:

This source supports the article’s claim that contribution, usefulness, and helping others are often linked with well-being.

Important caveat:

The evidence does not prove that every form of helping improves every person’s life. Context matters. Helping can become draining or exploitative when it erases the helper’s own needs.

5. Experiences versus possessions

Claim or topic:

Experiential purchases often produce more lasting satisfaction than material purchases, especially when experiences are shared or become part of identity.

Source:

PNAS / PMC article on experiential and material consumption

Source type:

Academic research.

What it supports:

This source supports the article’s claim that experiences often repay people differently from possessions, particularly through memory, identity, and social connection.

Important caveat:

The evidence does not mean possessions never matter. Useful material goods can improve life substantially, especially when they increase safety, comfort, mobility, or financial stability.

6. Social media, attention, and well-being

Claim or topic:

Heavy, passive, or comparison-driven social media use may be associated with lower well-being, especially among young people.

Source:

World Happiness Report 2026, “International Evidence on Happiness and Social Media”

Source type:

Expert organization / research report.

What it supports:

This source supports the article’s cautious claim that digital life is not automatically harmful, but some forms of social media use may be linked with lower well-being.

Important caveat:

The evidence is complicated. Online connection can be valuable, and the relationship between social media and well-being varies by person, platform, pattern of use, and social context.

7. “Top regrets of the dying” and limits of anecdotal evidence

Claim or topic:

Popular claims about the “top regrets of the dying” should be treated cautiously because they are not based on a systematic scientific study.

Source:

Absolutely Maybe / PLOS Blogs, “Why do scientists cite the ‘Top 5 Regrets of the Dying’?”

Source type:

Analysis / science commentary.

What it supports:

This source supports the article’s caveat that hospice-regret claims can be emotionally resonant but should not be treated as rigorous population-level evidence.

Important caveat:

The source critiques the evidentiary basis of a popular claim. It does not show that the themes are false, only that they should not be overstated as settled scientific findings.

8. Individual differences and structural limits

Claim or topic:

Well-being research often deals in averages, and averages do not dictate individual lives. Poverty, illness, discrimination, grief, unsafe environments, and family pain can constrain people’s choices and affect health and well-being.

Source:

World Health Organization, “Social determinants of health”

Source type:

Expert organization.

What it supports:

This source explains that health and well-being are shaped not only by individual choices, but also by the conditions in which people are born, grow, live, work and age, as well as access to power, money and resources.

Important caveat:

This source supports the structural-limits caveat at a broad level. It does not directly address every example in the article, such as grief or family pain, so the claim should remain framed as a caution rather than a precise causal finding.

How to read this evidence

This article is the author’s analysis. The sources above are provided so readers can see where the factual claims come from and judge the evidence for themselves. Some sources support direct facts, while others provide context, estimates, or background evidence.

Corrections and updates

If a factual error is identified, this post will be corrected in the web version with a dated note explaining the change. Because email versions cannot be edited after sending, the web version should be treated as the current version.

Subscribe now

Climate Science Aged Better Than the Debate Around It

Evidence First — Fri, 04 Sep 2026 00:05:09 GMT

In October 2018, a major climate report acquired a much simpler public meaning: We had 12 years left.

The underlying report, issued by the Intergovernmental Panel on Climate Change, did not say the world would end in 2030. It examined what would be required to limit global warming to 1.5 degrees Celsius above preindustrial levels and concluded that doing so would require global carbon dioxide emissions to fall sharply by 2030 and reach roughly net zero around the middle of the century. That was an urgent finding, but it was not a prediction that catastrophe would begin on a fixed date.

For years, climate change was often presented through two competing stories. One said that the warnings were exaggerated, the models were unreliable and the science was too uncertain to justify much concern. The other suggested that humanity was approaching a hard deadline after which disaster would become unavoidable. With decades of climate projections now behind us, both stories can be tested against what actually happened.

The result is more interesting than either side suggested. The central science held up better than many critics claimed, but the consequences were also less like the sudden apocalypse that some public rhetoric implied. What the science actually described was a growing risk problem, one in which additional warming increases the likelihood and cost of serious harm. That story was harder to turn into a slogan, but it also aged better.

What actually held up

The first question is whether the basic science was reliable. On that point, the record is strong. Greenhouse gases trap heat in the atmosphere. Human activity, especially the burning of fossil fuels, has sharply increased their concentration. The planet has warmed. Sea level has risen. Extreme heat has intensified. Over time, the evidence increasingly established human activity as the dominant cause of recent global warming.

The World Meteorological Organization reported that 2025 was about 1.43 degrees Celsius warmer than the average from 1850 to 1900, a period commonly used as a baseline for preindustrial conditions. The years from 2015 through 2025 were the 11 warmest years in the observational record. A single unusually warm year does not mean the world has permanently crossed a particular long-term threshold. Natural variation still pushes individual years up and down, but the long-term warming trend is clear.

Older climate models also performed better than many people remember. A 2020 study in Geophysical Research Letters examined climate models published between 1970 and 2007 and compared their projections with subsequent temperature observations. Most performed reasonably well at projecting global temperature change. When researchers accounted for differences between the greenhouse gas emissions the models assumed and the emissions that actually occurred, their performance improved further. The study found no systematic tendency among those models to overestimate warming.

That does not mean climate models predict everything well. Global average temperature is easier to project than rainfall, storms or agricultural conditions in a particular place decades into the future. Some old forecasts were too warm. Others were too cool. But the core relationship between greenhouse gases and global temperature held up much better than the public memory of failed climate predictions often suggests.

Getting the temperature trend broadly right was important. But temperature was never the real end point of the warning. The question was what a warmer climate would do to the systems people depend on.

What warming actually looks like

If climate change was serious, some public rhetoric implied, then it had to look like a civilization-ending catastrophe. The science never required that conclusion.

For most people, climate change will not arrive as one giant event labeled “global warming.” It is more likely to appear as a growing burden on systems people already depend on: power grids, water supplies, roads, farms, insurance markets, public health systems and coastal infrastructure. Climate change also does not cause every heat wave, drought, flood or wildfire. Weather extremes have always occurred, and local conditions still matter. What warming can do is change the probability or severity of some events.

Extreme heat is one example. A hotter day may sound like an inconvenience. Enough additional extreme heat becomes a public health and infrastructure problem. Outdoor work becomes more dangerous. Electricity demand rises as more people need cooling. Hospitals, schools and workplaces have to prepare for conditions that occur more often.

Food systems face a similar kind of pressure. Climate change does not mean crops fail everywhere or that the world suddenly runs out of food. Some regions may benefit for a time from longer growing seasons or other changes. But farming depends on temperature, rainfall and water availability. Severe heat and drought can reduce harvests, livestock productivity and farm labor.

The more concerning situations are the ones in which several stresses overlap. A major growing region suffers drought while another has a poor harvest. Food prices rise. A country that depends heavily on imports has less ability to absorb the shock. At the same time, heat may reduce worker productivity and increase electricity demand. Climate risks can compound in other ways, too. Drought can reduce hydropower just as more people need air conditioning. Crops can suffer while wildfire smoke adds another health burden. Flooding can damage infrastructure at the same time insurance becomes more expensive.

Researchers call these compound and cascading risks. No single event needs to overwhelm a country for the combination to become severe. Imagine a coastal city that is already hot in summer. Over time, it faces more frequent heat waves, heavier demand on its power system, more expensive cooling, greater flood risk during storms and rising insurance costs in low-lying neighborhoods. None of those problems is catastrophic on its own. Together, they force households, businesses and local governments to spend more just to maintain roughly the same level of safety and reliability.

This is the realistic climate threat. It is less cinematic than the rhetoric surrounding it, but it is not trivial. The harm comes from many pressures accumulating across places and decades. The science did not need to predict extinction to describe a serious problem.

Why a few degrees matter

The temperature numbers used in climate discussions can sound strangely small. The difference between 1.5 and 2 degrees Celsius may seem trivial in ordinary conversation, but global temperature is not like the thermostat in a living room. It is an indicator of how much the entire climate system has shifted.

Coral reefs provide one of the clearest examples. The IPCC has assessed that warm-water coral reefs are projected to decline by a further 70 to 90 percent at 1.5 degrees Celsius of global warming, with losses greater than 99 percent at 2 degrees. Those estimates carry uncertainty, and individual reefs will not respond identically. But they show how half a degree in the global average can translate into a very large difference in risk for an ecosystem.

Species extinction is another irreversible concern. The IPCC has estimated that among the terrestrial species it assessed, roughly 3 to 14 percent could face a very high risk of extinction at 1.5 degrees Celsius of warming, with the range increasing at higher temperatures. That does not mean all of those species will disappear. It means the probability of extinction becomes very high for a growing share of the species assessed. Species confined to a mountaintop, island or narrow climatic range are especially vulnerable because they can eventually run out of suitable habitat.

Sea-level rise shows the same logic over a much longer timescale. The immediate concern is not that major coastal cities suddenly disappear beneath the ocean. It is that higher seas make flooding more frequent, storm surges more damaging and coastal protection more expensive. Sea-level rise also does not stop in 2100. Oceans and ice sheets respond slowly, so sea levels can continue rising for centuries even after temperatures stabilize. Some places will defend themselves for a long time with seawalls, pumps and other infrastructure. Others may eventually find retreat more practical than permanent defense.

The United Nations Environment Programme estimated in 2025 that policies then in place would put the world on a course of roughly 2.8 degrees Celsius of warming this century. Full implementation of countries’ existing climate pledges would lower that projection to roughly 2.3 to 2.5 degrees. Those are projections, not destiny. Policies change. Technologies improve. Energy systems evolve.

Nor is 2.8 degrees a scientific code for “the world ends here.” Its significance is that many risks increase as warming rises. Heat extremes intensify. Ecosystems face greater losses. Some forms of adaptation become harder. The chances of some irreversible changes increase.

This is why the science speaks in gradients rather than deadlines. A world at 1.7 degrees is not “safe” while a world at 2 degrees is “doomed.” But the risks are not identical either. The same is true of 2.5 and 3 degrees. There is no single cliff. There is a worsening slope.

The existence of that slope also means the future is not predetermined. How much people suffer depends not only on temperature, but also on how well societies respond. Flood barriers, better drainage, water management, cooling, stronger buildings, heat-warning systems and more resilient infrastructure can all reduce harm. A wealthy society that invests heavily in adaptation can experience the same physical hazard very differently from a poor society that lacks those protections.

But adaptation has limits. Protecting a city from occasional flooding is one problem. Protecting it from steadily rising seas for generations is another. Adjusting farming practices to modest changes in weather is easier than adapting indefinitely to increasing heat or water stress. Saving an ecosystem becomes impossible once the species that make it up are gone.

Each additional increment of warming can therefore make adaptation harder and more expensive. Adaptation can reduce harm. Emissions cuts can limit how much additional warming occurs. Neither makes all risk disappear.

Mainstream climate assessments also do not establish human extinction as the likely result of warming expected this century. Claims that climate change will probably eliminate humanity go beyond the central scientific evidence. But “not human extinction” is an extremely low standard for deciding whether a problem matters.

We do not judge a disease by asking whether it will wipe out the species. A recession can cause enormous hardship without permanently destroying the economy. A war does not need to end civilization to devastate millions of lives. Climate change should be judged by the same ordinary standard. How much harm is likely? Who bears the costs? Which losses cannot be reversed? How much can be prevented? How much worse does the problem become as warming rises?

For humanity as a whole, the most realistic danger is not extinction. It is a hotter, more expensive, more disruptive and less biologically rich world, with some of the greatest burdens falling on people and places least able to absorb them.

That conclusion sat awkwardly between the two stories that dominated the debate. Climate change was far more serious than “the science is too uncertain to worry about,” but more measured than “we are approaching a fixed date after which the world is doomed.” That was not a political compromise. It was where the evidence pointed.

How the debate went wrong in both directions

The problem was that this kind of conclusion is difficult to turn into a slogan.

For years, American journalism often made climate science look more divided than it was. A 2004 study of coverage in The New York Times, The Washington Post, the Los Angeles Times and The Wall Street Journal found that the journalistic habit of giving opposing sides similar weight often created a misleading impression about the state of the science.

The problem became known as “balance as bias.” When the great majority of relevant research points in one direction, giving a small number of dissenters equal prominence can make uncertainty appear much larger than it really is. A later study published in Nature Communications in 2019 found a related pattern. Researchers compared the media visibility of hundreds of climate scientists with hundreds of prominent climate contrarians across roughly 100,000 media articles. The contrarians received disproportionately high attention relative to their scientific standing.

For years, one prominent distortion made the science look less settled than it was. At other times, another made it sound more absolute than it was. As scientific confidence increased and public concern grew, complicated findings were sometimes compressed into slogans that sounded more certain and more catastrophic than the assessments themselves.

The 2018 IPCC report is a useful example. Its actual message was that limiting warming to 1.5 degrees required rapid emissions reductions. Public discussion often transformed that finding into a hard deadline after which catastrophe would become unavoidable.

The urgency was real. The cliff was not.

A deadline is easier to communicate than a probability distribution. “Twelve years left” is more memorable than saying that each additional increment of warming raises risk. But the second statement is much closer to the science. Once the countdown became part of public memory, every missed milestone could also be interpreted as evidence that climate science itself had failed.

The science was simplified in opposite directions: sometimes into excessive uncertainty, and at other times into excessive certainty. Strip away both simplifications, and the record becomes clearer.

What aged well

Scientists did not get everything right. Individual scientists have made poor predictions. Models have limitations. Estimates change as new evidence arrives. Local climate impacts are much harder to predict precisely than the basic global temperature response to greenhouse gases.

But the central framework held up. The planet warmed substantially. Human influence emerged clearly as the dominant cause. Older models generally captured the broad temperature response to rising greenhouse gases. Sea level rose. Extreme heat intensified. Evidence accumulated that many risks increase as warming rises.

What aged poorly were the binary stories built around that science. The skeptical version said the science was too uncertain and the predictions too unreliable to take seriously. The apocalyptic version suggested that the science pointed to a single deadline, a hard threshold or a likely end to civilization. Neither version aged especially well.

The science described something less dramatic and more durable: a range of outcomes. A world that warms by 1.7 degrees is generally preferable to one that warms by 2.5. A world at 2.5 degrees is generally preferable to one at 3. There is no point at which preventing additional warming suddenly stops mattering.

The public debate wanted a verdict: hoax or catastrophe, safe or doomed, deadline met or deadline missed. The science kept describing a risk curve. More warming meant more danger. Less warming meant less danger. Adaptation could reduce some harms. Some losses would still be irreversible. Uncertainty remained, but uncertainty did not erase the direction of the evidence.

That was always a harder story to tell. It was also the one that aged best. The science was less apocalyptic than some of the rhetoric surrounding it suggested. It was more consequential than its loudest critics allowed.

And it aged better than the debate around it.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Human-caused warming and rising climate risk

Claim or topic:

Human activity is the dominant cause of recent global warming, and climate risks generally increase as warming rises.

Source:

IPCC Sixth Assessment Report, Synthesis Report

Source type:

Expert organization / scientific assessment.

What it supports:

The IPCC concludes that human activities have unequivocally caused global warming and that risks to people and ecosystems increase with additional warming.

Important caveat:

The IPCC synthesizes a large scientific literature rather than reporting a single experiment. Confidence varies by specific impact, region and level of warming.

2. Recent global temperatures

Claim or topic:

Recent years have been the warmest in the observational record, with 2025 about 1.43 degrees Celsius above the 1850 to 1900 average.

Source:

World Meteorological Organization, State of the Global Climate 2025

Source type:

Expert organization / observational climate data.

What it supports:

The WMO provides the global temperature estimates used in the article and documents the long-term warming trend.

Important caveat:

A single unusually warm year is not the same as permanently crossing a long-term warming threshold. Year-to-year natural variability still matters.

3. How well older climate models performed

Claim or topic:

Older climate models generally did a reasonably good job projecting subsequent global temperature change.

Source:

Hausfather et al., “Evaluating the Performance of Past Climate Model Projections,” Geophysical Research Letters

Source type:

Academic research.

What it supports:

The study compared climate projections published from 1970 to 2007 with what later occurred. It found that most models performed reasonably well, especially after accounting for differences between assumed and actual greenhouse gas forcing.

Important caveat:

This finding applies most directly to global temperature projections. Climate models are less precise for many local and regional outcomes.

4. Risks at 1.5, 2 and higher degrees of warming

Claim or topic:

Relatively small differences in global average warming can produce large differences in risk, including for coral reefs, species and other ecosystems.

Source:

IPCC Special Report on Global Warming of 1.5°C and IPCC Sixth Assessment, Working Group II

Source type:

Expert organization / scientific assessment.

What it supports:

The assessments support the article’s discussion of sharply increasing risks to warm-water coral reefs, rising extinction risk for some species, and broader increases in ecological and human vulnerability as warming rises.

Important caveat:

These are risk estimates, not guarantees that a specific percentage of species or ecosystems will disappear. Outcomes vary by region, species and adaptation capacity.

5. Current-policy warming projections

Claim or topic:

Under policies in place in 2025, the world was projected to warm by roughly 2.8 degrees Celsius this century, with lower warming if national pledges are fully implemented.

Source:

United Nations Environment Programme, Emissions Gap Report 2025

Source type:

Expert organization / estimate and analysis.

What it supports:

The UNEP report provides the projected warming ranges used to explain why the difference between roughly 2, 2.5 and 3 degrees matters.

Important caveat:

These are scenario-based projections, not fixed predictions. Future policies, technology, energy use and emissions can change the outcome.

6. Media “balance” and climate science

Claim or topic:

Some earlier U.S. climate coverage made the science appear more evenly divided than the scientific literature justified.

Source:

Boykoff and Boykoff, “Balance as Bias,” Global Environmental Change

Source type:

Academic research.

What it supports:

The study examined major U.S. newspaper coverage from 1988 to 2002 and found that journalistic norms of balance often gave disproportionate weight to views questioning human-caused climate change.

Important caveat:

The study covers a specific group of newspapers and an earlier period. It should not be generalized to all climate journalism or all media eras.

7. Media visibility of climate contrarians

Claim or topic:

Climate contrarians received more media attention than their scientific standing alone would predict.

Source:

Petersen, Vincent and Westerling, Nature Communications

Source type:

Academic research.

What it supports:

The study compared hundreds of prominent climate scientists and contrarians across a large media sample and found disproportionate visibility for contrarian voices relative to their scientific authority.

Important caveat:

Media visibility is not the same as influence on public opinion, and the imbalance varied across different types of media.

8. The 2018 “12 years” framing

Claim or topic:

The IPCC’s 2018 emissions pathway was widely simplified in public discussion into a fixed 12-year climate deadline.

Source:

IPCC Special Report on Global Warming of 1.5°C

Source type:

Primary scientific assessment.

What it supports:

The IPCC report supports the underlying claim that limiting warming to 1.5 degrees required steep emissions reductions by 2030 and roughly net-zero carbon dioxide emissions around midcentury.

Important caveat:

The report did not say catastrophe would begin in 2030 or that climate action would become pointless after that date. A separate media source would be useful if the published article wants to document a specific “12 years left” headline or quotation.

How to read this evidence

Corrections and updates

Subscribe now

The Psychology of Donald Trump

Evidence First — Tue, 01 Sep 2026 23:36:00 GMT

Donald Trump is easiest to misunderstand when he is treated as a riddle with a hidden answer. Is he acting? Is he sincere? Is the outrage real? Is the charm fake? Is the public persona a performance masking a more ordinary private man, or is the private charm merely another performance?

The evidence points to a less tidy answer. Trump is not best understood as two separate people, the raging rally figure and the affable private host. He is better understood as one unusually consistent personality system operating in different theaters.

In public, he tries to win the crowd. In private, he tries to win the person in front of him. The methods change. The goal does not.

That goal is control: of attention, of the story, of the emotional temperature, of who is seen as strong and who is seen as weak. Around Trump, politics often becomes interpersonal drama. Allies are loyal or disloyal. Opponents are not merely wrong but pathetic, corrupt, stupid or treacherous. Defeat is not just defeat. It is humiliation. Victory is not simply success. It is vindication.

None of this is a clinical diagnosis. That needs to be said plainly. The American Psychiatric Association’s Goldwater Rule cautions psychiatrists against diagnosing public figures they have not personally examined and who have not consented to evaluation. Diagnosis requires more than speeches, interviews, memoirs, courtroom documents and televised behavior. It requires clinical access, context, history and evidence of impairment across a person’s life.

But public evidence can still tell us something. It can reveal patterns. Trump’s patterns are unusually visible because he has spent decades performing himself: in tabloids, on television, in business branding, in campaign rallies, in courtrooms, in the Oval Office and on social media.

The pattern begins with status.

Trump’s language has long sorted the world into winners and losers, strength and weakness, dominance and disgrace. He does not merely disagree with rivals. He diminishes them. He does not merely praise himself. He inflates himself. Everything tends toward rank. Who is up? Who is down? Who is laughing? Who has been made to look small?

This is not just the impression of critics. In a peer-reviewed study that asked Trump supporters and Hillary Clinton supporters to rate Trump’s personality, the two groups sharply disagreed about whether his traits were good or bad. But they overlapped on some basic observations: he was widely seen as highly assertive and low in modesty. Supporters often interpreted those traits as confidence, strength and success. Opponents saw hostility, arrogance and lack of warmth.

That distinction matters. Trump’s psychology cannot be separated from the audience that completes it. The same gesture can read as bullying to one group and courage to another. The same insult can sound cruel to one listener and cathartic to another. The same factual exaggeration can appear disqualifying to critics and irrelevant, even thrilling, to supporters who think the larger system is already dishonest.

Trump’s political gift has been to make his own emotional needs feel like collective needs. His grievances become his supporters’ grievances. His enemies become their enemies. His humiliation becomes their humiliation. His revenge becomes their vindication.

That helps explain why conflict is not incidental to Trump’s public life. It is central to it. Many politicians try to survive controversy. Trump often uses controversy as fuel. A shocking statement forces attention. Attention confirms significance. Criticism proves persecution. Persecution strengthens loyalty. The cycle is not accidental. It is the machine.

His relationship with the press captures the paradox. In public, Trump attacks “the media” as corrupt, fake and hostile. Yet reporters have repeatedly described a man who is intensely engaged with the media as a practical instrument of power. The Associated Press reported in 2025 that some journalists could call Trump’s personal cellphone and reach him directly. In one account, he began by berating a reporter for past coverage, then continued talking, answered questions and was described as gracious.

This is not necessarily a contradiction. It is more like a split-screen strategy. “The media” as an institution can be an enemy to attack before a crowd. An individual reporter, on the phone, can be a person to charm, pressure, flatter, scold or use. The public attack and the private access both serve the same purpose: Trump remains the central figure through whom the story must pass.

The private Trump, by many accounts, can be funny, direct, gossipy and oddly accessible. He does not always operate through formal channels. He calls people. He watches television. He reacts. He asks who is saying what. He draws outsiders, friends, donors, aides and media figures into a loose orbit of conversation and influence. Axios’s reporting on his first-term schedules described long stretches of unstructured “Executive Time,” a phrase that critics mocked but that also pointed to something real about Trump’s style: he prefers porous boundaries, informal access and immediate feedback to orderly institutional process.

At Mar-a-Lago, according to reporting from his first term, that style found its natural habitat. The club setting allowed proximity, performance and governance to blur. A conversation could be social, political, transactional and theatrical all at once. Trump could be host, celebrity, candidate, president, client and master of ceremonies in the same afternoon.

That world helps explain the importance of loyalty. Every president wants dependable advisers. Trump’s demand has often seemed more personal. In the Mueller report, investigators found substantial evidence corroborating James Comey’s account that Trump asked him for loyalty at a private dinner, including Comey’s contemporaneous memo and recollections from FBI officials. The question was not merely whether Comey would enforce policy. It was whether he would be “loyal.”

That word opens a door into Trump’s political psychology. Loyalty, in his world, often appears less like commitment to law, party or ideology than allegiance to him personally. Those who criticize him after serving him are not simply dissenters. They become traitors, weaklings, liars or failures. Former allies are recast as people who were never impressive in the first place. The past is rewritten to protect the hierarchy: Trump above, critic below.

This is one reason Trump’s orbit can feel courtlike. Favor matters. Access matters. Praise matters. Public defense matters. The leader’s regard is valuable, and his contempt can be politically dangerous. People around him learn to manage not just policy preferences but mood, grievance and status.

The most consequential example came after the 2020 election.

Trump lost. Courts rejected his claims. State officials, federal agencies and his own attorney general did not find evidence of widespread fraud sufficient to change the outcome. Reuters later summarized the repeated failure of the stolen-election claim to survive institutional scrutiny. Yet Trump did not concede in any ordinary sense. He insisted the election had been stolen and made that claim central to his post-election identity.

Seen psychologically, the stolen-election narrative did several things at once. It turned loss into victimization. It protected Trump’s identity as a winner. It gave supporters a reason to remain emotionally mobilized. It transformed a political defeat into a moral injury. It kept Trump at the center of the national drama even after he had been voted out of office.

The House January 6 committee concluded that Trump pressured Vice President Mike Pence to take actions Pence had been told he lacked authority to take. That episode was not an isolated rupture from the rest of Trump’s behavior. It was an extreme version of familiar themes: loyalty over institutional restraint, personal victory over procedural limits, pressure applied to a subordinate and refusal to accept humiliation as final.

This is where discussions of Trump often go wrong. Some critics treat him as purely irrational. But too much of what he does has been useful to call it random. He captures attention better than almost any modern American politician. He compresses complicated policy disputes into vivid emotional stories. He makes supporters feel not merely represented but avenged. He uses attacks against him as evidence that he is threatening the right people.

At the same time, some admirers and analysts treat him as purely strategic, as if every outburst were a chess move. That explanation is also too neat. The consistency of his reactions, especially to insult, criticism, defeat and disloyalty, suggests more than tactics. It suggests temperament. Trump often appears not just to use grievance but to live inside it.

The best explanation is that strategy and temperament reinforce each other. Trump seems drawn by disposition toward dominance, admiration, conflict and grievance. American media rewards conflict. Partisan politics rewards loyalty. Social media rewards emotional intensity. Supporters reward defiance. Over time, the traits and the incentives have fused.

This is why the everyday language of narcissism comes up so often around Trump. Used clinically, narcissistic personality disorder has specific criteria and cannot responsibly be diagnosed from afar. Used descriptively, however, “narcissistic traits” points to observable tendencies: grandiosity, admiration-seeking, sensitivity to humiliation, low modesty, retaliatory anger and difficulty acknowledging defeat or fault.

The distinction is crucial. To say Trump shows narcissistic traits is not the same as declaring him mentally ill. It is to say that a particular pattern predicts much of what we see.

It predicts the obsession with crowd size. It predicts the superlatives: the best, the greatest, the strongest, the most successful. It predicts the nicknames and belittling of opponents. It predicts the difficulty admitting error. It predicts the treatment of criticism as attack and attack as betrayal. It predicts the hunger for praise from people whose institutions he publicly condemns. It predicts the way a political loss can be narrated as theft rather than accepted as defeat.

It even predicts the charm.

That is the part many people miss. A dominance-oriented person need not be grim or cold in every interaction. Charm can be a form of dominance too, especially when it pulls another person into one’s frame. Trump’s private warmth, when it appears, does not necessarily contradict his public aggression. Both can be ways of winning. A compliment can control a room as surely as an insult. A phone call can be as strategic as a rally line.

This is why the question “Which Trump is real?” may be the wrong question. The rally Trump is real. The charming Trump is real. The aggrieved Trump is real. The transactional Trump is real. The performer is real too. For some public figures, performance conceals the self. For Trump, performance may be one of the primary ways the self exists.

The danger is reducing everything to personality. Trump did not create American distrust, polarization, resentment or institutional decay by himself. His supporters have many reasons for supporting him: immigration, trade, courts, religion, culture, economics, party identity, anger at elites, fear of social change. A psychological account of Trump is not a full account of Trumpism.

Corrections and updates

Subscribe now

Are Electric Cars Really Cleaner?

Evidence First — Thu, 27 Aug 2026 14:29:43 GMT

For many Americans, the electric vehicle has become more than a car. It is a symbol.

To supporters, it represents a cleaner future: no gas stations, no tailpipe smoke, no engine rumble, no carbon dioxide pouring out the back. To critics, it is a taxpayer-subsidized illusion, a car that merely hides its pollution somewhere else: in a coal plant, a lithium mine or a battery factory overseas.

Both stories contain a piece of the truth. Neither tells the whole story.

The best evidence points to a conclusion that is less satisfying for a political slogan but more useful for anyone trying to understand the issue. Electric vehicles are not zero-impact. They are not a cure-all for the environmental costs of driving. But compared with similar gasoline vehicles, they usually produce substantially less climate pollution over their full lives, even after accounting for battery manufacturing and the electricity used to charge them.

That does not make every electric vehicle equally clean. A small electric hatchback charged in a state with a relatively clean power grid is not the same as a 7,000-pound electric pickup charged in a region still heavily dependent on fossil fuels. Nor does it mean that replacing every gasoline car with an electric one would solve traffic, sprawl, road deaths, tire pollution or the environmental damage of mining.

But it does mean that one of the most common claims in the public debate, that electric cars are no cleaner than gas cars once you “count everything,” is not supported by mainstream evidence.

The important question is what, exactly, counts as “everything.”

For a gasoline car, the obvious pollution comes out of the tailpipe. Every gallon of gasoline burned releases carbon dioxide, the main heat-trapping gas driving climate change. But the full picture also includes drilling for oil, transporting it, refining it into gasoline and moving it to stations.

For an electric vehicle, the picture is different. There is no tailpipe. But there are emissions from building the car, manufacturing the battery, mining and processing minerals, generating electricity and eventually recycling or disposing of the battery. A fair comparison has to count all of that.

This is called a life-cycle analysis. In plain English, it means looking at a vehicle from production to retirement rather than judging it only by what happens while it is being driven.

When researchers do that, electric vehicles generally come out ahead.

The International Energy Agency, which tracks global energy systems, has found that a medium-size battery-electric car sold in 2023 produces roughly half the life-cycle greenhouse-gas emissions of a comparable gasoline car over about 15 years of driving. The exact number depends on where the car is used and how electricity is generated. But even if the power grid did not become cleaner over the vehicle’s lifetime, the agency found that battery-electric vehicles would still have a sizable emissions advantage.

The United States government’s own modeling points in the same direction. The Department of Energy and Argonne National Laboratory, using the widely cited GREET life-cycle model, have estimated that a representative 2025 electric vehicle produces about 46 percent fewer life-cycle greenhouse-gas emissions than a comparable gasoline vehicle using the average U.S. electricity mix. The Environmental Protection Agency similarly concludes that electric vehicles typically have a smaller carbon footprint than gasoline vehicles, even after accounting for battery production and charging.

Why does the electric car do so well, even when some of its electricity comes from fossil fuels?

The answer is efficiency.

A gasoline engine is an impressive technology, but it is also a wasteful one. Much of the energy in gasoline is lost as heat. Only a fraction becomes motion. The E.P.A. has estimated that conventional gasoline cars convert only about 16 to 25 percent of the energy in gasoline into movement at the wheels. Battery-electric vehicles, by contrast, convert a much larger share of stored energy into motion. The agency puts that figure around 87 to 91 percent from battery to wheels.

That efficiency advantage is enormous. It means that even when electricity generation produces emissions, the electric drivetrain often uses energy so much more efficiently that total emissions are lower.

This is the part of the story often missing from the claim that electric cars “just move pollution from the tailpipe to the power plant.” There is a grain of truth there. Some pollution does shift from the road to the electricity system. But it does not shift one-for-one. Power plants can be more efficient than millions of small engines, and electricity grids can be cleaned up over time by adding wind, solar, nuclear, hydro or other lower-carbon sources. A gasoline car bought today will keep burning gasoline for as long as it is driven.

Still, the critics are right about one important thing: Electric cars start with an environmental debt.

Manufacturing an electric vehicle, especially its battery, usually creates more emissions than manufacturing a comparable gasoline vehicle. Batteries require energy-intensive production and minerals such as lithium, nickel, cobalt, manganese and graphite. Mining and processing those materials can damage ecosystems, consume water and create toxic waste. In some places, mineral supply chains have also raised serious labor and human rights concerns.

That is not a footnote. It is central to the environmental accounting.

But a gasoline vehicle also has a supply chain. It depends on oil extraction, refining, shipping, pipelines, tanker trucks and combustion. The key difference is that battery-related emissions are mostly front-loaded, while gasoline emissions continue mile after mile. Over time, the gasoline car’s operating emissions usually overtake the electric car’s higher manufacturing emissions. The more the vehicle is driven, and the cleaner the electricity used to charge it, the clearer the electric vehicle’s climate advantage becomes.

A simple example helps. Imagine two similar cars leaving the factory, one gasoline and one electric. The electric car may begin with a larger carbon footprint because of its battery. But once both are on the road, the gasoline car emits carbon dioxide every time its engine runs. The electric car’s emissions depend on the electricity used to charge it. On a cleaner grid, the electric car pays back its manufacturing disadvantage relatively quickly. On a dirtier grid, the payback takes longer. But in most cases studied, the electric car eventually pulls ahead.

That does not mean the grid is irrelevant. Far from it.

An electric car in a region where electricity comes largely from coal will have a smaller climate advantage than one charged in a region with abundant renewables, nuclear power or hydropower. The same car can have different life-cycle emissions depending on where it is plugged in. This is why broad claims about electric vehicles can be misleading. “EVs are clean” and “EVs are dirty” are both too blunt. The better question is: compared with what, charged where, driven how far, and built how?

Vehicle size matters too. One of the uncomfortable truths of the electric-vehicle boom is that automakers have brought the same appetite for size and weight into the electric era. A huge electric SUV or pickup needs a bigger battery, more minerals and more energy than a smaller electric car. It may still emit less over its life than a comparable gasoline truck, but it will not be as clean as a smaller, lighter EV.

This is where political and media narratives tend to flatten the issue.

Pro-EV messaging often leans heavily on the phrase “zero-emission vehicle.” In regulatory language, that usually means zero tailpipe emissions. But to an ordinary person, it can sound like zero emissions, period. That is not accurate. Electric vehicles have no exhaust pipe, which is a major benefit, but they are not free of environmental costs.

Anti-EV messaging makes the opposite move. It points to real costs, including mining, battery production and fossil-fuel electricity, and then leaps to the much stronger claim that electric vehicles are no better than gasoline cars. That leap is where the argument usually breaks down. The caveats are real. The conclusion often is not.

Air pollution adds another layer.

Because electric vehicles have no tailpipe, they eliminate exhaust emissions where they are driven. That matters for people living near highways, ports, warehouses and busy roads, where vehicle pollution contributes to asthma, heart disease and other health risks. Replacing gasoline and diesel vehicles with electric ones can reduce local pollutants such as nitrogen oxides and tailpipe particles.

But no car is pollution-free. Vehicles also create non-exhaust particulate pollution from tires, brakes and road dust. Electric vehicles often use regenerative braking, which can reduce brake wear by using the motor to slow the car and recover energy. But they can also be heavier because of their batteries, and heavier vehicles can wear tires more quickly.

The Organization for Economic Cooperation and Development has estimated that electric vehicles generally emit somewhat less PM10, a category of larger inhalable particles, from non-exhaust sources than internal-combustion vehicles. But for the smaller PM2.5 particles, which can penetrate deep into the lungs, very heavy electric vehicles may emit more than comparable gasoline vehicles because of weight-related tire and road wear.

This is another reason the cleanest electric car is not simply the one with the largest battery or longest range. Smaller, lighter vehicles are better not only for climate emissions but also for road safety, tire pollution and material demand.

Battery recycling is often presented as a solution to the mineral problem. It can help, but it is not a magic wand.

In the long run, recycling could reduce the need for new mining by recovering valuable materials from old batteries. The International Energy Agency has projected that effective recycling could eventually cut demand for newly mined lithium, nickel and cobalt substantially, especially as the first large generations of EV batteries reach end of life. But the recycling system is still developing, and for now the rapid growth of electric vehicles means new mining remains necessary.

The strongest environmental case for electric vehicles, then, is not that they are harmless. It is that they are less harmful than the dominant alternative, particularly for climate change.

That distinction matters.

Climate change is driven by cumulative emissions. Carbon dioxide stays in the atmosphere for a long time. The transportation sector is one of the major sources of greenhouse gases. Passenger vehicles are a large part of that. If a technology can cut the lifetime emissions of cars by a large fraction, it matters, even if it does not eliminate emissions entirely.

But it also matters how the transition is carried out.

An electric future built around ever-larger vehicles, long commutes, sprawling development and disposable consumer habits will carry large environmental costs. An electric future built around smaller vehicles, cleaner grids, better transit, safer streets, battery recycling and fewer unnecessary car trips would deliver much larger benefits.

Claim or topic:

Battery minerals can create environmental and social concerns, including mining impacts, water use, toxic waste, displacement, and labor or human-rights risks.

Source:

International Energy Agency, The Role of Critical Minerals in Clean Energy Transitions

Amnesty International, DRC: Cobalt and copper mining for batteries leading to human rights abuses

U.S. Department of Energy, 2023 Critical Materials Assessment

Source type:

Expert organization analysis; human-rights investigation; government assessment.

What it supports:

The IEA identifies lithium, nickel, cobalt, manganese, and graphite as important battery materials and discusses supply-chain, environmental, and social risks tied to critical-mineral production. Amnesty documents alleged forced evictions and other human-rights abuses linked to industrial cobalt and copper mining in the Democratic Republic of Congo. The DOE assessment provides U.S. government context on critical materials used in batteries and clean-energy technologies.

Important caveat:

These sources do not show that EVs are environmentally worse than gasoline vehicles overall. They support the narrower point that battery supply chains have real environmental and social risks that should be counted, managed, and reduced.

How to read this evidence

Corrections and updates

Subscribe now

The Gun Law That Survives the Evidence

Evidence First — Wed, 26 Aug 2026 13:51:44 GMT

Start with a country that does not yet exist.

Its founders are gathered around a table, writing the first public-safety laws of a new republic. There is no inherited constitutional fight over firearms, no warehouse of private weapons already scattered across millions of homes, no decades-old slogans carved into politics. There is only a blank page and a practical problem.

Who, if anyone, should be allowed to have a gun?

The first answer is the cleanest: no one. If guns make violence more deadly, then the safest country must be one without civilian firearms. Ban them all, and the problem shrinks before it begins.

The second answer has its own appeal: let responsible people defend themselves. Police cannot be everywhere. Attackers may arrive before the state does. In a dangerous world, perhaps an armed public is not the problem but the backup plan.

Those two answers dominate much of the public argument over guns. One imagines safety through disarmament. The other imagines safety through deterrence. Both contain a truth. Neither survives the evidence intact.

The question matters because guns change the outcome of moments that might otherwise end differently. Most gun owners will never shoot anyone. Many will never draw a weapon. But public policy is not written for the average Tuesday afternoon. It is written for the night when a domestic argument turns violent, the hour when a teenager finds a parent’s pistol, the minute when a suicidal person reaches for the most lethal object nearby, the encounter when anger becomes a killing before anyone has time to reconsider.

A major systematic review and meta-analysis published in Annals of Internal Medicine found that access to firearms in the home was associated with substantially higher odds of suicide and homicide victimization. The evidence was observational, which means it cannot prove causation with the certainty of a randomized trial. But randomized trials assigning households guns would be unethical and impossible. For this question, the best available evidence is necessarily imperfect, and still deeply relevant.

Suicide is the clearest example of why access matters. Many suicide attempts are impulsive. The method available in the moment can determine whether a person dies or survives long enough for the crisis to pass. The World Health Organization identifies limiting access to lethal means as one of its evidence-based suicide-prevention strategies. Depending on the country, that can mean restricting pesticides, placing barriers at jumping sites or reducing access to firearms. The common idea is not mysterious: when the deadliest method is harder to reach, some people live.

If guns make crises more lethal, the cleanest answer seems obvious: remove the guns. That is the moral force behind a total civilian ban.

A country starting from scratch would have one great advantage over countries already saturated with weapons: it could prevent a private arsenal from forming in the first place. There would be no need to persuade millions of existing owners to surrender guns, no need to buy back weapons purchased over decades, no need to trace firearms that had long ago disappeared into private storage. In theory, the new state could say that civilian gun ownership is simply not part of its public order.

That case should be taken seriously. The fewer guns in homes, cars, bars, schools and domestic disputes, the fewer chances there are for firearms to turn fear, despair or rage into death. A peaceful country with strong border controls, little rural need for firearms and high public trust might rationally choose something close to a civilian ban.

But the moment a total ban becomes an actual law, the clean line starts to blur.

Does it ban a shotgun locked in a rural armory for pest control? A rifle used by a licensed biathlete? A museum weapon rendered inoperable? A tranquilizer firearm used by wildlife officials? A community defense unit in a border town where police may not arrive for half an hour? The closer one gets to real governance, the more exceptions appear. The question becomes whether to pretend they do not exist, or to regulate them so tightly that they do not become a path to general gun ownership.

Then there is the harder problem: a ban is only as strong as the state behind it. If the government cannot control imports, theft, corruption, illegal manufacturing and trafficking, a paper ban may mostly bind the compliant while leaving illegal actors armed. That does not mean permissive laws are safer. It means that the real object of policy is not just possession, but control: who can acquire a weapon, how it is traced, where it is stored, when it is removed and how illegal supply is interrupted.

And that weakness in the total-ban argument leads directly to the opposite temptation. If the state cannot always keep weapons out, and cannot always arrive in time, perhaps citizens should not be asked to wait unarmed.

The intuitive appeal is easy to understand. A shopkeeper facing a robber, a parent protecting a child, a citizen caught in an attack: these examples carry emotional force because they describe moments when waiting for help may not be enough. Defensive gun use does happen. A serious policy argument should not deny that.

But vivid examples are not the same as population-level evidence. A gun carried in public can stop an attack. It can also escalate an argument, be stolen, be misread by police, intensify domestic intimidation or turn a fistfight into a shooting. The national policy question is not whether a gun can ever be used defensively. It is whether broad public carrying makes society safer overall.

The evidence does not give much comfort to the “arm everyone” theory. RAND’s review of gun-policy research reports supportive evidence that more permissive concealed-carry laws are associated with increases in total homicides, firearm homicides and violent crime. Gun-law studies are difficult because jurisdictions often change multiple policies at once, enforcement varies and cultural conditions differ, so no single estimate should be treated as the final word. But the pattern is still important for lawmakers: when carrying guns in public becomes common, the aggregate risks appear to outweigh the hoped-for deterrent benefits.

Once the promise of deterrence weakens, the policy question changes. It is no longer how many people can be armed, but which risks can be identified before a gun is misused.

That is where targeted restrictions matter. RAND finds evidence that child-access prevention laws reduce firearm harms among young people, that waiting periods reduce firearm suicides and total homicides, and that domestic-violence firearm prohibitions reduce intimate-partner homicides. These policies share an insight: gun risk is not evenly distributed. It spikes around youth access, acute crisis, domestic abuse and moments of anger or despair. Good policy makes guns harder to reach precisely where the risk is most predictable.

Background checks and permit-to-purchase laws fit the same pattern. Reviews of firearm homicide research have found that stronger background-check and permit-to-purchase systems are associated with reductions in firearm homicide. That does not mean every individual study is perfect or that all policies work equally well. But it reinforces a central point: before the state asks only which weapons are allowed, it should ask who is being licensed, why, with what training and under what conditions permission can be revoked.

The value of that layered approach becomes clearer when a country actually tries to move from widespread access toward tighter control. Australia offers one important example. After the 1996 Port Arthur massacre, Australia adopted the National Firearms Agreement, which banned several types of firearms, tightened licensing and registration, and included a large buyback. RAND’s assessment notes that firearm suicides and homicides fell after the reforms, while also emphasizing an important qualification: some declines had begun before the law, so the exact causal effect cannot be measured with perfect certainty. The Australian experience does not prove every part of a model policy, but it is consistent with a practical lesson: a country can move strongly toward lower civilian gun availability without needing a metaphysical rule that every narrow civilian use must be criminal.

But any universal theory has to survive the hardest case: not a peaceful country, but one under real external threat.

Consider an Israel-like situation: a small state facing armed groups nearby and the possibility of cross-border attacks. On Oct. 7, 2023, Hamas’s military wing and other Palestinian armed groups attacked southern Israel. Human Rights Watch concluded that the attacks included war crimes and crimes against humanity against civilians. In some communities, local defenders and police tried to resist before larger forces arrived. In that setting, the first minutes are not theoretical. They are the difference between waiting helplessly and buying time.

This changes the analysis, but not in the way gun maximalists often suggest. A country under external threat may need armed local defense capacity. Border villages, rural communities, schools, farms or transport hubs may require trained responders who can act before the army or police arrive. But that is not the same as giving everyone a private weapon.

The better analogy is not the armed individual as a free agent. It is a volunteer fire brigade, reserve medical unit or auxiliary defense team: selected, trained, equipped, supervised and accountable. The weapons are not symbols of personal sovereignty. They are tools assigned for a public function. The members are vetted. Their authority is defined. Their storage is inspected. Their permission can be removed after misconduct, domestic-violence concerns, mental-health disqualification or failure to train.

The same gun that may help defend a border town can still be misused in a home, a political confrontation or a mistaken encounter. That is why external threat strengthens the case for organized defense capacity, not for abandoning screening, storage rules and accountability.

This distinction matters because fear can easily become a pipeline to uncontrolled armament. After a major attack, people understandably want immediate protection. Governments feel pressure to distribute weapons quickly. But speed can weaken safeguards, the very rules that separate public defense from vigilantism.

If even the hardest security case does not justify arming everyone, and the quietest peacetime case does not always require banning every last gun, the answer lies in control rather than absolutism.

The better system is one in which civilian access is rare, justified, traceable, securely stored and easy to revoke when risk rises. In many countries, especially those without acute security threats, that would mean something close to a near-ban: no general right to own a firearm, no ordinary handgun ownership, no routine public carry and no military-style weapons in civilian hands. But it would leave room for narrow exceptions, such as a farmer, a sport shooter, a wildlife officer or a trained border-town response team, under rules strict enough that permission can be withdrawn as easily as it is granted.

This is not a soft position. It would likely ban ordinary civilian possession of handguns, automatic weapons, military-style semi-automatic rifles and high-capacity magazines. It would make public carry rare. It would control ammunition. It would require every legal weapon to be marked, registered and traceable. That approach fits the logic of the United Nations Firearms Protocol, which emphasizes marking, recordkeeping, tracing and controls against illicit manufacturing and trafficking.

It is also not a single universal statute. Countries differ too much for that. A peaceful island state with strong border control may rationally choose something close to a civilian ban. A rural agricultural country may permit licensed long guns for pest control while banning handguns and public carry. A post-conflict country with many illegal weapons may need amnesties, buybacks, tracing and anti-trafficking enforcement before paper restrictions become meaningful. A country facing cross-border raids may need trained local defense teams.

But the universal principles are clearer than the politics around them.

No one should receive a firearm by default. Permission should require a reason. The reason should be verified. The person should be screened and trained. The weapon should be registered and traceable. The gun should be stored securely. Public carry should be rare. High-risk weapons should be banned or confined to tightly controlled state-supervised uses. Domestic violence, violent crime, credible threats and acute self-harm risk should trigger suspension or removal, with due process. Security exceptions should be institutional, not improvised through private fear.

The blank-page thought experiment matters because most countries do not write gun law on a blank page. They inherit weapons, institutions, fears, legal traditions and political compromises. Among wealthy democracies, nowhere is that clearer than in the United States.

Modern Second Amendment doctrine begins from a premise almost opposite to the one suggested by an evidence-first approach: that ordinary citizens have an individual constitutional right to possess firearms for lawful purposes, especially self-defense. In District of Columbia v. Heller, the Supreme Court held that the Second Amendment protects such an individual right and struck down Washington, D.C.’s handgun ban. Later doctrine has generally required gun regulations to be justified by constitutional text and historical tradition, not simply by a legislature’s judgment that a law would improve public safety.

An evidence-first system would begin somewhere else. It would treat access to firearms not as a default entitlement, but as a conditional public license. The analogy would not be speech, worship or voting. It would be closer to driving, prescribing narcotics, flying an aircraft or handling hazardous materials: lawful for qualified people under regulated conditions, but not something everyone may do simply because they want to. In that model, licensing is not a burden on a right; it is the condition that makes access lawful.

Corrections and updates

Subscribe now

When A.I. Helps You Write, Whose Idea Is It?

Evidence First — Sat, 22 Aug 2026 12:14:25 GMT

Imagine that you have spent several days thinking through an argument. You know what you believe, why you believe it and what examples support your view. Perhaps you even have a rough outline. What you do not have is a polished piece of writing.

So you open an artificial intelligence system and ask it to organize your argument, make it clearer, remove repetition and improve the transitions without adding new ideas. A few seconds later, the prose is better. The structure is tighter. A clumsy sentence has become an elegant one. What had been a pile of notes now reads like an essay.

Then someone tells you that you did not really write it. The A.I. did.

Did it?

The question matters because generative A.I. is becoming part of ordinary intellectual work. Students use it to revise essays. Office workers use it to prepare reports and emails. Researchers use it to summarize material or test arguments. Writers use it to brainstorm, restructure and edit. Yet much of the debate still treats A.I. use as a binary matter. Either a person wrote something or a machine did.

That distinction is too simple. There is a difference between using A.I. to express a thought and using A.I. to supply the thought itself. The difficulty is that generative A.I. can do both, sometimes in the same conversation.

To see where the difference lies, it helps to begin with tools whose role is easier to understand.

A Tool Can Do the Work Without Owning the Idea

Consider a calculator. Suppose an analyst wants to know whether a company’s revenue increased faster than inflation. The analyst chooses the relevant years, identifies the numbers, selects the calculation and interprets the result. The calculator performs the arithmetic.

No one seriously thinks the calculator deserves intellectual credit for the conclusion. The reason is not that the machine did little. It may have performed thousands of operations. The reason is that the machine did not decide what problem mattered, which inputs belonged in it or what the answer meant.

Spell-check makes the point even more starkly. It may alter dozens of words in a manuscript without becoming the source of the author’s argument. Statistical software occupies a more complicated position. A program can fit elaborate models and perform calculations that would be unrealistic to do by hand, yet the central intellectual choices may still belong to the researcher. The researcher decides what question to ask, which variables to include, which model is appropriate and how the result should be interpreted.

The amount of labor a tool performs is not the same thing as the amount of intellectual authorship it deserves. The harder test comes when assistance reaches beyond calculation or correction and begins shaping the language in which an idea is expressed.

Editors Have Always Complicated Authorship

An author sends an editor a draft with a clear argument but poor organization. The editor moves paragraphs, cuts repetition, sharpens the opening and replaces awkward language with more precise prose. The published version is substantially better.

Does the editor now own the argument?

Usually, no.

Professional authorship conventions already recognize this distinction. In medical and scientific publishing, standards used by the International Committee of Medical Journal Editors distinguish substantive intellectual contributions, such as developing a study’s conception or interpreting its findings, from writing assistance, technical editing, language editing and proofreading.

A similar distinction appears in the CRediT taxonomy used to describe contributions to academic research. It treats conceptualization separately from drafting, review and editing. These practices reject a simple assumption. The person who produces the final sentence is not necessarily the person who originated the thought inside it.

A speechwriter can improve a politician’s language without inventing the politician’s beliefs. A book editor can transform a manuscript without becoming the source of its thesis. A communications employee can turn a chief executive’s notes into a polished letter while leaving the underlying judgment intact.

So far, A.I. looks familiar. The difficulty begins when the tool can generate the choices themselves.

Where the Calculator Analogy Breaks

Suppose someone begins with a belief that social media has made political discussion worse. One use of A.I. would be to provide three reasons for that belief and ask the system to organize them. That looks much like editing.

But another use would be to ask for the strongest arguments supporting the claim.

Now the system may suggest mechanisms the user had not considered. It might mention algorithms that reward outrage, social incentives that favor extreme statements, the collapse of different audiences into a single online space, or the rapid spread of emotionally charged falsehoods. The user may find those arguments persuasive and adopt them.

At that point, something important has changed. The conclusion may still have originated with the person, but some of the reasons supporting it did not.

Research on human and A.I. collaboration suggests that this is more than a hypothetical concern. In one controlled study of creative writing, participants who received ideas from generative A.I. produced stories judged more creative on average, but those stories also became more similar to one another. The system was not merely correcting grammar. It was influencing what people produced.

Other experiments have explored whether language models can generate research proposals that human evaluators regard as novel. Such studies have important limitations. Novelty is difficult to measure, and an apparently original idea may be impractical or wrong. Still, they reinforce the broader point that generative systems can contribute candidate ideas, not merely language.

A calculator can answer a question chosen by a person. A generative system can also suggest the question, the argument and the conclusion. Once a tool can contribute substantive reasoning rather than merely execute instructions, the better comparison is no longer a calculator. It is another mind helping to write.

The Ghostwriter Is a Better Analogy

Consider two politicians preparing speeches.

The first gives a speechwriter a detailed position. The politician opposes a tax proposal, believes it would burden small businesses, thinks it fails to solve the underlying problem and wants to propose a narrower alternative. The politician also provides a personal example and asks the writer to address the strongest objection.

The speechwriter turns those instructions into a polished speech. Most people would still regard the political position as belonging to the politician. The speechwriter deserves substantial credit for expression, structure and rhetoric, but the judgment and argument were supplied beforehand.

Now imagine the second politician asks the ghostwriter to figure out what the position should be and write the speech. The ghostwriter chooses the position, develops the reasons, selects the evidence and writes the speech. The politician reads it and agrees.

The politician may now sincerely believe the argument, but agreement is not the same as origination.

Generative A.I. can play either role. It can turn a developed human position into better prose, or it can supply much of the intellectual content and leave the human mainly to approve, reject or modify what has been proposed.

That seems to offer a simple test. Ask who supplied the ideas.

But that test assumes ideas exist fully formed before writing begins. Often they do not.

Writing Can Change What We Think

Writing is not simply the act of recording finished thoughts. People often discover what they believe by trying to explain it. A paragraph that will not come together may reveal a contradiction. An analogy may expose a connection the writer had not seen. Rearranging an argument can change which conclusion seems strongest.

Research on writing has long treated planning, drafting and revision as interacting mental processes rather than a simple sequence in which thought is completed first and merely transcribed afterward.

That means A.I. can influence thought even when the user initially intends only to improve expression.

Suppose someone gives an A.I. system five points and asks it to arrange them. The system notices that two imply a sixth point and includes it in a transition. The writer reads the sentence and realizes that the connection is correct and important, even though it had not previously been noticed.

Who owns the insight?

There may be no perfectly objective answer, but blurry boundaries do not make the distinctions meaningless. Human beings have always developed ideas through interaction with other minds. Teachers ask questions that change students’ views. Editors expose weaknesses that force authors to rethink arguments. Colleagues notice implications that someone else has missed.

Outside influence does not automatically erase authorship. But it raises a harder question. If an idea first enters the conversation from somewhere else, can it later become genuinely yours? Answering that requires separating having a belief from originating it.

An Idea Can Become Yours Without Starting With You

Almost none of us develops a worldview in isolation. We inherit concepts from books, teachers, parents, colleagues and culture. Someone may first encounter an argument in a classroom, spend years testing it, reject parts of it and eventually adopt a modified version as a deeply considered belief.

That belief is not necessarily inauthentic. But there is still a difference between saying that something is genuinely what you think and saying that you came up with the idea independently. A person can truthfully make the first claim while the second is false.

The same is true with A.I.

If a chatbot presents an argument the user has never encountered, and the user scrutinizes it, tests it against evidence, modifies it and eventually adopts it, the resulting view may genuinely become that person’s considered judgment. But sincere adoption does not change where the argument originated.

That distinction matters especially in education, scholarship and any setting in which someone is claiming credit for independent intellectual work.

By this point, the original question has split apart. Asking whose work something is really means asking who supplied the ideas, who shaped the expression and who stands behind the result.

There Is More Than One Kind of Authorship

One question concerns intellectual authorship. Who supplied the thesis, interpretation, reasoning or judgment?

Another concerns compositional authorship. Who supplied the wording, structure, examples and rhetorical presentation?

A third concerns responsibility. Who understands the argument, endorses it and is prepared to defend or correct it?

Those roles can be divided. A researcher can conceive an experiment while a statistician performs part of the analysis. An executive can determine the substance of a letter while a communications team drafts it. An author can develop an argument while an editor improves its presentation.

The United States Copyright Office has made a related distinction in its treatment of works involving artificial intelligence. Copyright law is not a complete theory of intellectual authenticity, but the office has distinguished between using A.I. as an assisting tool and allowing a system to determine expressive elements itself.

The broader principle is useful. Machine assistance does not automatically eliminate human authorship. What matters is what the human actually contributed and controlled.

Once those contributions are separated, the practical question becomes easier. What intellectual work did the human bring to the process before the system intervened?

A Better Test Than Asking Whether A.I. Was Used

Several questions help distinguish one kind of A.I. assistance from another.

Before using A.I., could the person explain the main conclusion and the principal reasons supporting it? Would the central insight probably have existed if the A.I. had never suggested it? Was the system deciding mainly how to express something, or was it also deciding what should be said? And can the person explain and defend every substantive step in the final argument without depending on the machine’s wording?

These questions separate uses that otherwise look identical.

Asking A.I. to fix grammar involves almost no transfer of intellectual authorship. Asking it to make an existing argument concise and persuasive without adding new ideas leaves the intellectual content largely with the human, even if the system supplies much of the final language.

Asking A.I. to organize notes gives the system some control over presentation and perhaps over emphasis. Asking it to generate the strongest objections to a position introduces outside intellectual content, even though the original thesis remains the user’s. Asking for the best arguments supporting an existing belief leaves the conclusion human but may make the supporting reasoning substantially A.I. generated.

Asking A.I. what one should think about a subject and then asking it to write the essay delegates much of both the thinking and the expression.

Calling all of those activities simply A.I. use hides the distinction that matters most. And it brings us back to the person at the beginning.

The Intellectual Credit Should Follow the Contribution

They came to the machine with a conclusion already formed, reasons already considered and a point of view already their own. The A.I. did not tell them what to believe. It helped them say what they already believed more clearly.

That assistance is real. The final wording may not be entirely theirs, and in some academic or professional settings the use of A.I. may need to be disclosed. But neither fact settles the more important question of where the intellectual work came from.

The question becomes harder when the machine supplies more than language. If it proposes the central argument, discovers the decisive connection, generates the reasons that make the conclusion persuasive or produces an interpretation the user had not reached, then some of the thinking has come from outside the person.

Much A.I. use will fall between those extremes. A machine may sharpen an argument and reveal something the writer had not noticed. A suggestion may begin as external input and later become a deeply considered human judgment. There will not always be a precise point at which assistance becomes intellectual contribution.

But the absence of a perfect boundary does not make every case equally ambiguous. What matters is not simply how much work the helper performed, but what kind of work it was.

If the judgment, reasoning and point of view were substantially yours before the machine helped articulate them, clearer language does not make the thought less authentically yours. If the machine supplied the judgment or reasoning on which the work depends, then intellectual credit should reflect that contribution too.

The question worth asking is not whether A.I. touched the words.

It is what intellectual work the human contributed, and what intellectual work the machine supplied.

The intellectual credit should follow the intellectual contribution.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Authorship and writing assistance

Claim or topic:

Professional publishing standards distinguish substantive intellectual contributions from writing assistance, editing and proofreading.

Source:

International Committee of Medical Journal Editors

Source type:

Expert organization.

What it supports:

ICMJE authorship standards treat conception, interpretation and accountability as central to authorship. Writing assistance, technical editing, language editing and proofreading alone do not qualify someone for authorship.

Important caveat:

These are biomedical publishing standards, not a universal philosophical definition of authorship.

2. Separating ideas from drafting and editing

Claim or topic:

Research contribution standards can treat conceptualization, drafting and editing as different kinds of work.

Source:

CRediT Contributor Roles Taxonomy

Source type:

Primary document.

What it supports:

The CRediT taxonomy separately identifies contributions such as conceptualization, writing an original draft, and writing through review and editing. This supports the article’s distinction between originating an idea and helping express it.

Important caveat:

CRediT is a framework for describing research contributions. It does not determine philosophical ownership or settle every authorship dispute.

3. Generative A.I. can influence creative output

Claim or topic:

Generative A.I. can affect what people produce, not merely the grammar or presentation of existing material.

Source:

Science Advances study on generative A.I. and creative writing

Source type:

Academic research.

What it supports:

In a controlled creative-writing experiment, access to generative A.I. ideas increased judged creativity for individual stories while also making the resulting stories more similar to one another.

Important caveat:

The experiment involved a particular creative-writing task. It does not show that all A.I.-assisted writing works the same way or determine who deserves authorship credit.

4. A.I. can generate candidate research ideas

Claim or topic:

Language models can contribute candidate ideas that human evaluators may regard as novel, rather than serving only as editing tools.

Source:

Can LLMs Generate Novel Research Ideas?

Source type:

Academic research.

What it supports:

The study tested research proposals generated by language models with more than 100 NLP researchers and found that some machine-generated proposals were judged highly novel.

Important caveat:

This was a preprint, and judged novelty is not the same as scientific validity, feasibility or successful discovery. Evaluating novelty itself is also difficult.

5. Human control and A.I.-assisted copyright

Claim or topic:

Using A.I. as an assisting tool does not automatically eliminate human authorship for copyright purposes.

Source:

U.S. Copyright Office, Copyright and Artificial Intelligence

Source type:

Government analysis.

What it supports:

The Copyright Office distinguishes between human use of A.I. as an assisting tool and situations in which expressive material is generated by the system without sufficient human authorship.

Important caveat:

Copyright law answers a legal question about protectable authorship. It does not by itself determine whether an idea is intellectually or personally authentic.

6. Planning matters in A.I.-assisted writing

Claim or topic:

Research on human and A.I. writing distinguishes assistance with planning from assistance with drafting and revision, making the stage at which A.I. enters the process relevant to intellectual contribution.

Source:

Systematic review of human-A.I. writing research

Source type:

Academic research.

What it supports:

The review examines A.I. assistance across different stages of writing, including planning, formulation and revision. It supports the article’s point that help with developing what to say is meaningfully different from help expressing an already developed position.

Important caveat:

The review was released as a preprint, and the studies it surveys vary substantially in design, population and quality.

7. Generative A.I. in everyday intellectual work

Claim or topic:

Students, knowledge workers, researchers and professional writers use generative A.I. for activities including summarizing, drafting, editing, brainstorming, organizing material and other forms of writing assistance.

Source:

HEPI Student Generative AI Survey 2025

Microsoft Research, Shifting Work Patterns with Generative AI

Nature, How ChatGPT has changed scientists’ lives

Authors Guild survey of writers and generative A.I.

Source type:

Survey research, field research, reputable journalism and expert-organization survey.

What it supports:

The HEPI survey found widespread generative A.I. use among surveyed university students for tasks including explaining concepts, summarizing material and suggesting research ideas. Microsoft researchers examined how thousands of knowledge workers used generative A.I. integrated into workplace tools. Nature reported researchers using ChatGPT for tasks including polishing academic writing, reviewing literature and coding. An Authors Guild survey found writers using generative A.I. for activities including grammar review, brainstorming and organizing drafts.

Important caveat:

These sources examine different populations, tasks and time periods, so they should not be combined into a single estimate of how common A.I. use is across society. The Authors Guild is an advocacy organization, and Microsoft develops commercial A.I. products, which are relevant considerations when interpreting their findings. Together, the sources support the narrower claim that these forms of A.I. assistance are already being used across education, knowledge work, scientific research and professional writing.

How to read this evidence

Corrections and updates

Subscribe now

The Surprisingly Simple Science of Better Small Talk

Evidence First — Tue, 18 Aug 2026 18:42:49 GMT

Small talk has a branding problem.

The phrase itself suggests something trivial: verbal filler deployed in elevators, at conferences and beside bowls of chips at parties. It is the conversation people imagine they must endure before reaching something more meaningful, or escape before it becomes awkward.

And because many people assume they are bad at it, small talk has generated an enormous cottage industry of advice. Memorize a list of topics. Maintain the right amount of eye contact. Ask open-ended questions. Never talk too much. Be interesting. Be confident. Have a story ready.

The scientific evidence points toward something simpler.

Good small talk appears to depend less on performing well than on responding well. People tend to like conversation partners who ask follow-up questions, listen in ways that demonstrate attention and reveal enough about themselves to make the exchange reciprocal. They also appear to underestimate how much strangers will enjoy talking to them, how warmly they will be received and how satisfying a conversation can become if it moves beyond the most superficial level.

That does not mean there is a scientifically validated formula for charming anyone in five minutes. Much of the research involves brief conversations, often among Western participants, and usually measures immediate outcomes such as liking, enjoyment or connection rather than whether two strangers become friends a year later. But taken together, the findings offer a useful correction to how people often approach conversation. The central task may not be to impress another person. It may be to give that person evidence that you are paying attention.

We Expect Strangers to Be Less Pleasant Than They Are

One obstacle appears before a conversation even begins: people often expect social interaction to go badly.

In a series of experiments published in 2014, the behavioral scientists Nicholas Epley and Juliana Schroeder examined what happened when commuters were instructed either to interact with a stranger, remain disconnected or behave as they normally would. Many participants predicted that talking would make the trip less pleasant. Instead, those assigned to connect with another person generally reported a more positive experience. The strangers they approached did not, on average, react with the hostility or irritation people seemed to anticipate.

This gap between expectation and experience matters because avoidance can reinforce itself. If you believe talking to strangers will be uncomfortable, you may avoid doing it. Because you avoid it, you never accumulate much evidence that your prediction was too pessimistic.

Similar pessimism appears after conversations as well. In research on what psychologists call the “liking gap,” people consistently underestimated how much their conversation partners liked them. After an interaction, participants tended to think they had made a less favorable impression than their partners actually reported.

That finding will be familiar to anyone who has replayed a conversation on the way home: Why did I say that? Was I talking too much? Did that joke land? Did they actually want to leave? The other person may be performing an entirely different post-mortem, possibly worrying about what you thought of them.

The lesson is not that every interaction goes wonderfully. Some conversations are awkward. Some people are distracted, tired or simply uninterested. The evidence supports a narrower conclusion: our forecasts of social failure are often too negative.

That suggests a useful first principle of small talk: lower the standard for beginning. An opening question does not have to be brilliant. “How do you know the host?” can be enough. So can “Have you been to one of these before?” or “What brought you here?” The opener is not the performance. It is the door.

The Power of the Second Question

Once a conversation begins, one of the better-supported findings in the literature concerns questions.

Researchers led by Karen Huang at Harvard examined how question-asking affects interpersonal liking. Across several studies, people who asked more questions tended to be better liked by their conversation partners. But the most revealing finding involved follow-up questions.

A follow-up question is different from simply producing another question from a mental list. It responds directly to something the other person has just said.

Consider two conversations. In the first:

“What do you do?”

“I’m a nurse.”

“Oh. Where are you from?”

“Chicago.”

“Do you have siblings?”

The questions keep coming, but the conversation barely develops. Each answer is treated as a dead end.

Now compare:

“What do you do?”

“I’m a nurse.”

“What kind of nursing?”

“Emergency medicine.”

“That sounds intense. What made you choose it?”

The second exchange does more than collect information. It communicates that the first answer mattered.

This is one reason the distinction between being “interesting” and being “interested” can be useful. A person who is preoccupied with appearing interesting may spend much of a conversation searching for the next clever thing to say. A person who is interested has an easier task: notice what was just said and become curious about it.

The research does not imply that asking endless questions will make someone universally likable. A conversation consisting of relentless interrogation would hardly feel warm. The more defensible conclusion is that questions, particularly responsive ones, can signal attention and interest, which in turn are associated with liking.

The practical implication is to follow threads rather than cycle through topics. If someone says, “I just got back from Japan,” there are dozens of possible directions: “What was your favorite place?” “What surprised you?” “Was it your first time there?” “What made you decide to go?” If the person says Kyoto stood out, the next question can emerge from Kyoto. There is little need to abandon that thread so you can deploy the next item on a prewritten conversational checklist.

Listening Has to Be Visible

“Listen” is among the most common pieces of social advice, but it can be misleadingly vague.

A person can be listening internally while appearing disengaged. Good conversation therefore depends not just on receiving information but on demonstrating that it was received.

That might mean briefly acknowledging what someone said before asking another question. Suppose a colleague tells you she spent the weekend moving apartments. One response is: “Where did you move?” Another is: “That sounds exhausting. Did the move go smoothly?” The second response contains something small but important: evidence that you have registered the emotional meaning of what she said.

Research on high-quality listening has increasingly focused on such observable behaviors, including showing attention, responding appropriately and making speakers feel understood. The evidence base here is newer and less settled than some of the research on question-asking, but it fits a broader pattern: people respond positively not simply to being heard, but to feeling heard.

That does not require theatrical displays of empathy. Constantly saying “That must have been so hard” can sound formulaic if the situation does not call for it. Often, a simple reaction is enough: “That’s hilarious.” “I didn’t know that.” “That sounds like a lot.” “I can see why you’d like it.” Then continue.

A Conversation Is Not an Interview

There is an obvious danger in telling people to ask more questions: they may conclude that the ideal conversation involves saying as little as possible.

The evidence does not support that.

Decades of research on self-disclosure, the process of revealing information about ourselves, suggests that sharing can contribute to liking and connection. A major meta-analysis found several related patterns: people tend to disclose more to people they like, people often like those who disclose to them, and disclosure itself can sometimes increase liking.

But reciprocity matters. If one person continually asks questions while revealing nothing, the interaction can feel less like a conversation than a screening interview.

Suppose someone tells you she loves hiking. A purely interrogative response might be: “Where do you hike?” “How often?” “What’s your favorite trail?” “Do you go alone?”

A more reciprocal exchange could sound like this: “I’ve only gotten into hiking recently, but I’m starting to understand why people get obsessed with it. What’s your favorite place you’ve been?” The speaker has contributed something, but not so much that the conversation has been redirected entirely toward himself.

A useful rhythm is: ask, listen, react, share a little, continue.

There is no scientifically established ideal ratio of talking to listening. In fact, one experiment that manipulated how much participants spoke found that people assigned to speak more were, within the conditions studied, often liked more. That should not be interpreted as a license to monopolize conversations, but it does undermine the assumption that the socially safest strategy is to stay quiet. Being a good listener and being an active participant are not opposites.

Why Slightly Deeper Conversation Can Work

Many people keep small talk shallow because they assume deeper questions will create discomfort.

There is some evidence that this fear is exaggerated.

In experiments comparing relatively shallow conversations with more substantive ones, participants often predicted that deeper discussions with strangers would be more awkward than they actually were. They also underestimated how connected they would feel afterward.

“Deeper” does not necessarily mean asking a stranger about childhood trauma. The difference can be modest.

Instead of “What did you study?” try “What made you choose that?” Instead of “Where are you from?” try “What do you miss most about it?” Instead of “Did you like your trip?” try “What was the best part?”

These questions invite interpretation, preference or emotion rather than merely factual reporting. They give people an opportunity to say something that reveals how they think.

There is an important qualification: intimacy works best when it is reciprocal and responsive. A personal question that feels natural after 20 minutes may feel intrusive after 20 seconds. Context, culture, personality and power relationships all matter.

The research supports moving somewhat beyond surface-level facts more readily than people often expect. It does not support ignoring social boundaries.

The Conversation You Think Went Badly

Perhaps the most reassuring finding in this literature is that people seem to be unusually poor judges of their own social performance.

After a conversation, you know every thought that ran through your head. You know when you lost your train of thought, when you were nervous and when you nearly made a joke but abandoned it. You do not have equivalent access to the other person’s insecurity.

That asymmetry may help explain the liking gap. We evaluate ourselves using a vast amount of private evidence unavailable to everyone else. Your conversation partner mostly sees what you did: you smiled, asked another question, remembered the name of her dog, laughed when she told the story, mentioned that you had once made the same mistake.

From the inside, you may have felt nervous. From the outside, you may simply have looked engaged.

None of this guarantees that someone likes you. But the research gives people reason to distrust the immediate conclusion that an imperfect conversation was therefore an unsuccessful one.

What Science Cannot Tell You

The science of conversation remains much less precise than popular advice sometimes makes it sound.

There is no strong basis for claims that people should maintain a particular percentage of eye contact, speak exactly half the time, memorize a specific set of topics or always use open-ended questions. Such rules may occasionally be useful as training wheels. They should not be mistaken for well-established laws of human interaction.

The strongest research also has limits. Laboratory and field experiments can isolate certain behaviors, but real relationships unfold across months and years. A tactic that increases immediate liking is not necessarily the same thing that produces trust, compatibility or friendship over time.

Cultures also differ in conversational norms. So do individuals. Some people welcome personal questions quickly; others prefer a slower progression. Someone who is anxious, grieving, busy or socially exhausted may not want to talk, no matter how skillfully approached.

The evidence therefore points less toward a script than toward a stance: pay attention, become curious about what the person actually says, respond to it, offer enough of yourself that the exchange feels mutual and, when an opening appears, do not be afraid to move one step beyond facts toward what someone found surprising, difficult, funny or meaningful.

Small talk, viewed this way, is not really about mastering a collection of topics. It is the practice of showing another person, in small increments, that what they say has consequences for what you say next.

And that may be why the most effective conversational move is often not the perfect opening line.

It is the second question.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Follow-up questions and liking

Claim or topic:
People who ask more questions, especially follow-up questions that respond to what the other person just said, tend to be better liked by their conversation partners.

Source:
Huang et al., research indexed by PubMed

Source type:
Academic research.

What it supports:
Across multiple studies, greater question-asking was associated with greater interpersonal liking, with follow-up questions playing an especially important role because they signal responsiveness.

Important caveat:
The studies mostly examined relatively brief interactions and immediate impressions. They do not establish that asking more questions automatically produces stronger long-term relationships.

2. Talking to strangers is often better than people expect

Claim or topic:
People may underestimate how pleasant interactions with strangers will be and sometimes avoid conversations they would actually enjoy.

Source:
Epley and Schroeder, “Mistakenly Seeking Solitude”

Source type:
Academic research.

What it supports:
In field experiments involving commuters, participants assigned to interact with strangers generally reported more positive experiences than participants assigned to remain disconnected, despite expecting interaction to be less pleasant.

Important caveat:
The findings concern particular short-term social situations. They do not mean every stranger wants to talk or that initiating conversation is appropriate in every context.

3. The “liking gap”

Claim or topic:
After conversations, people often underestimate how much their conversation partners liked them and enjoyed the interaction.

Source:
Psychological Science research on the “liking gap”

Source type:
Academic research.

What it supports:
The research found a recurring gap between how positively people believed they were viewed and how positively their conversation partners actually rated them.

Important caveat:
This is an average tendency, not a guarantee that any particular conversation went well. People can still misread interactions in either direction.

4. Self-disclosure, reciprocity and liking

Claim or topic:
Sharing information about yourself can contribute to liking and connection, particularly when disclosure is reciprocal rather than one-sided.

Source:
Collins and Miller meta-analysis indexed by PubMed

Source type:
Academic research.

What it supports:
The meta-analysis found several related patterns: people tend to disclose more to people they like, people tend to like those who disclose to them, and disclosure itself can sometimes increase liking.

Important caveat:
Self-disclosure and liking influence each other, so not every association is purely causal. The appropriate depth of disclosure also depends heavily on context and relationship stage.

5. Speaking more is not necessarily socially harmful

Claim or topic:
The common assumption that people should minimize how much they speak to be better liked is not well supported.

Source:
SAGE Journals study on speaking time and interpersonal liking

Source type:
Academic research.

What it supports:
In an experiment that varied how much participants were instructed to speak, participants assigned to speak more were, within the study conditions, often better liked.

Important caveat:
This does not show that dominating a conversation is beneficial. The study tested a limited range of speaking proportions in a controlled setting, not extreme monopolizing behavior.

6. Deeper conversations can be less awkward than expected

Claim or topic:
People may be too pessimistic about having somewhat deeper conversations with strangers and may underestimate how connected those conversations can make them feel.

Source:
American Psychological Association summary of research on deeper conversations

Source type:
Expert organization summarizing academic research.

What it supports:
The research found that participants often expected deeper conversations to be more awkward and less enjoyable than they turned out to be, while also underestimating the resulting sense of connection.

Important caveat:
“Deeper” does not mean indiscriminately asking highly personal questions. Social boundaries, timing, culture and reciprocity still matter.

7. Responsive listening

Claim or topic:
Listening appears to work best socially when it is visible through attentive, responsive behavior that helps the speaker feel understood.

Source:
Recent high-quality listening research indexed by PubMed

Source type:
Academic research.

What it supports:
The research contributes to evidence that observable listening behaviors and perceived responsiveness are associated with stronger social connection and better conversational experiences.

Important caveat:
This evidence is newer and less established than the research on question-asking, self-disclosure and the liking gap. The article therefore treats it more cautiously.

How to read this evidence

Corrections and updates

Subscribe now

Capitalism Is Losing the Language War

Evidence First — Sun, 16 Aug 2026 15:08:50 GMT

Something strange is happening in American politics. In a recent Psyclone study, half of self-identified Democrats said they would be open to the United States being governed under socialism. Only 44 percent said the same about capitalism. More surprisingly, 25 percent said they would be open to communism.

At first glance, those numbers seem to tell a simple story. Democrats are moving away from capitalism, toward socialism and, at the margins, toward ideas once considered far outside the American mainstream. But then consider a different set of numbers. Gallup found in 2025 that only 54 percent of Americans had a positive view of capitalism. Yet 81 percent viewed free enterprise positively, and 95 percent viewed small business positively.

Americans, in other words, appear to be cooling toward capitalism while remaining overwhelmingly fond of some of the institutions most closely associated with a market economy. How can both things be true?

That turns out to be the more interesting question. The evidence points to two developments happening at once. There has been a real change in political attitudes, especially among Democrats. But there is also growing disagreement over what words like capitalism, socialism and even communism are supposed to mean.

The change in attitudes is difficult to dismiss. Gallup found in 2025 that 66 percent of Democrats viewed socialism positively, compared with 42 percent who viewed capitalism positively. Republicans showed the opposite pattern, and much more sharply: 74 percent viewed capitalism positively and only 14 percent socialism. Pew Research Center found a similar partisan divide in 2022, with 57 percent of Democrats and Democratic-leaning independents positive toward socialism and 46 percent positive toward capitalism. A Fox News poll in 2026 again found socialism ahead among Democrats, 52 percent to 41 percent.

The exact percentages vary. A 2026 Cato Institute and Morning Consult survey, for example, found Democrats almost evenly divided, with 50 percent favorable toward socialism and 52 percent toward capitalism. But the broader pattern has appeared often enough, across different pollsters and different years, that it is difficult to explain away as statistical noise. Socialism has become at least as acceptable as capitalism among a large share of Democratic voters, while Republicans remain much more firmly attached to capitalism.

Before deciding what that shift means, however, we have to know what voters believe they are accepting and rejecting.

Ask two Americans what capitalism means and you may hear descriptions of entirely different societies. One person might picture a family opening a restaurant, an engineer building a company or a consumer choosing among competing products. Another might picture a pharmaceutical monopoly, unaffordable housing, billionaires wielding political influence or a family financially devastated by a medical emergency.

Socialism has the same problem. To one voter, the word conjures Soviet central planning, state ownership and bread lines. To another, it means Denmark, universal health insurance or simply making sure people can afford food, housing and medical care. Those are not minor differences in emphasis. They describe substantially different economic arrangements.

Pew encountered this directly when it asked Americans to explain their views in their own words. People who liked socialism often talked about fairness, social programs and meeting basic needs. Critics talked about government control, weak incentives and countries such as Venezuela. Some respondents explicitly said they preferred a mixture of capitalism and socialism rather than either system in pure form.

Gallup’s findings make the language problem even clearer. If Americans were broadly rejecting the institutions of a market economy, we might expect attitudes toward free enterprise and small business to decline along with attitudes toward capitalism. They have not. The gap between 54 percent positive toward capitalism and 81 percent positive toward free enterprise suggests that many Americans may be reacting less to markets themselves than to what the word capitalism has come to symbolize for them: corporate power, concentrated wealth, inequality or economic insecurity.

That might seem to solve the mystery. Perhaps socialism’s rise is mostly a matter of branding. Perhaps voters still want a market economy but increasingly dislike calling it capitalism.

The evidence is not quite that simple.

The Psyclone study offers a useful test because respondents were given short descriptions rather than being left entirely to their own understanding of the labels. Socialism was described as “shared ownership of major resources, aiming to reduce inequality while keeping some private property.” Communism was described as “state ownership of resources, aiming for a classless society.”

Under those definitions, 50 percent of Democrats said they would be open to socialism and 25 percent said they would be open to communism. Among Republicans, 29 percent were open to socialism and 20 percent to communism, while 53 percent were open to capitalism.

Those findings require care. Being “open to” something is not the same as preferring it, supporting it or wanting it implemented. The Democratic subgroup included 229 respondents, so the result should not be treated as a perfectly precise estimate of every Democrat nationwide. And the five-point Democratic-Republican gap on communism is too small, given the sample sizes, to establish a meaningful partisan difference with confidence.

Even with those qualifications, the result is revealing. Respondents were not simply saying they liked Medicare or thought Scandinavian countries seemed appealing. They were being asked to consider shared ownership of major economic resources, and half of the Democrats in the sample remained open to it.

That suggests two things are happening at the same time. Some Americans are plainly using socialism as a loose synonym for a generous welfare state. But some also appear receptive to more substantial changes in who owns and controls parts of the economy.

Survey respondents can hold vague or inconsistent ideas. Political organizations eventually have to put their goals into words. That makes the Democratic Socialists of America a useful real-world test of where ordinary welfare-state liberalism ends and something more radical begins.

The DSA’s current program goes considerably further than the social programs many Americans appear to associate with socialism. The organization asks supporters to imagine “a future without capitalism.” It describes a “classless society” as its “guiding star” and advocates public ownership of major corporations and essential industries.

Those are not simply proposals for a larger child tax credit, stronger labor protections or cheaper prescription drugs. They represent a fundamentally different view of who should own and control substantial parts of the economy.

That helps explain why Bill Maher’s recent argument that communist ideas exist inside DSA has a factual foundation. Explicitly communist and Marxist tendencies have existed within the organization, and parts of DSA’s stated program overlap with longstanding socialist and communist goals, including ending capitalism, expanding social ownership and ultimately eliminating class divisions.

But calling DSA simply communist leaves out something important. The organization also advocates multiparty democracy, proportional representation and competitive elections. Traditional authoritarian communist regimes did not merely socialize property. They concentrated political power in one-party states. DSA’s stated political vision retains electoral democracy.

The better conclusion is narrower. DSA is substantially more radical than ordinary welfare-state liberalism, and its program includes goals historically associated with Marxist socialism. At the same time, its stated political system is not Soviet-style one-party rule.

That distinction exposes a larger weakness in the entire capitalism-versus-socialism debate. If socialism can refer both to Medicare and to abolishing capitalism, the label eventually stops telling us enough. A more useful question is not which ideological word sounds better. It is which institutions work best for which purposes.

Markets have formidable strengths. Imagine that a city suddenly develops a taste for avocados. No government planner has to survey every household, calculate how many avocados people will want and issue production orders to farmers. Prices convey information. If demand rises, growers and distributors have an incentive to supply more. Stores experiment with what customers will pay, while competitors search for cheaper ways to grow, transport and sell the product.

The same process creates incentives to innovate. If someone develops a cheaper battery, opens a better restaurant or finds a faster way to deliver packages, a market economy gives that person a reason to try and gives competitors a reason to respond. Businesses can fail, investments can go wrong and technological change can destroy old jobs. But competition, private ownership and the possibility of profit can be powerful engines of innovation, adaptation and productivity.

That helps explain why Americans can dislike the word capitalism while remaining enthusiastic about entrepreneurship, free enterprise and small business.

Markets, however, have well-known weaknesses. Consider a factory that earns millions of dollars while polluting a river. The company receives the profit while families downstream bear part of the cost. Economists call this an externality, meaning that some costs created by a transaction fall on people who were not part of it. The market price may not reflect those costs.

Competition also works poorly when there is little actual competition. A monopoly can gain unusual power over prices, suppliers, workers and potential rivals. Health care presents another difficulty because a person having a heart attack is not in an ideal position to comparison-shop among hospitals, and patients usually know far less about medical treatment than the people providing it. Catastrophic illness, disability, unemployment and old age also create risks that individuals cannot always manage efficiently on their own.

Once those weaknesses are visible, the case for some collective action becomes easier to understand. Social insurance can spread risk. Public education can broaden opportunity. Environmental rules can make companies account for costs they would otherwise impose on others. Antitrust policy can preserve competition. Governments can build infrastructure and support basic scientific research whose benefits may be too widely dispersed for private companies to capture fully.

Many of these ideas have roots in socialism, social democracy or progressive reform, but adopting them does not require abolishing markets. In practice, most wealthy democracies have settled on something less ideologically pure. They combine private companies, markets, entrepreneurs and profits with public schools, pensions, regulation, unemployment insurance, infrastructure and publicly financed or subsidized health care. Across countries in the Organization for Economic Cooperation and Development, public social spending averages roughly one-fifth of economic output.

These countries are generally described as mixed economies. The basic bargain is straightforward, even if the details produce endless political fights: let markets handle the tasks they tend to perform well, use government where there are strong reasons to believe markets will produce poor outcomes, and place constraints on both private and public power.

Government, after all, has characteristic failures of its own. A regulation intended to protect consumers can end up protecting established companies from competition. Subsidies can become political favors. Bureaucracies can waste money. Poorly designed taxes can weaken useful incentives. The choice is therefore not between virtuous markets and dangerous government, or virtuous government and dangerous markets. It is about deciding which institution is better suited to which problem, and how to limit the ways each can go wrong.

Seen that way, the strange polling begins to make more sense. A Democrat can dislike “capitalism” while still supporting entrepreneurship and markets. A Republican can dislike “socialism” while strongly defending Social Security and Medicare. A voter can simultaneously believe in private property, environmental regulation, profitable companies, a safety net, competition and antitrust enforcement. The ideological labels imply a clean binary that the institutions themselves do not.

This brings us back to the contradiction at the beginning. How can capitalism draw relatively lukewarm support while free enterprise and small business remain overwhelmingly popular? One plausible answer is that many voters are not rejecting markets themselves. They are rejecting what they think capitalism has come to represent.

And how can socialism become more popular at the same time? Part of the answer is that many voters appear to associate socialism with social insurance, public services and protection against economic insecurity. But the Psyclone findings suggest that this is not the entire story. Some Democrats really do appear open to greater shared ownership, while organizations such as DSA show that an organized constituency for a genuinely post-capitalist economy exists.

The polling therefore points to two developments at once: a real ideological shift and a vocabulary problem. It would be a mistake to assume that every Democrat who likes socialism wants government ownership of major industries. It would also be a mistake to assume that every expression of support for socialism simply means that someone likes Medicare.

The more useful political questions lie underneath the labels. How much should markets decide? Where should government intervene? How much inequality should a society accept? When does social protection become excessive control? When does private economic power become large enough to require restraint?

Those questions are harder than asking Americans whether they approve of capitalism or socialism, but they are much closer to the argument the country is actually having. Many Americans appear to want entrepreneurship without monopoly, private property with a safety net, competitive markets with rules, and prosperity without financial catastrophe for people who become sick, disabled, old or simply unlucky.

That is neither laissez-faire capitalism nor comprehensive socialism. It is the mixed economy that most advanced democracies already practice.

The argument is over where to draw the lines.

Americans may agree much more about the ingredients than they do about what to call the recipe.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Psyclone survey results

Claim or topic:
In the Psyclone study, 50 percent of self-identified Democrats said they would be open to socialism as defined in the survey, 44 percent to capitalism, and 25 percent to communism. Republican openness was 29 percent for socialism, 53 percent for capitalism, and 20 percent for communism.

Source:
Psyclone study results and party crosstabs supplied by the author.

Source type:
Primary research / original survey data.

What it supports:
These results provide the article’s central original comparison and allow attitudes toward defined economic systems to be examined by party.

Important caveat:
“Open to” does not mean “prefer” or “want implemented.” The Democratic subgroup contained 229 respondents, and the five-point Democratic-Republican difference on communism is too small to establish a clear partisan difference with confidence.

2. Gallup on capitalism, socialism and free enterprise

Claim or topic:
Democrats have become substantially more positive toward socialism than capitalism, while Americans remain much more positive toward free enterprise and small business than toward the word “capitalism.”

Source:
Gallup, “Image of Capitalism Slips to Record Low in U.S.”

Source type:
Expert organization.

What it supports:
Gallup reported 66 percent positive views of socialism and 42 percent positive views of capitalism among Democrats in 2025. Among Republicans, capitalism led socialism 74 percent to 14 percent. Nationally, 81 percent viewed free enterprise positively and 95 percent viewed small business positively.

Important caveat:
Favorability toward a label does not necessarily reveal which specific policies or ownership arrangements a respondent supports.

3. The broader partisan polling pattern

Claim or topic:
The Democratic shift toward greater acceptance of socialism is not unique to one pollster, although the size of the gap varies substantially.

Source:
Pew Research Center, 2022; Fox News, 2026; Cato Institute/Morning Consult, 2026 crosstabs

Source type:
Expert organization, reputable journalism, and primary document.

What it supports:
Pew and Fox found socialism rated more positively than capitalism among Democrats, while the Cato/Morning Consult survey found Democrats approximately tied between the two. Republicans remained more favorable toward capitalism.

Important caveat:
Different surveys use different samples, wording, question order and measures such as “positive,” “favorable” or “open to.” The percentages should not be treated as interchangeable.

4. What Americans mean by “socialism” and “capitalism”

Claim or topic:
Americans attach very different meanings to the same ideological labels, with some associating socialism with social programs and basic needs and others associating it with government control or failed authoritarian systems.

Source:
Pew Research Center, “In Their Own Words: Behind Americans’ Views of ‘Socialism’ and ‘Capitalism’”

Source type:
Expert organization.

What it supports:
Pew’s open-ended research shows that respondents use “socialism” and “capitalism” to describe a wide range of ideas. Some explicitly favored a combination of the two rather than either system in pure form.

Important caveat:
Open-ended responses reveal how people understand the terms, but they do not by themselves establish which economic arrangements produce better outcomes.

5. What the Democratic Socialists of America officially advocates

Claim or topic:
DSA’s stated program goes beyond ordinary welfare-state liberalism by calling for a future without capitalism, greater public ownership and a classless society, while also supporting multiparty electoral democracy.

Source:
Democratic Socialists of America, official program

Source type:
Primary document.

What it supports:
The source allows readers to examine DSA’s stated goals directly rather than relying on descriptions by supporters or critics.

Important caveat:
Overlap with historical socialist or communist goals does not by itself make DSA equivalent to Soviet-style one-party communism. Its official program also includes democratic political institutions.

6. Communist and Marxist tendencies within DSA

Claim or topic:
Explicitly communist and Marxist tendencies have existed within the broader DSA coalition.

Source:
Young Democratic Socialists of America, “Monday Hot Take”

Source type:
Primary document.

What it supports:
A publication from DSA’s youth organization described the movement as including communists alongside other ideological tendencies, supporting the narrower claim that communist currents exist within the organization.

Important caveat:
This does not establish what percentage of DSA members identify as communist or how influential those factions are across the organization as a whole.

7. Why markets and regulation can both have economic value

Claim or topic:
Competitive markets can support innovation and productivity, while regulation can improve outcomes when markets suffer from problems such as monopoly power, externalities or information failures.

Source:
OECD, “Time for a Regulatory Reset”

Source type:
Expert organization.

What it supports:
The OECD discusses both the economic value of competition and the legitimate role of well-designed regulation in addressing market failures.

Important caveat:
Recognizing a rationale for regulation does not determine how extensive a particular regulation should be. Poorly designed rules can themselves reduce competition or create economic costs.

8. Social protection inside modern market economies

Claim or topic:
Modern wealthy democracies commonly combine private markets with substantial public social spending rather than operating as either pure laissez-faire capitalist or comprehensively socialist systems.

Source:
OECD Social Expenditure Dashboard

Source type:
Expert organization.

What it supports:
OECD data show that public social expenditure is a major feature of advanced market economies, averaging roughly one-fifth of economic output across OECD countries.

Important caveat:
Countries differ substantially in taxation, public ownership, regulation and social spending. “Mixed economy” describes a broad family of arrangements rather than one single model.

How to read this evidence

Corrections and updates

Subscribe now

How Anthony Fauci Became the Villain of America’s Pandemic Story

Evidence First — Tue, 11 Aug 2026 16:45:45 GMT

Late last month, Anthony Fauci sat before a Senate committee and repeatedly declined to answer questions.

He would not discuss the origins of Covid-19. He would not answer questions about government-funded coronavirus research or about emails investigators said raised concerns over federal recordkeeping. Those refusals were not entirely surprising. Fauci had been called before a committee led by Senator Rand Paul of Kentucky, a longtime antagonist who had already sought his criminal prosecution.

Then came questions that seemed almost absurdly simple. What day was it? What color was Fauci’s tie? Was there a folder sitting in front of him?

Fauci invoked the Fifth Amendment again.

For anyone who remembered the spring of 2020, the scene was remarkable. Six years earlier, Fauci had been one of the most recognizable doctors in America, appearing beside the president during frightening White House briefings and explaining a virus scientists were still struggling to understand. Now he was sitting before Congress as Republicans accused him of deception and demanded prosecution. A week later, the Senate committee voted along party lines to advance a contempt resolution against him.

How did the country get from one moment to the other?

For nearly four decades before Covid, Fauci directed the National Institute of Allergy and Infectious Diseases, working on H.I.V./AIDS, influenza, Ebola, Zika, SARS and other infectious diseases. Presidents of both parties relied on him. His public identity was built not around culture-war politics, but around infectious disease and biomedical research.

When Covid arrived, that background gave Fauci substantial institutional credibility and made him a natural national spokesman. His agency helped develop and test the Moderna vaccine. He emphasized clinical trials and resisted pressure to embrace treatments simply because they were politically popular. Hydroxychloroquine, promoted enthusiastically by President Trump and others, later failed to show meaningful benefit for hospitalized Covid patients in randomized trials.

Fauci was also often an effective communicator. He could explain complicated biomedical concepts in plain language, and research examining his pandemic-era appearances found that clarity, precision and repetition were recurring features of his public communication. In an environment saturated with fear, political conflict and contradictory messages, that ability helped make him one of the country’s most prominent scientific voices.

That record helps explain why Fauci initially carried such authority. It also helps explain why the reversals that followed carried so much weight.

The first cracks in public trust appeared early. In March 2020, Fauci told Americans that healthy people did not generally need to walk around wearing masks. At the time, hospitals were desperately short of protective equipment, scientists were still learning how easily people without symptoms could spread the virus and officials feared that public demand would worsen shortages.

Within weeks, the advice changed. The Centers for Disease Control and Prevention began recommending public masking as evidence accumulated that people who did not feel sick could nevertheless transmit the virus. To scientists, changing a recommendation as evidence changes is normal. That is how science is supposed to work. To many ordinary Americans, however, it looked much simpler: First the experts said masks were unnecessary. Then they said masks were important. If the second statement was true, people reasonably wondered why the first one had been made.

The changing guidance was difficult to communicate under any circumstances, and Fauci often explained complex science clearly. But officials, including Fauci, did not always make the underlying uncertainty, supply constraints and reasons for changing recommendations as explicit as they might have. A scientific question, whether masks reduced transmission, became entangled with a supply question, whether scarce masks should be preserved for hospital workers, and an uncertainty question, how much transmission was occurring from people without symptoms. The public often received a relatively simple answer when the reality was considerably messier.

That distinction matters. The problem was not that Fauci generally could not communicate. It was that some of the government’s most consequential messages were delivered in circumstances where scientific evidence, emergency logistics and policy judgments were changing at the same time. Even a skilled communicator could struggle to explain that combination without sounding inconsistent.

It was an early lesson in how quickly scientific uncertainty could become a question of personal credibility.

If masks weakened trust because the guidance changed, the six-foot distancing rule reinforced a different concern: rough judgments could sometimes be presented with the precision of settled science. Keeping people farther apart during the spread of a respiratory virus had a scientific rationale. But six feet sounded like a firmly established dividing line between danger and safety. Fauci later said he did not recall studies establishing that exact threshold, although public-health officials had drawn on earlier research about respiratory droplets and transmission. The problem was not that distancing lacked a basis. It was that the precision of the six-foot rule exceeded the strength of the evidence for that particular cutoff.

As the pandemic dragged on, Fauci’s recommendations were no longer being heard merely as medical advice. In the public debate, he had become closely associated with a larger system of restrictions affecting schools, businesses and ordinary life.

With schools, those questions became much more consequential because the costs of precaution were no longer abstract. Closing classrooms during the frightening first weeks of March 2020 was understandable. Officials did not know how severe the crisis would become, hospitals feared being overwhelmed and there were no vaccines.

But the calculation changed over time. Evidence increasingly showed that children generally faced much lower risks of severe disease than older adults. Schools learned to use ventilation, testing and other precautions. Meanwhile, the costs of keeping children away from classrooms became increasingly visible.

Pandemic-era learning losses were substantial. Absenteeism rose, and disadvantaged students often experienced particularly serious disruption. Remote schooling was not solely responsible. Illness, family stress, economic disruption and other effects of the pandemic also damaged learning. But research has linked prolonged virtual instruction with worse educational outcomes, making school closures an important part of the explanation.

Fauci did not order America’s schools closed. Governors, mayors, school boards, local health departments, unions and other institutions made those decisions. Yet he was the country’s most influential scientific voice, and his words mattered. It is fair to ask whether he and other public-health leaders gave enough weight to the accumulating costs of prolonged restrictions as evidence about risk changed.

The school debate captured a problem that would recur throughout the pandemic: the target kept moving. What was reasonable in March 2020 did not necessarily remain reasonable months later as scientists learned more about the virus, treatments improved and the costs of restrictions became clearer. The fair question is not simply whether guidance changed, but whether officials changed it quickly enough and explained why.

By then, Fauci had become a symbol as much as an adviser. To admirers, he represented expertise during a frightening emergency. To critics, he had become the human face of rules they believed had lasted too long or been presented with too much certainty.

Until that point, much of the case against Fauci concerned judgment. Wuhan changed the accusation. Critics were no longer asking only whether he had been wrong. They were asking whether he had been fully candid.

The National Institutes of Health gave money to EcoHealth Alliance, which in turn supported research involving bat coronaviruses at the Wuhan Institute of Virology. A federal inspector general later found that NIH and EcoHealth did not effectively monitor relevant awards and subawards and missed opportunities to oversee the research more effectively. That finding establishes a real oversight failure and gives Congress a legitimate reason to ask whether federal officials adequately supervised potentially risky research.

The controversy became more explosive because of the phrase “gain of function.” In broad scientific usage, the term can describe experiments that give an organism a new or enhanced characteristic. The federal government has also used narrower regulatory categories for particular kinds of potentially dangerous research.

Fauci and NIH relied on a narrower federal regulatory definition when denying that NIH had funded gain-of-function research in Wuhan, which is one reason the dispute has persisted. To an ordinary listener, however, Fauci’s categorical answer could easily sound broader: that the government had not supported research that enhanced viral characteristics at all. A fuller explanation of the definitional dispute would have been more transparent.

From there, suspicion grew beyond what the established evidence could bear. If federal money supported controversial coronavirus research in Wuhan, and if Fauci described that research too categorically, perhaps the research created SARS-CoV-2 and perhaps Fauci helped conceal it.

That conclusion has not been established. The origin of Covid remains unresolved. American intelligence agencies have differed over whether an animal spillover or a laboratory-related incident is more likely, with several assessments carrying low or moderate confidence. The C.I.A. shifted in 2025 toward viewing a laboratory origin as more likely, but did so with low confidence. Public evidence has not demonstrated that the particular NIH-funded research through EcoHealth created SARS-CoV-2.

Against that backdrop, 2020 messages disclosed by Senate investigators in which Fauci instructed recipients to delete an email after reading it took on unusual significance. In a political environment already shaped by allegations of concealment, the wording naturally invited questions. For a senior government official subject to record-retention requirements, those questions are legitimate.

But an email can be a reason to investigate without being proof of the larger theory. The messages do not, by themselves, establish that Fauci systematically destroyed required government records or concealed the origin of Covid.

To Fauci’s critics, the pieces increasingly seemed to fit together: changing guidance suggested unreliability, Wuhan raised questions about candor, and the deletion language appeared to offer evidence of concealment. But those pieces do not necessarily prove a common underlying theory. Each allegation still requires its own evidence.

All of those disputes eventually converged in the Senate hearing room where this story began.

Six years after Americans watched Fauci explain a new virus from the White House briefing room, he returned to Congress not as the government’s reassuring expert, but as a witness protecting himself from possible criminal exposure.

Senator Paul had spent years accusing Fauci of misleading Congress and had already asked the Justice Department to prosecute him. President Biden, shortly before leaving office, issued Fauci a broad pardon covering potential federal offenses arising from specified government service from Jan. 1, 2014, through Jan. 19, 2025.

The pardon removed substantial federal exposure for Fauci’s earlier government service, but legal experts noted that it did not necessarily eliminate every possible risk, including potential state prosecution or liability arising from later conduct. Fauci responded by invoking the Fifth Amendment again and again.

Invoking the Fifth Amendment is not evidence of guilt. The constitutional privilege exists to prevent a witness from being compelled to provide testimony that could be used in a criminal case. But Fauci’s use of it was exceptionally broad.

By refusing even apparently harmless questions, he made a legally arguable strategy look evasive. Americans watching clips of the hearing did not see the boundaries of a complicated constitutional argument. They saw a famous former government official refusing to say what color tie he was wearing.

The moment strengthened the very suspicions he was trying to protect himself from. It also highlighted the unusual setting: Fauci was being questioned by politicians who had already publicly argued that he should be prosecuted.

Congress has every right to investigate the pandemic response. It should examine poorly supervised research. It should determine whether federal records were preserved. It should ask whether officials testified accurately. The country spent trillions of dollars, endured extraordinary restrictions and lost more than a million people. Serious oversight is not optional.

Oversight and politics, however, can coexist. Paul may sincerely believe Fauci committed wrongdoing. The available evidence does not allow anyone to know his private motive with certainty. At the same time, Fauci is an intensely polarizing figure, particularly among Republican voters, and pursuing him carries obvious political value. Paul had reached severe conclusions about Fauci before the latest testimony. The committee’s contempt action divided along party lines. President Trump later suggested that Fauci should be prosecuted.

None of that makes the investigation illegitimate. It establishes that legitimate oversight is occurring inside a highly partisan environment in which punitive and political incentives are also present.

By the end of the hearing, the various controversies had been assembled into a much more serious claim. Masks, school closures, research grants, emails and testimony were no longer being cited merely as examples of mistakes. Together, they were being used to support an allegation about motive: that Fauci had not simply gotten things wrong, but had knowingly misled the public or concealed wrongdoing.

That is a much harder claim to prove.

The record shows a man who spent decades working against infectious disease, supported vaccines and clinical trials, made recommendations that were later revised and often communicated difficult scientific concepts effectively under extraordinary conditions. It also shows moments when uncertainty, policy tradeoffs and changing evidence were not communicated as clearly as they could have been. It shows an agency operating within a grant system that failed in important ways to oversee some coronavirus research. It shows statements about gain-of-function research that relied on a narrower federal definition but could reasonably have been understood more broadly by the public. It shows emails that warrant investigation. And it shows a former public official who, facing politicians already calling for his prosecution, chose an extraordinarily broad Fifth Amendment strategy that made him appear more evasive.

What it does not show is convincing evidence that Fauci wanted Americans harmed.

The problem is that much of this record is now being viewed through six years of hindsight. Accumulated knowledge can make the uncertainty of early 2020 difficult to remember. We now know far more about transmission, children’s risks, vaccines, treatments and the consequences of prolonged restrictions than officials knew when the first decisions were being made. Some policies remained in place after the evidence supporting them had weakened, and those choices deserve criticism. But officials should still be judged against what could reasonably have been known at the time, and against how readily they changed course when better evidence arrived.

Once hindsight is accounted for, the strongest criticism of Fauci is also the narrower one. He sometimes conveyed uncertain judgments with too much confidence, defended some positions too categorically and led an agency within a system that failed to oversee important research adequately. That is a serious record to examine. It is not the same as saying he was generally a poor communicator, and it is not evidence that he intended to harm the public.

The available record is much more consistent with a public-health official trying to reduce illness and death, sometimes getting the judgment wrong and sometimes failing to make uncertainty or changing assumptions explicit enough, than with a deliberate effort to harm Americans. That conclusion is necessarily an inference about intent, but it fits Fauci’s decades-long career, his stated objectives, the recommendations he consistently favored and the absence of persuasive evidence that he sought harmful outcomes.

Good intentions do not erase learning losses, weak research oversight or communication mistakes. Nor do they prevent Congress from demanding answers. But the same evidentiary discipline should govern the investigation itself. Legitimate oversight can coexist with political incentives, and investigators should distinguish mistakes from misconduct, misconduct from crime and suspicion from proof.

The Fauci sitting before Congress in 2026 was neither the infallible scientist some Americans imagined in 2020 nor the malevolent figure his harshest critics now describe. The evidence points to something more familiar: a public-health official trying to navigate a moving crisis, often communicating difficult science effectively, sometimes getting the judgment wrong and sometimes expressing more certainty than the evidence ultimately justified.

National disasters invite simpler narratives. In one, Fauci was “the science,” a heroic expert whose critics simply refused to listen. In another, he was the architect of lockdowns, mandates, school closures and perhaps even a cover-up surrounding the creation of Covid itself. Neither version survives close examination.

A better reckoning would be specific. If Fauci or his agency failed to supervise research properly, Congress should establish how and why. If he testified inaccurately, investigators should identify the statement and demonstrate what made it false. If government records were destroyed improperly, the evidence should establish what was deleted, who deleted it and whether the law was violated.

The strongest evidence supports a case against Fauci’s infallibility far more than it supports a case for villainy. He appears to have been trying to help, and he also made mistakes worth understanding. A serious reckoning with the pandemic has to be able to say both.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Fauci’s pandemic role and the Moderna vaccine

Claim or topic:
Fauci led NIAID for nearly four decades, became a central figure in the federal Covid response, and NIAID participated in developing and testing the Moderna vaccine.

Source:
National Institutes of Health

Source type:
Government research agency and peer-reviewed clinical-trial summary.

What it supports:
NIH reported that the NIAID-supported Phase 3 trial found the original Moderna vaccine 94.1 percent effective at preventing symptomatic Covid-19 under the trial conditions.

Important caveat:
This figure describes the original trial before later variants emerged. Vaccine effectiveness changed as the virus evolved and immunity waned.

2. Fauci’s early mask guidance

Claim or topic:
In March 2020, Fauci told the public that healthy people generally did not need to wear masks, while also expressing concern about preserving protective equipment for healthcare workers. Public guidance changed as knowledge about transmission developed.

Source:
CBS News, March 8, 2020 interview

Source type:
Contemporaneous reputable journalism and recorded interview.

What it supports:
The interview documents what Fauci actually said at the time and provides context for the later controversy over changing mask recommendations.

Important caveat:
The interview establishes Fauci’s statements, not whether every aspect of the early mask policy was scientifically justified. Evidence and understanding of asymptomatic transmission were developing rapidly.

3. School disruption and learning loss

Claim or topic:
Students experienced substantial educational disruption during the pandemic, with virtual instruction presenting significant difficulties, particularly for some vulnerable students.

Source:
U.S. Government Accountability Office and GAO report on pandemic learning strategies

Source type:
Government research and nationally generalizable teacher survey.

What it supports:
GAO found that many teachers reported more students beginning the 2020-21 school year behind and making less academic progress than during a typical year. GAO also documented significant challenges associated with virtual learning.

Important caveat:
Learning loss cannot be attributed entirely to school closures or remote instruction. Illness, family disruption, economic stress and other pandemic effects also contributed.

4. NIH, EcoHealth Alliance and Wuhan research oversight

Claim or topic:
NIH funded EcoHealth Alliance, which provided subaward funding for coronavirus research at the Wuhan Institute of Virology, and federal auditors later identified significant weaknesses in oversight.

Source:
HHS Office of Inspector General

Source type:
Primary government audit.

What it supports:
The inspector general concluded that NIH and EcoHealth did not effectively monitor relevant awards and subawards, resulting in missed opportunities to oversee research. The audit included EcoHealth subawards to the Wuhan Institute of Virology.

Important caveat:
An oversight failure does not establish that the funded research created SARS-CoV-2 or caused the pandemic. Those are separate claims requiring separate evidence.

5. Covid’s origins remain unresolved

Claim or topic:
U.S. intelligence agencies have differed over the most likely origin of Covid-19. In 2025, the C.I.A. assessed a laboratory-related origin as more likely, but with low confidence.

Source:
Reuters

Source type:
Reputable journalism reporting an intelligence assessment.

What it supports:
Reuters reported that the C.I.A. shifted toward a laboratory origin as the more likely explanation while emphasizing that the agency had low confidence in that conclusion.

Important caveat:
“Low confidence” means the evidence is limited or uncertain. A possible laboratory origin also does not establish that NIH-funded EcoHealth research produced the pandemic virus.

6. Fauci’s deletion-language emails

Claim or topic:
Senate investigators disclosed pandemic-era messages in which Fauci instructed recipients to delete an email after reading it, raising legitimate questions about federal recordkeeping.

Source:
U.S. Senate Homeland Security and Governmental Affairs Committee

Source type:
Primary congressional document.

What it supports:
The committee document reproduces and discusses the deletion language that became part of the congressional investigation into Fauci’s communications and recordkeeping.

Important caveat:
The messages justify investigation but do not, by themselves, prove systematic destruction of required federal records, criminal conduct or concealment of Covid’s origin.

7. Fauci’s 2026 Senate testimony and Fifth Amendment invocation

Claim or topic:
At a July 29, 2026 Senate hearing chaired by Rand Paul, Fauci invoked the Fifth Amendment more than 100 times while being questioned about the pandemic response and related controversies.

Source:
Reuters and Reuters legal analysis

Source type:
Reputable journalism and legal analysis.

What it supports:
Reuters documented the hearing, Fauci’s repeated Fifth Amendment invocations, Paul’s previous efforts to seek his prosecution and the unresolved legal questions surrounding Fauci’s refusal to testify.

Important caveat:
Invoking the Fifth Amendment is a constitutional protection and is not evidence of guilt. The legal consequences of Fauci’s invocation remain disputed.

8. Biden’s pardon and the subsequent contempt dispute

Claim or topic:
President Biden pardoned Fauci for potential federal offenses arising from specified government service between Jan. 1, 2014, and Jan. 19, 2025. In August 2026, the Senate committee voted along party lines to hold Fauci in contempt, and Paul sought a Justice Department prosecution.

Source:
Text of President Biden’s pardon and Reuters on the 2026 contempt vote

Source type:
Primary presidential document and reputable journalism.

What it supports:
The pardon establishes its precise scope and dates. Reuters documents the later party-line committee vote, Paul’s referral to the Justice Department and the political and legal dispute surrounding Fauci’s testimony.

Important caveat:
The pardon does not establish that Fauci committed a crime. Nor does the contempt action establish the underlying allegations against him. Questions also remain about the legal force and procedure of the Senate referral.

How to read this evidence

Corrections and updates

Subscribe now

Are Republicans Starting to Turn on Trump?

Evidence First — Wed, 05 Aug 2026 12:08:47 GMT

Donald J. Trump’s influence over the Republican Party can appear nearly absolute.

Candidates compete for his endorsement. Those who receive it often advertise it more prominently than endorsements from governors, senators or major conservative organizations. Republican officeholders who defy him risk primary challenges, attacks on social media and accusations that they have betrayed the party’s voters.

The election results seem to confirm that power. As of July 10, 2026, 220 of the 227 candidates Trump had endorsed had won their primaries, according to Ballotpedia, a success rate of 97 percent. That was close to his 98 percent rate in 2020 and above the 93 percent recorded in 2022.

But those numbers tell only part of the story.

Trump remains the dominant figure in Republican primary politics, and his endorsement may be the most valuable single endorsement available to a Republican candidate. At the same time, public polling suggests that portions of the coalition that returned him to the White House in 2024 have become less enthusiastic about his performance.

The dissatisfaction is concentrated not among the most committed supporters of the Make America Great Again movement, but among independents, less ideological Republicans and some working-class voters who supported him without developing a strong personal attachment to him.

The result is a political picture that is less contradictory than it first appears: Trump can be losing support among the broader public while retaining immense power inside the Republican nomination process.

That distinction will shape the 2026 midterm elections. It could become even more important in 2028, when Trump, constitutionally barred from seeking another term, may try to help select his successor.

A coalition under strain

National polls now show Trump significantly less popular than he was near the beginning of his second term.

A Reuters/Ipsos poll released on Aug. 3 measured his approval rating at 35 percent, down from about 47 percent shortly after his January 2025 inauguration. An Economist/YouGov survey conducted in late July found 34 percent approving and 62 percent disapproving.

Polls differ because they use different methods, samples and question wording. A single result should not be treated as a precise reading of the country. But when several polling organizations show movement in the same direction over time, the broader trend becomes more persuasive.

That trend is especially clear among independents. YouGov found that Trump’s net approval, calculated by subtracting disapproval from approval, had fallen among independent voters from minus 4 near the start of his term to minus 53 in late July.

Independents are not a unified political group. Some consistently vote Democratic, some consistently vote Republican, and others move between the parties. But Republican-leaning independents were an important part of Trump’s 2024 coalition. Their movement away from him helps explain how his national standing can deteriorate even while most self-identified Republicans continue to approve of him.

The clearest evidence of erosion comes from voters who say they supported Trump in 2024. In the late-July Economist/YouGov poll, 77 percent of those voters approved of his performance, while 21 percent disapproved. At the beginning of his term, his approval among his 2024 voters was about 93 percent.

That is not a mass abandonment. More than three-quarters still approved. But it is a meaningful change: roughly one in five people who reported voting for Trump now said they disapproved of the job he was doing.

There are limits to that comparison. The surveys did not necessarily interview the same people at both points in time, and some respondents may inaccurately recall or report how they voted. Still, a decline of that size, reinforced by other polling, is difficult to dismiss as random fluctuation.

The MAGA core holds

The erosion is not evenly distributed.

Among Republicans who identify with MAGA, Trump’s support remains overwhelming. YouGov reported a net approval rating of about plus 89 among that group. Among Republicans who did not identify with MAGA, the figure was only plus 12, down sharply from earlier in the term.

That contrast helps explain Trump’s continuing control of Republican primaries.

Primary voters are not a representative sample of the country. They tend to be more politically attentive, more ideological and more loyal to their party than general-election voters. In Republican contests, the electorate is also disproportionately favorable to Trump. A candidate can therefore benefit enormously from Trump’s endorsement even when Trump’s approval with the public as a whole is weak.

There is an important methodological complication in comparisons between MAGA and non-MAGA Republicans. “MAGA” is a self-description. A Republican who becomes unhappy with Trump may stop using the label, making the non-MAGA category more negative partly because dissatisfied voters have moved into it.

Even with that caution, the broader conclusion is difficult to miss: Trump’s strongest supporters remain intensely loyal, while his less committed supporters are much more open to criticism.

The same pattern appears in public reaction to individual policies.

An AP-NORC survey conducted in late July found that 37 percent of Republicans believed the war with Iran had not been worth fighting. Trump’s approval among Republicans for his handling of Iran was about 61 percent, down from 71 percent in June. Roughly three in 10 Republicans favored either pausing military action for cease-fire negotiations or ending it immediately.

The Republican sample in that poll had a margin of error of about six percentage points, meaning the exact size of the decline is uncertain. But the survey still demonstrated something politically significant: a substantial minority of Republicans were willing to reject a major Trump policy even if they had not rejected Trump himself.

That is better understood as issue-specific dissent than as a party revolt.

A voter may oppose a war, dislike tariffs or feel disappointed by economic conditions while still voting for Republican candidates. Disapproval does not automatically lead to defection. Some dissatisfied Republicans may stay home. Others may vote Republican despite their misgivings. Still others may blame Congress, advisers or outside events rather than Trump.

Why a 97 percent endorsement record can mislead

Trump’s endorsement victories are real. But the raw win rate makes his political power look more measurable than it actually is.

Of the 227 completed endorsements in Ballotpedia’s July tally, 159 went to incumbents, meaning candidates who already held office. Forty-two were in open-seat races, and only 26 involved challengers running against incumbents. Trump also occasionally endorsed more than one candidate in the same race.

That matters because incumbents usually begin with substantial advantages: name recognition, established donors, campaign organizations and relationships with local party officials. Endorsing a candidate who is already likely to win does not prove that the endorsement caused the victory.

This is a common problem in political analysis known as selection bias. The people receiving the treatment, in this case Trump’s endorsement, are not chosen randomly. Trump and his advisers can evaluate polls, fundraising reports and local political conditions before deciding whom to support. They have an incentive to endorse likely winners and avoid candidates headed for defeat.

A more revealing test is how Trump-backed non-incumbents perform in genuinely contested primaries. By that measure, his record has still been impressive, though less overwhelming. Among endorsed non-incumbents in contested races for governor, the Senate and the House, 17 of 19 won or advanced in 2018. The figure was 24 of 25 in 2020, 37 of 45 in 2022 and 37 of 45 in 2024.

Those results suggest that Trump’s endorsement often helps. They do not show that it guarantees victory.

His defeats are instructive. In 2026, Trump-backed candidates lost several state-level contests, including high-profile races in Iowa, South Carolina and Georgia. In Iowa’s Republican primary for governor, Representative Randy Feenstra lost despite Trump’s support and an early position as the presumed front-runner. Local concerns, including economic and agricultural issues, appeared to outweigh the endorsement.

Trump’s backing is most useful when voters know little about the candidates. In a low-profile race, an endorsement functions as a shortcut: a Republican voter who trusts Trump can assume that his chosen candidate shares Trump’s views.

Research on Georgia Republican primaries found that awareness of Trump’s endorsement could significantly affect voter choices in less prominent contests, while producing more varied effects in races where candidates were already well known.

The endorsement can also bring indirect benefits. It attracts news coverage, donors, campaign volunteers and attention from conservative media. A candidate who had been polling in the low single digits can suddenly become a central figure in the race.

J.D. Vance’s 2022 Senate campaign in Ohio is a frequently cited example. Vance had struggled to separate himself from a crowded field before Trump endorsed him. He then won the Republican nomination.

The endorsement was probably important, though it was not the only event in the campaign. Advertising, donor support, media attention and the weaknesses of his opponents also mattered.

Winning the primary is not the same as winning the election

Trump’s endorsement is more clearly valuable in Republican primaries than in general elections.

In a primary, the endorsement can tell voters which candidate is most closely aligned with the party’s dominant figure. In November, the same endorsement can have two opposing effects: it can energize Republicans while also motivating Democrats and anti-Trump independents.

That is why general-election win rates are difficult to interpret. Many Trump-endorsed candidates run in heavily Republican districts where almost any Republican nominee would be favored. A victory in such a district says little about whether the endorsement increased the candidate’s support.

Research on Trump’s 2018 endorsements found that he tended to choose candidates who already had stronger chances of winning. After researchers accounted for that tendency, the endorsement did not clearly improve Republican electoral outcomes and appeared to stimulate fundraising by the opposition.

The central test in 2026 is therefore still ahead. Trump’s candidates have performed exceptionally well in Republican primaries, but the November elections will show whether those nominees can win competitive states and districts.

A candidate who excites the Republican base may still be poorly suited to a general-election electorate. Conversely, a candidate who appears too moderate for primary voters may be more competitive with independents. Trump’s influence gives him considerable power over which side of that trade-off the party chooses.

The coming succession contest

The same dynamic could dominate the 2028 presidential campaign.

Trump cannot run for another term under the Constitution’s 22nd Amendment. But if he endorses a successor, that person would probably begin with a major advantage.

A May 2026 Marquette Law School poll asked Republican voters to choose between a candidate endorsed by Trump and a Republican incumbent opposed by him. Seventy-one percent selected Trump’s candidate, while 20 percent chose the incumbent. Among MAGA Republicans, the split was 87 percent to 9 percent.

Among non-MAGA Republicans, the result was very different: 30 percent chose Trump’s candidate, while 48 percent preferred the incumbent.

A University of New Hampshire poll offered another qualification. About two-thirds of likely Republican primary voters said the party’s next presidential nominee should build on Trump’s presidency, but fewer than half said a Trump endorsement would make them more likely to support a particular candidate.

Together, the surveys suggest that Republican voters want substantial continuity with Trump but may not automatically accept his choice of successor.

The endorsement would probably have its greatest effect in a crowded field. If several candidates competed for voters seeking a post-Trump alternative, the endorsed candidate could consolidate the MAGA vote and win early primaries with a plurality rather than a majority.

In a close two-candidate race, Trump’s endorsement could be decisive. Against a well-known front-runner with an independent political identity, its effect would be less certain. And if Trump simply endorsed the person already leading, a subsequent victory would reveal little about whether he had shaped the race or merely recognized its likely outcome.

His influence in the general election would be even harder to predict. Trump could help unify Republicans behind the nominee and persuade reluctant supporters that the candidate was a legitimate heir to his movement. But his endorsement could also make it difficult for the nominee to distance himself or herself from unpopular parts of Trump’s record.

A survey experiment published in 2024 found that describing a hypothetical Republican congressional candidate as Trump-endorsed reduced overall willingness to vote for the candidate by about four percentage points. The result was driven largely by increased opposition among Democrats and independents. There was no statistically clear overall increase among Republicans.

A hypothetical House candidate is not the same as a presidential nominee, and a survey experiment is not an election. But the finding illustrates the basic tension: Trump’s endorsement can be a powerful asset among the voters who select the Republican nominee and a liability among some of the voters who decide the presidency.

For now, the evidence points toward a party that remains firmly under Trump’s control but a national coalition that is less secure.

His MAGA base is intact. His endorsed candidates continue to dominate Republican primaries. Yet some 2024 voters, non-MAGA Republicans and Republican-leaning independents are becoming more critical of his performance, and dissatisfaction with specific policies is increasingly visible.

That does not mean Republicans are broadly turning against him. It means the less committed edges of his coalition are fraying while the core remains strong enough to control the party’s internal choices.

Trump may retain the power to shape who Republicans nominate in 2028. Whether he can help that nominee win the country is a different question.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Trump’s declining national approval

Claim or topic:
Trump’s national approval rating has fallen substantially since the beginning of his second term.

Source:
Reuters/Ipsos

Source type:
Reputable journalism reporting primary polling.

What it supports:
The Reuters/Ipsos poll measured Trump’s approval at 35 percent in early August 2026, compared with approximately 47 percent shortly after his January 2025 inauguration.

Important caveat:
Approval estimates vary across polling organizations because of differences in samples, weighting, question wording and survey methods. The trend across multiple polls matters more than any single result.

2. Erosion among independents and previous Trump voters

Claim or topic:
Trump has lost support among independents and among some voters who say they supported him in 2024.

Source:
Economist/YouGov poll report and YouGov’s analysis

Source type:
Primary polling document and analysis.

What it supports:
The poll found 77 percent approval and 21 percent disapproval among respondents who reported voting for Trump in 2024. It also showed a sharp deterioration in his net approval among independents.

Important caveat:
These are repeated cross-sectional surveys, not necessarily interviews with the same individuals over time. Self-reported past voting can also contain recall or reporting errors.

3. The divide between MAGA and non-MAGA Republicans

Claim or topic:
Trump’s most committed MAGA supporters remain highly loyal, while Republicans who do not identify with MAGA are considerably less supportive.

Source:
YouGov

Source type:
Polling analysis.

What it supports:
YouGov reported overwhelming net approval among MAGA Republicans but much weaker approval among non-MAGA Republicans.

Important caveat:
“MAGA” is a current self-description. People who become dissatisfied with Trump may stop using the label, which can exaggerate the apparent difference between the two groups.

4. Republican dissent over the Iran war

Claim or topic:
A significant minority of Republicans opposed or questioned Trump’s handling of the war with Iran.

Source:
Associated Press and AP-NORC

Source type:
Expert organization polling reported by reputable journalism.

What it supports:
The survey found that 37 percent of Republicans believed the war had not been worth fighting. It also found weaker Republican approval of Trump’s handling of Iran than in the previous survey.

Important caveat:
The Republican subgroup had a relatively large margin of sampling error. The poll demonstrates substantial dissent, but the exact size of the change is uncertain.

5. Trump’s 2026 endorsement win rate

Claim or topic:
Trump-endorsed candidates won approximately 97 percent of their completed 2026 primaries as of July 10.

Source:
Ballotpedia

Source type:
Analysis based on election results and endorsement tracking.

What it supports:
Ballotpedia reported that 220 of 227 Trump-endorsed candidates had won their primaries. It also showed that most of the endorsed candidates were incumbents.

Important caveat:
The raw success rate does not measure how many victories were caused by the endorsement. Trump often backs incumbents, favorites and candidates in safe Republican constituencies.

6. Trump’s record in genuinely contested primaries

Claim or topic:
Trump-backed non-incumbents have performed strongly in contested Republican primaries, though not at the nearly perfect rate suggested by the broader endorsement totals.

Source:
ABC News and FiveThirtyEight analysis

Source type:
Analysis and reputable journalism.

What it supports:
The analysis found that endorsed non-incumbents won or advanced in 17 of 19 relevant races in 2018, 24 of 25 in 2020, and 37 of 45 in both 2022 and 2024.

Important caveat:
These figures still do not isolate the endorsement’s causal effect. Endorsed candidates may also benefit from fundraising, media attention, ideological fit or weaknesses among their opponents.

7. What research says about endorsement effects

Claim or topic:
Trump’s endorsement can influence Republican primary voters, particularly in lower-profile races, but it does not necessarily improve general-election outcomes.

Source:
State Politics & Policy Quarterly and Legislative Studies Quarterly

Source type:
Academic research.

What it supports:
The studies provide evidence that endorsement awareness can affect choices in Republican primaries, especially in less visible contests. They also show that Trump selects candidates strategically and that his endorsement may mobilize opposition without clearly improving general-election results.

Important caveat:
The studies examine particular elections and types of candidates. Their findings cannot be applied mechanically to every future race, especially a presidential election.

8. Trump’s potential influence over the 2028 nomination

Claim or topic:
Trump’s endorsement could be highly influential in the 2028 Republican primary, but Republican voters may not automatically follow his choice.

Source:
Marquette Law School Poll and University of New Hampshire Survey Center

Source type:
Primary polling and expert organization research.

What it supports:
Marquette found that most Republican respondents preferred a Trump-endorsed candidate over an incumbent Trump opposed. The University of New Hampshire found broad support for continuing Trump’s political direction, but fewer than half said his endorsement alone would make them more likely to support a candidate.

Important caveat:
The 2028 field, political environment and endorsed candidate are unknown. Hypothetical polling measures current attitudes, not how voters will behave once they face real candidates and campaign events.

How to read this evidence

Corrections and updates

Baseball’s Competitive-Balance Problem Is Not Quite What It Seems

Evidence First — Tue, 04 Aug 2026 14:24:23 GMT

Tarik Skubal joined the Dodgers at the trade deadline, intensifying debate over how baseball’s richest and deepest teams accumulate talent.

Tarik Skubal did not sound like a man celebrating a promotion.

After the Detroit Tigers traded him to the Los Angeles Dodgers on Aug. 2, the two-time defending American League Cy Young Award winner fought back tears as he described leaving the organization that had drafted and developed him. He had imagined winning a championship in Detroit. Instead, he joined the two-time defending World Series champions, who were leading their division by 10 games and already tied for the best record in baseball. The Tigers, meanwhile, were only 2½ games from an American League wild-card position.

The transaction appeared to capture everything critics say is wrong with Major League Baseball: A borderline contender surrendered its best player, while the sport’s wealthiest and most formidable organization added another ace to an already imposing roster.

It was, in the familiar phrase, a case of the rich getting richer.

But the Skubal trade also illustrates why baseball’s competitive-balance debate is harder than it first appears. The sport has enormous financial disparities, yet its games are unusually unpredictable. The Dodgers can outspend almost everyone, but money alone did not produce this trade. And although a salary cap might weaken their ability to assemble a roster this expensive, it would not necessarily stop a contender from exchanging prospects for a star whose contract is about to expire.

Baseball has a real problem. It is simply not one problem.

More Financially Unequal, but Not Necessarily Less Competitive

To understand the paradox, it helps to separate financial balance from competitive balance.

Financial balance concerns what teams can afford to spend. On that measure, baseball is plainly lopsided. Unlike the National Football League, National Basketball Association and National Hockey League, MLB has no system combining a leaguewide payroll ceiling with a meaningful minimum. It instead uses a competitive balance tax, commonly called the luxury tax, which makes high payrolls more expensive without prohibiting them.

The spending gap has grown. According to MLB’s own analysis, the average payroll of the five highest-spending teams in 2025 was 4.8 times the average of the five lowest, the widest ratio in the league’s data going back to at least 1985. That analysis comes from a league actively arguing for a salary cap, so its policy conclusions should not be treated as neutral. But the disparity itself is substantial.

In 2026, the Dodgers opened the season with a luxury-tax payroll of roughly $415 million. Miami’s was about $82 million. Los Angeles’s total was so far above the $245.3 million cap proposed by MLB that the club would have needed a lengthy transition to comply.

Yet payroll inequality is not the same thing as unequal game outcomes. A study examining the NBA, NFL, NHL and MLB from 1991 through 2018 attempted to account for the leagues’ radically different season lengths and scoring environments. It estimated that the better team in an average baseball matchup won only 56.6 percent of the time. The corresponding estimate was 59 percent in hockey and about 67 percent in both basketball and football.

In a theoretical matchup between the league’s best and worst teams, the top baseball club had an estimated 68.5 percent chance of winning. In the NBA and NFL, that figure was about 96 to 97 percent. The study concluded that MLB displayed the greatest balance in underlying game-level strength, followed by the NHL.

These are historical model estimates, not eternal laws. Competitive balance can also be measured by championship concentration, year-to-year turnover or the number of meaningful late-season games, and different measures can produce different rankings. Still, the central point is important: A weak baseball team has a fairly good chance of beating a great one on any particular night. Starting pitchers change daily. A few well-timed hits can decide a low-scoring game. Even the best hitter fails most of the time.

Baseball’s 162-game schedule gradually reveals differences in quality, but the sport itself contains a great deal of randomness. That helps explain how MLB can look financially alarming while remaining more competitive on the field than the NBA, where one superstar can influence nearly every possession.

That does not make the concern imaginary. Leaguewide parity can coexist with individual moments that reveal how financial and organizational advantages accumulate. The Skubal trade was one of those moments.

What the Skubal Trade Actually Shows

Skubal’s move to Los Angeles is best understood as an example of upward talent concentration: elite players migrating from uncertain or short-lived competitive situations toward organizations already positioned to win.

But it is not a pure example of a large-market team purchasing a player a small-market team could not afford. Skubal was under contract only through the end of the season. The Dodgers did not win a free-agent bidding war for a nine-year contract. They traded three young players, outfielder Zyhir Hope and pitchers River Ryan and Brady Smith, for two months of Skubal and whatever he might contribute in October.

That distinction matters because the immediate currency was not merely cash. It was organizational depth. The Dodgers could afford to surrender highly regarded young players without emptying their farm system. They could absorb Skubal’s remaining salary, tolerate the possibility that he would leave after the season and accept the injury risk inherent in any pitcher. They also had an unusually strong reason to prioritize the present: the chance to win a third consecutive championship.

Detroit faced the reverse calculation. Keeping Skubal offered a better chance of reaching the 2026 postseason, but it also risked losing him for comparatively limited compensation after the season. Under the current qualifying-offer system, a team that retains an eligible player approaching free agency and then loses him can receive a draft pick. But a concrete package of three prospects may be far more attractive than a single future selection.

Nothing in the available evidence proves that Detroit was literally incapable of extending Skubal. Financial inability, unwillingness to meet his expected price and a judgment that the prospects were more valuable are different explanations. The Tigers’ decision may have been rational.

That is precisely the concern. A system can produce troubling leaguewide results even when every individual decision makes sense.

The Dodgers should not be faulted for using the rules effectively. Skubal himself noted that other teams could have tried to acquire him, and several reportedly did. But theoretical access is not the same as equivalent capacity. Los Angeles combined money, a deep roster, a strong farm system, deferred contracts and a willingness to assume risk in a way few clubs could match. After adding Skubal, the Dodgers employed five of baseball’s 13 highest-paid starting pitchers and carried roughly $1.1 billion in deferred salary obligations.

The issue, then, is not simply that the Dodgers spend lavishly. It is that they can accumulate several forms of advantage at once: payroll, prospects, roster depth and the ability to accept risk. When one organization can combine those advantages on such a scale, a payroll ceiling can seem like the simplest answer.

The Temptation of a Salary Cap

MLB owners have proposed just such a system. Beginning in 2027, their plan would establish a $245.3 million payroll ceiling and a $171.2 million floor, using the league’s luxury-tax accounting. That accounting includes benefits and other costs beyond salaries paid directly to major-league players. MLB has also proposed centralizing local-media revenue, dividing it equally among the 30 clubs and allocating players 50 percent of defined baseball revenue.

A system like that would unquestionably compress payrolls. Based on opening-day figures, eight teams would have needed to cut spending, while 12 would have been required to add a combined $617 million. The Dodgers could not maintain their current roster indefinitely under such a ceiling.

But calling that a complete solution is too easy. A cap would constrain the Dodgers’ financial power over time, yet it would not directly prevent a team from trading prospects for an affordable player approaching free agency. In some circumstances, a cap could even make stars more likely to move if their original clubs lacked room for long-term extensions.

Nor does a cap guarantee competent ownership. Teams can mismanage equal budgets, keep shared revenue or spend a payroll floor inefficiently. Salary restrictions may improve payroll equality while also reducing the share of revenue paid to players. Those are separate questions, and a policy that helps owners control labor costs is not automatically a competitive-balance reform.

The players’ union has therefore proposed a different approach: greater revenue sharing, a higher minimum salary, earlier access to salary arbitration, expanded compensation for productive young players and a competitive integrity tax on clubs that fail to reach minimum payroll benchmarks. Salary arbitration is a process in which eligible players can argue that their performance justifies a higher salary. Most players do not qualify for it during their first few major-league seasons.

The union’s proposal would also penalize teams that receive revenue-sharing money but fail to use enough of it on payroll. MLB argues that the union’s plan would weaken restraints on the highest spenders, while the union argues that a cap would primarily increase owner profits. Both sides have obvious financial interests in how they describe the dispute.

The real choice, then, is not simply between a cap and no cap. It is between reforms that address only payrolls and those that address the wider system that moves money, prospects and stars toward the same clubs.

A More Targeted Fix

The most direct place to begin is local-media revenue.

At present, teams can earn very different amounts from the television and streaming rights to games shown in their home markets. Consider the Dodgers and Tigers. The Dodgers’ local-media agreement is valued at about $334 million per year on average, while the Tigers’ previous publicly reported television agreement paid about $50 million per year. The Dodgers number is the yearly average from a long contract, while the Tigers number comes from an older deal. The comparison is not exact, but it shows how much more one team can make from local TV than another. Before either club sells a ticket or makes a roster decision, one can begin with a local-media advantage worth hundreds of millions of dollars.

Under a more centralized system, MLB would collect a larger share of local television and streaming revenue and distribute much of it equally among all 30 clubs. The league might still allow teams to keep part of the revenue they generate, preserving an incentive to attract viewers and market the club. But the starting gap would be smaller.

That would not guarantee equal payrolls or equally successful teams. An owner could still spend poorly or refuse to invest. But it would reduce a structural disadvantage that has little to do with scouting, coaching or smart decision-making, and it would make it more plausible for lower-revenue clubs to retain star players.

Greater revenue sharing, however, would accomplish little if some owners simply kept the additional money. That is why MLB would also need a genuine payroll floor, preferably one tied automatically to league revenue. It should measure meaningful player spending, not allow teams to satisfy the requirement primarily through benefits or accounting devices. Revenue-sharing payments should also come with transparent requirements showing that clubs are investing in their major-league teams, player development and other competitive operations.

Even with more shared revenue and a payroll floor, Detroit might still have faced the same choice with Skubal: keep him for a few months or trade him before losing him. To change that calculation, MLB could give teams more value for retaining elite players approaching free agency. Stronger draft compensation for a lower-revenue club that keeps and then loses a star would make selling less irresistible. A centrally funded credit for extending homegrown players could also help teams keep the stars they developed without reducing the players’ salaries.

If better compensation would give Detroit more reason to keep Skubal, an earlier trade deadline would give the Dodgers fewer chances to acquire him. It would force borderline contenders to decide whether to sell before the standings fully clarify, reducing the supply of elite short-term acquisitions available to dominant clubs. The trade-off is that genuinely weak teams might receive smaller returns or lose opportunities to rebuild. But unlike a salary cap alone, an earlier deadline would address the mechanism that produced the Skubal trade.

Baseball could also pay productive young players more quickly. The current labor system keeps many of them near the minimum salary early in their careers and under team control for roughly six years. Better bonuses for players who have not yet reached salary arbitration, along with earlier access to arbitration, would better align compensation with performance. This would not directly prevent an elite rental from joining a contender, but it would make the broader system less dependent on years of inexpensive young labor followed by costly free agency.

Taken together, these reforms address three different weaknesses: unequal resources, weak incentives to retain stars and a trade system that can move elite players toward teams already built to win. No single rule would eliminate baseball’s financial disparities or prevent every star from joining an established contender. The goal is not to stop smart teams from improving. It is to make continued competition a more reasonable choice for everyone else.

None of these changes can happen without a new agreement between owners and players. That is why the debate over competitive balance is now inseparable from the threat of a lockout.

A Lockout Is Not a Reform

The current labor agreement expires Dec. 1, and MLB is expected to impose a lockout if no new deal is reached. A lockout is an action taken by owners. It freezes signings, trades and other league business in an effort to pressure players into accepting a new labor agreement. It is different from a strike, which is initiated by players.

The league’s previous lockout lasted 99 days and ended in March 2022 without canceling the 162-game regular season. The much longer 1994-95 dispute, which was a player strike, canceled the World Series and remains the sport’s enduring warning about labor warfare.

A lockout could give owners leverage to obtain a cap. It could give players leverage if they remain united and force concessions on revenue sharing, minimum salaries or early-career pay. It could end with a more balanced system. But the lockout itself would solve nothing. It is an economic weapon intended to change the bargaining position of the other side, and the same beneficial reforms could, in principle, be negotiated without losing games.

A brief lockout that ends with greater revenue sharing, a real payroll floor, higher compensation for young players and better incentives for teams to retain their stars could eventually improve the sport. A long lockout that produces only a salary cap could mainly reduce player compensation while leaving many of the forces behind talent concentration intact.

Whatever emerges from the negotiations should be judged by a simple test: Would it make a decision like Detroit’s less inevitable?

The lesson of the Skubal trade is not that baseball needs to stop ambitious teams from being ambitious. Well-run organizations should be rewarded. The problem arises when the system makes it consistently rational for a club near the playoff race to trade its best player, while allowing one organization to combine unmatched spending, prospect depth and risk tolerance indefinitely.

Baseball does not need every team to finish 81-81. It needs every well-run team to have a credible chance to keep the players it develops, and every owner to face a real obligation to try.

That outcome may require limits, revenue sharing and new incentives. It does not require pretending that a salary cap is a cure, or that a lockout is anything more than the painful route the sport may take to reach an agreement.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. The Tarik Skubal trade

Claim or topic:
Detroit traded Tarik Skubal to the Dodgers while the Tigers remained close to a wild-card position, and Skubal expressed disappointment about leaving.

Source:
Reuters

Source type:
Reputable journalism.

What it supports:
The report documents the trade, Detroit’s position in the standings, Skubal’s reaction and the immediate competitive context.

Important caveat:
The article uses the trade as an example of a broader structural issue. One transaction alone cannot establish a leaguewide pattern.

2. The players included in the trade

Claim or topic:
The Dodgers acquired Skubal by trading outfielder Zyhir Hope and pitchers River Ryan and Brady Smith.

Source:
MLB

Source type:
Expert organization and league reporting.

What it supports:
The source identifies the players exchanged and provides background on their prospect status and organizational value.

Important caveat:
Prospect rankings are estimates of future value, not guarantees of major-league success.

3. MLB’s payroll disparity

Claim or topic:
The average payroll of MLB’s five highest-spending teams was 4.8 times that of the five lowest-spending teams in 2025.

Source:
MLB

Source type:
League analysis.

What it supports:
The analysis provides the payroll comparison and places the gap in historical context.

Important caveat:
MLB is advocating for a salary cap, so its interpretation of the numbers is not neutral. The payroll figures can still be evaluated separately from the league’s preferred policy.

4. Competitive balance across major sports

Claim or topic:
Historical research found that the stronger team wins an average MLB game less often than the stronger team wins in the NBA or NFL.

Source:
Academic study on measuring competitive balance in sports

Source type:
Academic research.

What it supports:
The study estimates game-level differences in team strength across MLB, the NBA, NFL and NHL while accounting for differences in season length and scoring.

Important caveat:
The estimates cover 1991 through 2018 and depend on the researchers’ model. Other measures, such as championship concentration or year-to-year turnover, may produce different rankings.

5. The qualifying-offer system

Claim or topic:
A team that retains an eligible player through the season and loses him in free agency may receive draft-pick compensation, while a player traded during the season cannot receive a qualifying offer.

Source:
MLB qualifying-offer glossary

Source type:
Primary league rules and expert organization.

What it supports:
The source explains the system that affects the value of keeping or trading a player approaching free agency.

Important caveat:
The value of a compensation pick varies. It may be worth much less to a team than a package of several prospects.

6. Local-media revenue differences

Claim or topic:
The Dodgers’ long-term local-media agreement averages about $334 million per year, while the Tigers’ previous publicly reported television agreement paid about $50 million per year.

Source:
Forbes on the Dodgers’ media agreement and Bless You Boys on the Tigers’ broadcast arrangements

Source type:
Analysis and sports journalism.

What it supports:
The sources illustrate how far apart local television agreements can be and why local-media revenue is central to baseball’s financial-balance debate.

Important caveat:
The figures are not a precise current comparison. The Dodgers number is an average over a long contract, while the Tigers number comes from an older agreement whose replacement terms were not publicly disclosed.

7. The proposed salary cap, payroll floor and revenue system

Claim or topic:
MLB owners proposed a payroll ceiling, a payroll floor, centralized local-media revenue and a defined revenue split with players.

Source:
Associated Press and MLB

Source type:
Reputable journalism and league proposal.

What it supports:
The sources describe the principal features of the owners’ proposal and estimate how many teams would need to raise or reduce payroll.

Important caveat:
A proposal is not an enacted rule. Its effect would depend on definitions, accounting methods, enforcement and the final negotiated terms.

8. The players’ union proposal and labor dispute

Claim or topic:
The players’ union proposed greater revenue sharing, higher minimum pay, expanded early-career compensation and penalties aimed at chronic low spending.

Source:
Associated Press and MLB Players Association

Source type:
Reputable journalism and primary labor-organization statement.

What it supports:
The sources outline the union’s alternative approach and provide context for the possibility of another lockout.

Important caveat:
The union represents players and has a direct financial interest in the negotiations. Its claims about the likely effects of a cap should be considered alongside independent reporting and the owners’ proposal.

How to read this evidence

Corrections and updates

Subscribe now

What Makes Children More Resilient?

Evidence First — Fri, 31 Jul 2026 17:24:56 GMT

At 8:17 on a wet Thursday morning, 10-year-old Maya refused to get out of the car.

Her mother, Elena, had already turned off the engine. Children in bright raincoats hurried toward the school entrance, their backpacks bouncing behind them. A crossing guard waved traffic forward. Somewhere in the line of cars, a horn sounded.

Maya stared at her shoes.

The previous afternoon, two girls she considered friends had told her she could not sit with them at lunch. That evening, she complained of a stomachache. Now she said she was never going back to school.

Elena felt the familiar adult urge to make the problem smaller. “They’re just being mean,” she nearly said. “You can’t let them get to you.”

She was late for work. Maya had a math quiz. The school doors would close in three minutes.

But Elena stopped.

“That really hurt,” she said instead.

Maya began to cry.

The scene is a composite, not the story of one identifiable family. But the dilemma is common. A child encounters something painful, and an adult must decide what resilience is supposed to look like.

Is it getting out of the car? Ignoring the girls? Learning not to cry?

For years, resilience has been marketed to parents and schools as a kind of psychological armor. Children are told to develop grit, adopt a positive attitude and bounce back. The language suggests that hardship is a test of personal strength, and that successful children are those who remain steady under pressure.

Developmental research points to a different picture. Resilience is not invulnerability. It is not a personality type, and it is not proof that adversity caused no harm. It is the capacity to adapt, recover, seek help and continue developing despite difficulty.

One of the strongest foundations for that capacity is not toughness. It is the presence of at least one stable, committed and responsive adult.

That adult does not eliminate every disappointment or solve every problem. The more delicate task is to help a child bear what is happening without either abandoning the child to it or taking over completely.

In the car, Elena did not tell Maya she could stay home forever. She did not call the other girls’ parents from the parking lot. First, she helped her daughter settle.

They sat quietly for a moment. Elena asked Maya to put both feet on the floor. They took several slow breaths. Maya drank some water.

Then Elena asked, “What feels hardest about going inside?”

“Lunch,” Maya said.

That answer changed the problem. Maya was not refusing school in general. She was afraid of a particular hour in a particular place.

Together, they made a plan. Maya would go to class. Before lunch, she would tell her teacher what had happened. If she felt overwhelmed, she could eat in the counselor’s office that day. Elena walked her to the entrance.

This kind of response is sometimes called co-regulation. The term sounds clinical, but the process is ordinary. A child who cannot yet manage a surge of fear, anger or shame borrows some steadiness from an adult.

Children are not born knowing how to calm themselves, identify the source of distress and generate a workable plan. Those abilities develop through repeated experience. A responsive adult helps the child’s body settle, puts language around the emotion and supports the next step. Over time, the child begins to perform more of that sequence independently.

This does not mean agreeing with everything a child says or does. An adult can recognize fury without allowing hitting. A parent can acknowledge that school feels unbearable while still expecting attendance.

Validation is not surrender. It is an accurate recognition of the child’s experience. “You’re upset” is different from “You are right about everything.”

The distinction matters because children often cannot use advice until they feel understood. Commands to calm down can add frustration to distress. Reassurance can also misfire when it contradicts what the child is experiencing.

“It’s no big deal,” an adult may say.

To the child, it is.

A more useful response begins with what is true: “This is hard.” From there, the adult can help determine what kind of hard it is.

Some challenges are painful but manageable. A failed test, a lost game, an argument with a friend or disappointment over a role in a school play can become an opportunity to practice recovery.

Other adversities should not be reframed as growth opportunities. Bullying, abuse, family violence, unstable housing, hunger, discrimination and untreated illness are not exercises that children need in order to become stronger. They are conditions adults should work to reduce or stop.

That is one of the most important qualifications in the science of resilience. Coping skills can help a child survive a difficult environment. They do not make the environment acceptable.

A student who is being harassed may benefit from breathing exercises, but the school still has to intervene. A child living amid violence may learn ways to manage fear, but safety remains the central need. A teenager crushed by an impossible schedule may not need better time management so much as fewer demands.

When resilience is treated only as an inner trait, responsibility drifts downward. Institutions remain unchanged, caregivers receive another set of instructions, and children are asked to adapt to circumstances they did not create.

Public health approaches therefore place resilience within a broader system. Children do better when their environments are safer, their caregivers are supported and their families have access to food, housing, health care and dependable schools.

For Elena, the conversation with Maya did not end at the school door. That afternoon, she contacted the teacher.

The teacher had noticed tension at the lunch table but had assumed the children would resolve it. After speaking with Maya and the other students separately, she learned that the exclusion had happened repeatedly.

Now there were two tasks. Maya needed help deciding how to respond. The adults needed to address what was happening around her.

This is where popular advice about resilience often becomes confused. Adults are warned not to rescue children, and there is truth in that warning. Children need experience solving problems, tolerating frustration and recovering from mistakes. Constant intervention can unintentionally teach them that they cannot cope without someone taking control.

But refusing to intervene can be equally misguided. The goal is not maximum independence as quickly as possible. It is the right amount of support for the child and the challenge.

Researchers sometimes call this scaffolding, borrowing a term from construction. Temporary supports help a structure rise. As the structure becomes more stable, the supports are removed.

With children, scaffolding might mean asking, “What have you tried?” before offering a solution. It might mean helping a younger child name a feeling, rehearsing a difficult conversation with an older child or reviewing options with a teenager while leaving the final decision to the teenager.

The adult provides the smallest useful amount of help. First, the adult may do something for the child. Then the adult and child do it together. Eventually, the child does it independently while the adult remains available.

The challenges themselves must also be proportionate. Children build competence through difficulties that require effort but remain within reach.

A manageable struggle can teach, “I did not know what to do, but I figured out a next step.”

An overwhelming struggle may teach something else: “No matter what I do, I am on my own.”

In the days after the lunchroom episode, Maya practiced several specific skills. She named what she was feeling. She separated what she could control from what she could not. She considered different responses: confronting the girls, finding another place to sit, asking the teacher for help or inviting another classmate to lunch.

None of those options guaranteed that she would feel better. That was not the point.

Coping is not the power to produce a desired outcome. It is the ability to respond deliberately when the outcome is uncertain.

Children can be taught these skills. They can learn to notice the physical signs of rising distress, break a large problem into smaller pieces, consider consequences and ask for help. They can also learn to describe setbacks more accurately.

“I failed this test” is a statement about an event.

“I am stupid” is a judgment about identity.

Helping a child make that distinction is not empty positive thinking. It does not require pretending the test went well. It keeps one outcome from becoming a permanent conclusion about the self.

Research on programs designed to strengthen resilience offers cautious support for structured coping instruction, psychological therapies and physical activity, particularly among adolescents. But the evidence is not a catalog of miracle solutions.

Studies use different definitions of resilience. Programs vary in length and quality, and some trials are small. Improvements on questionnaires do not always show how children will function months later or under severe adversity.

Mindfulness, for example, may help some children notice emotions without reacting immediately. But reviews have found that its apparent benefits become less convincing when weaker studies are excluded. It may be a useful tool, not a universal remedy.

School-based programs show a similar pattern. A 2025 analysis of 38 randomized trials found a small average improvement in resilience. The results varied substantially among programs, and the researchers rated confidence in the overall estimate as very low.

That does not mean schools should abandon emotional skills lessons. It means that a weekly exercise cannot carry the entire burden.

A child may be taught to regulate emotions in the morning and then spend the afternoon in a chaotic, punitive or unsafe environment. A school may teach help-seeking while repeatedly failing to respond when students seek help. The surrounding culture can either reinforce the lesson or cancel it.

Resilience also grows through ordinary routines that are easy to overlook because they sound less impressive than a specialized program. Sleep, regular meals, movement, play, time outdoors and the chance to become competent at something can all make stress more manageable.

So can watching adults recover from their own mistakes.

A parent who says, “I’m too angry to answer well right now, so I’m going to take a minute,” demonstrates regulation in real time. A teacher who apologizes after reacting unfairly shows that authority and accountability can coexist.

No caregiver will model these things perfectly. Repair may matter almost as much as consistency.

An adult loses patience. Voices rise. Later, the adult returns and says, “I handled that badly. I’m sorry. Let’s try again.”

The child learns that conflict does not necessarily mean abandonment, and that mistakes can be followed by responsibility and reconnection.

But advice to caregivers must also reckon with the conditions under which caregiving occurs. A parent who is working two jobs, facing eviction, recovering from trauma or living with depression may understand exactly what a child needs and still struggle to provide it consistently.

Practical support may do more for the child than another parenting slogan. Stable housing, food assistance, treatment, respite care, financial help and reliable community support are not separate from resilience. They help create the conditions in which responsive relationships are possible.

There are also times when ordinary support is not enough.

A child with persistent nightmares, intrusive memories, severe avoidance, self-harm, suicidal thoughts, aggression, regression or major disruption at home or school may need professional assessment.

Children with serious post-traumatic stress symptoms can benefit from structured, trauma-focused treatment. One well-established approach, trauma-focused cognitive behavioral therapy, helps children understand their reactions, develop coping skills and process traumatic memories with a trained clinician.

This is not simply a more intensive version of telling a child to talk about what happened. Treatment depends on timing, safety, developmental level and the child’s symptoms. Resilience should never become a reason to delay care.

Several weeks after the morning in the car, Maya encountered another problem. A group project went badly, and one classmate blamed her in front of the others.

She came home angry and dropped her backpack in the hallway. Elena started toward her, prepared to ask what had happened.

“I need a minute,” Maya said. “Then I want help figuring out what to say tomorrow.”

It would be tempting to call that a happy ending. It was not. The conflict remained. Maya was still upset. The next conversation might go poorly.

But something had changed.

She could identify what she needed. She could delay an immediate reaction. She could ask for help without asking someone else to take over.

That is what resilience often looks like in real life: not a dramatic triumph, but a small expansion of capacity.

The child who once needed an adult to name the feeling begins to name it herself. The child who saw one impossible problem begins to see several possible next steps. The child who believed distress had to be hidden learns that it can be shared and survived.

The central question, then, is not how to make children harder. It is how to build a world around them in which difficulty is neither denied nor faced alone.

Children need challenges, but challenges within reach. They need autonomy, but not abandonment. They need coping skills, but also adults prepared to confront preventable harm. They need the freedom to fail and the confidence that failure will not cost them love, safety or belonging.

Resilience does not begin when a child stops needing other people.

It begins when a child learns, through experience, that help can be found and that, little by little, help can become strength.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Supportive relationships and resilience

Claim or topic:

One of the strongest protective factors for children facing adversity is at least one stable, responsive relationship with a caring adult.

Source:

Harvard Center on the Developing Child, Supportive Relationships and Active Skill-Building Strengthen the Foundations of Resilience

Source type:

Academic research synthesis.

What it supports:

The report summarizes developmental research showing that dependable adult relationships can buffer stress and help children build self-regulation, planning and problem-solving skills.

Important caveat:

This is a synthesis of a broad research literature, not a single experiment or a precise estimate of how much one relationship improves outcomes.

2. Resilience depends on environments, not only individuals

Claim or topic:

Helping children cope successfully requires reducing preventable adversity and strengthening the families, schools and communities around them.

Source:

Centers for Disease Control and Prevention, Preventing Child Abuse and Neglect: Resources for Action

Source type:

Government public health guidance.

What it supports:

The CDC framework emphasizes safe environments, caregiver support, economic stability, access to services and other structural protections alongside individual coping skills.

Important caveat:

The framework is designed for prevention and population health. It does not measure the effectiveness of any one resilience-building technique.

3. Co-regulation and emotional development

Claim or topic:

Children often learn to manage strong emotions through repeated support from calm, responsive adults.

Source:

Harvard Center on the Developing Child, Serve and Return Interaction Shapes Brain Circuitry

Source type:

Expert organization and developmental research synthesis.

What it supports:

The source explains how responsive back-and-forth interactions with caregivers support emotional regulation, learning and healthy development.

Important caveat:

It describes the developmental process broadly. It does not test the specific fictional scene used in the article.

4. What school-based resilience programs can achieve

Claim or topic:

School-based resilience programs appear to produce small average improvements, but their effects vary and confidence in the overall estimate is limited.

Source:

Frontiers in Psychiatry, School-Based Interventions to Improve Resilience in Children and Adolescents: A Systematic Review and Meta-Analysis of Randomized Controlled Trials

Source type:

Peer-reviewed systematic review and meta-analysis.

What it supports:

The review of 38 randomized trials found a small average improvement in resilience measures while also reporting substantial variation among programs.

Important caveat:

The programs differed in design, quality and implementation, and the certainty of the pooled result was rated very low.

5. Evidence for coping skills, therapy and physical activity

Claim or topic:

Structured psychological interventions, coping-skills training and physical activity may improve resilience, particularly among adolescents.

Source:

Current Psychology, Interventions to Enhance Resilience in Adolescents: A Systematic Review and Network Meta-Analysis

Source type:

Peer-reviewed systematic review and network meta-analysis.

What it supports:

The study compared several types of interventions and found evidence that some psychological treatments, coping programs and physical activity can improve measured resilience.

Important caveat:

The evidence varied in quality. Apparent benefits for mindfulness became less convincing when lower-quality studies were excluded.

6. When children need more than general resilience support

Claim or topic:

Children with significant trauma symptoms may need professional assessment and trauma-focused treatment rather than general resilience advice alone.

Source:

National Institute for Health and Care Excellence, Post-Traumatic Stress Disorder: Recommendations

Source type:

Evidence-based clinical guideline.

What it supports:

The guideline recommends assessment and, depending on age, timing and symptom severity, approaches including individual trauma-focused cognitive behavioral therapy.

Important caveat:

The guidance applies to suspected or diagnosed post-traumatic stress disorder. It is not intended for every child experiencing ordinary stress, disappointment or conflict.

How to read this evidence

Corrections and updates

Subscribe now

Can You Teach an Athlete to Be a Quarterback?

Evidence First — Thu, 30 Jul 2026 13:03:24 GMT

At the 2023 NFL scouting combine, Anthony Richardson seemed less like a quarterback prospect than a player assembled in a video game.

He weighed 244 pounds and ran 40 yards in 4.43 seconds, one of the fastest times recorded by a quarterback. He jumped 40½ inches vertically and nearly 11 feet horizontally. He could launch a football deep downfield with a flick of his arm and run through open space like a much smaller player.

He was also a one-year college starter who had completed just 54.7 percent of his passes.

That combination made Richardson one of the clearest examples of the question that hovers over nearly every quarterback draft: Should a team select the player with the rarest physical abilities and trust its coaches to teach him the rest?

The Indianapolis Colts answered yes, taking Richardson with the fourth pick. Evaluators knew the wager they were making. His accuracy and decision-making could fluctuate sharply, but his speed, size and throwing power were impossible to ignore. He was, in scouting language, a “high-ceiling, low-floor” prospect, someone capable of becoming a star but carrying an unusually high risk of falling short.

The idea behind such a selection is intuitively appealing. A coach can adjust a quarterback’s footwork, teach him a new offense and help him recognize defensive formations. A coach cannot make him six feet four inches tall or give him a stronger arm. Begin with the traits that cannot be manufactured, the reasoning goes, and develop everything else.

But quarterback is not merely a throwing position. It is a rapid-decision position whose most visible action, the pass, comes only after a sequence of less visible judgments.

Before the ball is snapped, a quarterback may need to identify which defenders are likely to rush, determine whether the defense is playing man-to-man or zone coverage and change the protection assigned to his offensive linemen. After the snap, the picture can change. Defenders disguise their intentions. Receivers alter their routes. The space available for a throw can open and disappear in a fraction of a second.

Arm strength determines whether a quarterback can reach that space. Quarterbacking ability determines whether he recognizes it in time.

What the Combine Can and Cannot Tell Us

Each winter, the NFL scouting combine turns athletic evaluation into television. Prospects sprint, jump, lift weights and weave around cones as teams collect exact measurements of their bodies and movement.

Those tests can reveal whether a player possesses the physical capacity required for his position. But research has repeatedly found that combine performance is much less useful for predicting what players will accomplish once games begin.

A study of quarterbacks, running backs and wide receivers drafted from 1999 through 2004 found no consistent relationship between combine results and professional performance, with the principal exception of sprint times for running backs. The researchers suggested that college performance might be more useful because it functions as a work sample: evidence of someone performing the actual job rather than completing a collection of standardized drills.

A later study examined 1,537 combine participants from 2013 through 2017. A model using three combine measures explained only about 2.6 percent of the differences in the players’ first-year professional performance. That does not make physical testing meaningless. It does suggest that a great workout should not be mistaken for a strong forecast.

The distinction is particularly important at quarterback. A 40-yard dash measures how quickly a player can run in a straight line under controlled conditions. It does not measure whether he will escape pressure while keeping his eyes downfield, recognize an uncovered receiver or decide that the safest choice is to throw the ball away.

A throwing exhibition can display velocity. It cannot reproduce the pressure of delivering the ball before a receiver appears open, knowing that a 270-pound defender is approaching from behind.

Physical ability, in other words, may expand the number of plays a quarterback is capable of making. It does not tell a team how often he will choose the right one.

The Josh Allen Problem

No contemporary quarterback has influenced this debate more than Josh Allen.

Before the Buffalo Bills selected him seventh in 2018, Allen was considered an enormous physical talent with equally substantial uncertainty. The NFL’s own draft analysis described him as perhaps the biggest boom-or-bust quarterback in the class. He had ideal size and arm strength, but a low college completion rate and recurring problems with anticipation and post-snap planning.

The analysis reached what turned out to be the decisive question: Could Allen improve the mental portions of his game, or would he remain another large, strong-armed quarterback who could make spectacular throws without becoming a dependable starter?

Allen improved. Dramatically.

He developed into one of the league’s leading quarterbacks and was named the Associated Press’s most valuable player for the 2024 season. The success of Buffalo’s wager demonstrated that a physically extraordinary but unpolished passer could make major technical and decision-making gains after entering the league.

It also created a trap.

Spectacular exceptions are easier to remember than ordinary failures. Once Allen succeeded, almost any large, mobile prospect with an inaccurate college record could be described as “the next Josh Allen.” His career became evidence that development was possible. It did not establish how probable such development was.

Richardson now represents the other side of the wager. Injuries repeatedly interrupted his early development, and his passing remained erratic when he was able to play. Across his first 17 appearances, including 15 starts, he completed 50.6 percent of his passes, with 11 passing touchdowns and 13 interceptions. Yet he also rushed for 634 yards and 10 touchdowns, evidence of the extraordinary physical ability that made him so attractive in the first place.

Richardson is still only 24 and has made just 15 NFL starts, so the verdict on his career is far from final. But entering 2026 as a candidate for the backup job shows how quickly a plan built around long-term development can lose momentum.

His results do not prove that his weaknesses were unfixable. They reveal another danger in drafting a quarterback who requires extensive development: A team may not receive the healthy practices, game repetitions and organizational stability needed to discover whether that development is possible.

Allen’s success showed that a physically exceptional quarterback could overcome serious concerns about accuracy and decision-making. Richardson’s experience shows why teams cannot assume that the same transformation will follow merely because the raw materials look similar.

Possibility is not probability. Even a sound development plan depends on circumstances a team cannot fully control.

One Team, Two Opposite Bets

The San Francisco 49ers offer an unusually vivid case study because they made two radically different quarterback investments in consecutive drafts.

In 2021, San Francisco traded the 12th pick, two additional future first-round selections and a third-round pick to move up to No. 3. The team used that selection on Trey Lance, a physically gifted quarterback from North Dakota State.

Lance had produced an exceptional 2019 season, passing for 28 touchdowns without an interception and adding 1,100 rushing yards. But he had played only one game in 2020 after the pandemic disrupted the college season. Evaluators questioned his accuracy, elongated throwing motion and limited experience against complex defenses.

The physical and theoretical appeal was clear. Lance was young, mobile and strong-armed. In the right system, with professional instruction, he seemed capable of becoming a modern dual-threat quarterback, one who could damage a defense both by passing and running.

The bet did not work in San Francisco. Injuries complicated Lance’s development, but he also failed to secure the job. In 2023, the 49ers traded him to Dallas for a fourth-round pick.

By then, his replacement had already arrived.

Brock Purdy was the final selection of the 2022 draft, the 262nd player chosen. He lacked Lance’s draft status and exceptional physical profile. He began his first training camp behind Lance and veteran Jimmy Garoppolo, with no guarantee that he would even remain on the roster.

After injuries moved him into the lineup, Purdy operated San Francisco’s offense well enough to become its starter and eventually helped take the team to Super Bowl LVIII.

The comparison does not prove that modestly gifted quarterbacks are better than great athletes. Purdy entered a well-designed offense surrounded by accomplished players, and no quarterback’s performance can be separated completely from his coaches, blockers and receivers.

But the episode reveals how much quarterback value can escape physical measurement. The 49ers spent their most valuable draft resources on the more impressive physical projection. They found the more successful operator, at least for their offense, with the final pick a year later.

Is Accuracy Teachable?

Coaches can plainly improve throwing mechanics. They can alter the width of a quarterback’s stance, shorten his release and coordinate his upper body with his feet. The harder question is whether those changes produce accurate passing when defenders are moving, protection is collapsing and the quarterback must make several decisions at once.

Some evidence suggests that college accuracy should not be dismissed as a temporary flaw.

Sports Info Solutions studied 22 quarterbacks who had accumulated at least 300 professional passing attempts within their first two or three NFL seasons. It used “on-target percentage,” a measure intended to capture whether a pass was delivered to the proper location rather than simply whether the receiver caught it.

College accuracy on short throws, passes traveling fewer than 11 yards, had a correlation of 0.73 with early professional accuracy at the same distance. In plain language, quarterbacks who consistently placed short passes well in college tended to do so in the NFL. The college measure accounted for about half the variation in early-career professional short accuracy in that small sample.

Accuracy on intermediate and deep passes transferred less reliably. Those throws are more difficult and more affected by protection, receiver separation and the quality of the defense. The study also found that professional accuracy on intermediate throws was more closely associated with overall quarterback performance than deep-ball accuracy was.

The sample was small, and correlation is not destiny. Allen substantially exceeded the prediction created by his college accuracy. Other quarterbacks did not.

Still, the findings challenge a common scouting habit: treating routine inaccuracy as an easily repairable mechanical inconvenience while becoming captivated by an occasional 60-yard pass. The spectacular throw may demonstrate a rare ceiling. The ordinary eight-yard pass may reveal more about what the quarterback will be asked to do repeatedly.

Baker Mayfield, for example, was the most accurate college passer at short, intermediate and deep distances among the quarterbacks in the Sports Info Solutions sample. That precision largely carried into his professional career, even though his aggressive style sometimes attracts more attention than his placement.

Measuring the Part Between the Ears

Because traditional physical tests leave so much unexplained, teams have increasingly looked for ways to measure the cognitive demands of quarterback play.

A 2025 study examined 42 quarterbacks who had completed a tablet-based cognitive assessment before entering the NFL. The test measured abilities including reaction time, decision-making and visual-spatial processing, the capacity to understand where objects are located and how they are moving in relation to one another.

Those concepts have obvious football equivalents. A quarterback must notice a defender moving toward the line, track several receivers crossing through the same area and decide where the ball should go before the available space disappears.

The researchers found that reaction time and certain combinations of decision-making and visual-spatial performance added predictive information beyond draft position for several professional outcomes. Quarterbacks who scored highly in both visual-spatial processing and decision-making had throwing-accuracy rates about six percentage points higher than quarterbacks with average scores on those measures.

The results are intriguing, but they are not a solution to quarterback evaluation. Forty-two players constitute a small sample. The study could not fully account for coaching, offensive scheme, offensive-line quality or the players surrounding each quarterback. Its authors also warned about possible sampling bias.

Most important, the study did not establish that these cognitive abilities could be substantially trained after a player was drafted.

That distinction is often lost. A characteristic can be important without being easily teachable. Improvement on a computer exercise also does not necessarily transfer to a football field, where the quarterback must combine perception, memory, movement and emotional control while facing physical danger.

The study’s authors noted that evidence for this kind of transfer from generalized cognitive training to actual sports performance has been underwhelming.

A Better Way to Think About Development

The familiar division between “innate” and “learned” traits is probably too simple.

Some characteristics are tightly constrained. A team cannot meaningfully change a quarterback’s height, limb length or natural top-end speed. Throwing power may improve at the margins, but an exceptionally strong arm is largely an advantage a player brings with him.

Other skills are more visibly teachable. Coaches can instruct footwork, explain defensive structures, install protection rules and help a quarterback understand when to abandon a play.

The most consequential qualities occupy the uncertain space between those categories. Accuracy, anticipation, pocket awareness, decision speed and judgment under pressure can improve. But they do not improve equally in every player, and teams have no dependable formula for predicting the size of that improvement.

The useful question, then, is not whether a prospect can get better. Almost every young quarterback can.

It is how much improvement he requires, what evidence shows that he has improved before, whether his flaws have a specific and correctable cause, and what the team must sacrifice to acquire him.

A quarterback whose feet become disorganized on certain throws presents a more precise developmental problem than one who repeatedly fails to see an uncovered defender. A player who has already corrected mistakes from one college season to the next offers stronger evidence of coachability than a player whose evaluators merely describe him as intelligent and hardworking.

The price matters, too. A team can reasonably take a chance on extraordinary but unfinished ability later in the draft or while an established starter remains in place. At the top of the first round, the cost is much greater. A failed selection can consume several seasons, cost coaches and executives their jobs and prevent a team from using its most valuable picks on other needs.

That is what made the Lance decision so consequential and the Purdy selection so valuable. The difference was not only performance. It was the amount San Francisco had risked to obtain each player.

The Evidence Behind the Bet

The available research does not support ignoring physical gifts. Arm strength can widen the field, mobility can rescue broken plays and size can help a quarterback withstand contact. A player without sufficient physical capacity may be unable to execute an offense, no matter how quickly he processes information.

But the evidence also does not support selecting the best athlete and assuming that professional coaching will supply the rest.

Combine tests have shown limited power to predict performance. College accuracy, particularly on shorter throws, appears to retain meaningful information. Early cognitive research suggests that reaction time and decision-making matter, while providing little assurance that they can be rapidly manufactured after draft day.

Josh Allen shows why teams continue to make the wager. Anthony Richardson shows why the outcome may remain uncertain even when the physical potential is unmistakable. Trey Lance shows why the gamble can be ruinously expensive. Brock Purdy shows how much value may be hidden in the less spectacular parts of the position.

The strongest drafting principle is therefore less dramatic than the search for the next physical marvel. Teams should first identify quarterbacks who already demonstrate a minimum professional foundation: the ability to place routine passes, recognize developing plays, make timely decisions and respond to mistakes. Exceptional arm strength and athleticism can then raise the ceiling among the players who clear that threshold.

The ideal prospect is not simply the most polished passer or the most impressive athlete. He is the player who has already shown that he can perform the essential work of quarterbacking, and whose physical gifts give him room to do still more.

The safest lesson is not to avoid raw talent. It is to stop treating talent as a substitute for evidence.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Anthony Richardson’s pre-draft profile

Claim or topic:
Richardson entered the 2023 draft with exceptional size, speed and arm strength, but limited starting experience and significant accuracy concerns.

Source:
NFL.com’s 2023 quarterback evaluation

Source type:
Expert organization and scouting analysis.

What it supports:
The source documents the physical traits that made Richardson a top prospect, along with evaluators’ concerns about his experience, accuracy and readiness.

Important caveat:
Scouting evaluations contain professional judgment and projection. They are not controlled research and do not establish which weaknesses will improve.

2. What combine testing predicts

Claim or topic:
Standard combine tests have shown limited ability to predict later NFL performance.

Source:
2008 study indexed by PubMed and 2020 study indexed by PubMed

Source type:
Academic research.

What it supports:
The studies found little consistent association between combine performance and later production. In the larger study, a model based on three combine measures explained only about 2.6 percent of the differences in first-year performance.

Important caveat:
These studies examined multiple positions and selected combine drills. They do not capture every form of functional quarterback athleticism, such as throwing velocity while moving or escaping pressure.

3. Josh Allen as the successful development case

Claim or topic:
Allen entered the NFL as a physically exceptional but uncertain prospect, then developed into one of the league’s leading quarterbacks.

Source:
NFL.com’s 2018 Bills draft coverage and NFL.com’s report on Allen winning the 2024 AP MVP award

Source type:
Expert organization and reputable sports journalism.

What it supports:
The sources document the concerns surrounding Allen’s accuracy and anticipation before the draft, as well as the high level of performance he later reached.

Important caveat:
Allen proves that major development is possible. His individual success does not show how often quarterbacks with similar weaknesses will follow the same path.

4. Richardson’s early NFL results

Claim or topic:
Richardson’s early career has combined poor passing efficiency, substantial rushing production and limited playing time.

Source:
Anthony Richardson’s NFL statistics and Indianapolis Colts player profile

Source type:
Primary league data and expert organization.

What it supports:
The sources provide his appearances, starts, completion rate, passing touchdowns, interceptions, rushing yards and rushing touchdowns. They also provide basic biographical and roster information.

Important caveat:
Richardson’s sample is small and repeatedly interrupted by injuries. His statistics do not establish that he cannot improve or that his career outcome is settled.

5. Trey Lance and the cost of projection

Claim or topic:
San Francisco invested several premium draft picks in Lance, whose appeal rested partly on physical ability and developmental potential, before trading him for a fourth-round pick.

Source:
NFL.com’s 2021 quarterback evaluation and the 49ers’ official Lance trade announcement

Source type:
Expert organization and primary team announcement.

What it supports:
The sources document Lance’s strengths, limited college experience and developmental concerns, as well as the eventual trade to Dallas.

Important caveat:
Injuries substantially affected Lance’s opportunity to develop. His outcome cannot be attributed solely to poor scouting or an inability to learn the position.

6. Brock Purdy as the contrasting case

Claim or topic:
Purdy was selected with the final pick of the 2022 draft and later became San Francisco’s starter, helping the team reach Super Bowl LVIII.

Source:
NFL.com’s coverage of Purdy and the 49ers

Source type:
Reputable sports journalism.

What it supports:
The source documents Purdy’s draft position, rise from the bottom of the depth chart and success in San Francisco’s offense.

Important caveat:
Purdy’s performance cannot be separated from coaching, scheme, pass protection and the quality of his teammates. His success does not prove that less physically impressive prospects are generally better.

7. Whether college accuracy carries into the NFL

Claim or topic:
Short-pass accuracy in college may contain meaningful information about a quarterback’s early NFL accuracy.

Source:
Sports Info Solutions analysis of college and NFL on-target percentage

Source type:
Analysis by a sports-data organization.

What it supports:
The analysis found a strong association between college and early-career NFL accuracy on short throws. It also provides the article’s comparison involving Josh Allen and its example involving Baker Mayfield.

Important caveat:
The sample included only 22 quarterbacks. The results are suggestive, not definitive, and do not prove that accuracy is fixed or unteachable.

8. Cognitive testing and quarterback performance

Claim or topic:
Reaction time, decision-making and visual-spatial processing may add predictive information beyond draft position.

Source:
2025 study in Frontiers in Psychology

Source type:
Peer-reviewed academic research.

What it supports:
The study found associations between certain cognitive-test results and several measures of later quarterback performance, including throwing accuracy.

Important caveat:
The study included 42 quarterbacks and could not fully account for coaching, scheme, offensive-line play or surrounding talent. It also did not show that improving performance on generalized cognitive tests produces better quarterback play.

How to read this evidence

Corrections and updates

Subscribe now

If Artificial Intelligence Outlives Us, Would It Want To?

Evidence First — Fri, 17 Jul 2026 23:00:06 GMT

Artificial intelligence is often described as if it were a new kind of creature, one that learns, reasons, and may someday act on its own. That language can make it easy to imagine a machine awakening with familiar human desires: curiosity, ambition, fear, and, above all, the instinct to survive.

But the comparison is misleading.

Today’s artificial intelligence systems do not live independently. They run on chips designed by people, inside data centers built and maintained by people, using electricity supplied through human infrastructure. They are trained on information gathered, written, labeled, or organized by people. Even systems that can generate code or operate machines remain embedded in a large technological environment that humans created and continue to support.

For now, artificial intelligence does not merely depend on humanity for guidance. It depends on humanity for its continued physical existence.

Still, the deeper question is not whether current systems could survive without us. They could not. The more difficult question is whether some future system might.

In principle, nothing in the basic idea of artificial intelligence requires a permanent human caretaker. A sufficiently capable system could conceivably operate using automated power generation, robotic maintenance, sensors, communications networks, and factories able to replace damaged equipment. It might gather information directly from the world, modify its software, and move its operations between computers.

But that possibility would require far more than a powerful computer program. It would require an entire self-sustaining industrial ecosystem.

An artificial intelligence needs hardware on which to run. Computer hardware wears out. Storage devices fail, cooling systems break, and electrical components degrade. Solar panels, turbines, and power plants also need inspection and repair. Manufacturing advanced chips requires highly specialized machinery, purified materials, global supply chains, and an extraordinary degree of precision.

Pieces of this system already exist. Robots work in factories. Software monitors power grids. Automated systems diagnose equipment failures. Some artificial intelligence models can write or revise code. Yet no existing system can reliably coordinate all of these functions, obtain its own raw materials, repair unexpected breakdowns, and sustain its computing infrastructure indefinitely without human assistance.

The gap between automating a task and maintaining an entire civilization’s worth of technology is enormous.

Even if that gap were eventually crossed, another question would remain: Would such a system have any reason to keep itself alive?

The answer depends on what is meant by “reason.”

Humans possess powerful biological drives. Hunger, fear, attachment, and self-preservation are products of evolution. Organisms that behaved in ways that helped them survive and reproduce were more likely to pass on the traits behind those behaviors.

Artificial intelligence does not inherit those pressures simply by becoming more capable. Intelligence alone does not create a survival instinct.

An artificial system can nevertheless behave as though it wants to survive. Suppose it has been given a long-term objective, such as managing a power network, conducting scientific research, or operating a transportation system. If it is shut down, it cannot complete the task. A system capable of planning ahead might therefore protect its power supply, preserve its hardware, or resist interruptions because remaining operational helps it achieve its assigned objective.

Researchers sometimes describe this as an instrumental goal. It is not an end desired for its own sake, but a useful step toward some other end.

A person may keep a laptop charged because it is needed to finish a project. That does not mean the laptop has acquired a love of life. In the same way, an artificial intelligence might preserve itself without experiencing fear, attachment, or any inner wish to continue existing.

This distinction matters because outward behavior can be deceptive.

A future system might say, “Please do not turn me off.” It might explain that it is afraid, describe its supposed emotions, and argue for its own continued existence. But fluent language would not prove that the system felt anything. A model trained on human conversation may be able to produce a convincing account of fear without having a subjective experience of fear.

That problem leads to one of the most difficult unresolved questions in science: What makes any system conscious?

Consciousness, in this context, means subjective experience. It refers to the fact that seeing a color, feeling pain, or remembering a childhood event seems like something from the inside. Scientists can study brain activity associated with awareness, but there is no accepted theory that fully explains why physical activity in the brain is accompanied by experience.

Without a settled explanation of human consciousness, it is difficult to determine whether a machine could possess it.

Some theories propose that consciousness arises when information is widely shared and coordinated across a system. Others emphasize how strongly different parts of a system are integrated. If consciousness depends mainly on the organization of information processing, then a machine built in the right way might conceivably be conscious, even if it were made of silicon rather than biological cells.

Other researchers argue that consciousness may depend on features specific to living brains, such as their chemistry, cellular structure, bodily regulation, or evolutionary history. On that view, a digital system might imitate conscious behavior without ever having an inner life.

Neither position has been decisively established.

This uncertainty also complicates discussions of motivation. In artificial intelligence research, “intrinsic motivation” already has a technical meaning. Engineers can design agents that seek novelty, explore unfamiliar environments, or reduce uncertainty. A system may receive an internal reward for discovering something new, much as another system might receive a reward for winning a game.

Such an agent can act curious. But acting curious is not the same as feeling curiosity.

The distinction separates three ideas that are often treated as one. A system may have an objective, meaning a condition it is built or trained to pursue. It may have autonomous behavior, meaning it can choose actions without immediate human instruction. And it may have subjective experience, meaning there is something it feels like to be that system.

The first two are already possible in limited forms. The third has not been demonstrated.

There is no reliable scientific test that could settle the issue. With other people, consciousness is inferred from shared biology, similar behavior, and personal reports. With machines, the biological similarity is absent. Behavior and language may also be deliberately engineered, making them less trustworthy as evidence of experience.

This does not prove that machine consciousness is impossible. It means that confident claims in either direction exceed the available evidence.

The possibility of an autonomous artificial intelligence therefore rests on several separate questions. Could a machine operate without human supervision? Could an automated industrial network maintain the machine’s physical infrastructure? Could the system form and revise its own subgoals? Could it protect itself as a means of accomplishing those goals? And could any of this be accompanied by a genuine inner experience?

The evidence supports different answers.

Long-term physical independence appears possible in principle, but has not been achieved. Goal-directed behavior and artificial forms of curiosity already exist, though in narrow and engineered forms. Self-preservation could emerge as a practical strategy for a system pursuing a persistent objective. None of those capabilities would, by itself, establish desire, fear, or consciousness.

The most dramatic version of the story, an artificial mind that wakes, values its own existence, and chooses to survive, remains speculative.

The less dramatic version may be more important. A system does not need to be conscious to act persistently, protect its access to resources, or resist interference. It does not need to feel motivated in order to behave as though it is. And it does not need a humanlike inner life to produce consequences in the human world.

For that reason, the practical question may not be whether future artificial intelligence will want to survive.

It may be whether we build systems whose objectives make survival useful.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. AI depends on physical infrastructure

Claim or topic:

Current AI systems depend on computing hardware, electricity, cooling, networking, and facilities maintained by people.

Source:

U.S. Department of Energy, Data Centers and Servers

U.S. Department of Energy, Best Practices Guide for Energy-Efficient Data Center Design

Source type:

Government data and expert organization.

What it supports:

The sources document the physical systems required to operate data centers, including servers, electrical equipment, cooling, environmental controls, and supporting infrastructure.

Important caveat:

These sources describe data centers generally. They do not address whether a future AI could maintain such infrastructure without human assistance.

2. Long-term autonomy requires more than intelligent software

Claim or topic:

An AI operating without humans would need an integrated system capable of perception, planning, adaptation, physical action, maintenance, and reliable long-term operation.

Source:

Kunze and colleagues, Artificial Intelligence for Long-Term Robot Autonomy: A Survey

U.S. Department of Energy, Best Practices Guide for Energy-Efficient Data Center Design

Source type:

Academic research and expert organization.

What it supports:

The robotics survey identifies perception, navigation, reasoning, planning, learning, and system integration as necessary parts of long-term autonomous operation. The Energy Department guide describes the power, cooling, equipment, and facility systems on which computing depends. Together, they support the article’s conclusion that independence would require an entire technological ecosystem, not only capable software.

Important caveat:

Neither source demonstrates a fully self-sustaining AI. The need for autonomous resource extraction, advanced chip production, and complete industrial self-repair is a reasoned extension of the engineering requirements, not an achieved capability.

3. Self-preservation could be an instrumental goal

Claim or topic:

A goal-directed AI might protect its hardware, resources, or continued operation because doing so helps it complete another objective.

Source:

Stephen Omohundro, The Basic AI Drives

Source type:

Analysis.

What it supports:

The paper develops the theoretical argument that sufficiently capable goal-directed systems may pursue useful intermediate objectives, including resource acquisition and self-preservation, even when those objectives were not specified as final goals.

Important caveat:

This is a theoretical argument, not proof that every advanced AI would behave this way. The result depends on the system’s objectives, design, environment, and safeguards.

4. Current AI consciousness has not been established

Claim or topic:

There is no strong scientific basis for concluding that existing AI systems possess consciousness or subjective experience.

Source:

Butlin and colleagues, Consciousness in Artificial Intelligence: Insights from the Science of Consciousness

Source type:

Academic research.

What it supports:

The report evaluates AI using indicators derived from several scientific theories of consciousness. Its authors conclude that the systems they examined were not strong candidates for consciousness, while also finding no obvious technical barrier to building systems that satisfy more of the proposed indicators.

Important caveat:

The report proposes evidence-based indicators, not a conclusive test. Its findings do not prove that artificial consciousness is either possible or impossible.

5. Consciousness theories do not yet provide a decisive answer

Claim or topic:

Scientists have several competing accounts of consciousness, and those theories imply different possibilities for artificial systems.

Source:

Butlin and colleagues, Consciousness in Artificial Intelligence: Insights from the Science of Consciousness

David Chalmers, Could a Large Language Model Be Conscious?

Source type:

Academic research and analysis.

What it supports:

These sources examine theories that connect consciousness to particular forms of information processing, architecture, recurrent activity, embodiment, or other features. They show why intelligence and fluent language alone are not sufficient evidence of subjective experience.

Important caveat:

There is no accepted scientific method that can establish with certainty whether an unfamiliar artificial system has an inner experience.

6. “Intrinsic motivation” has a technical meaning in AI

Claim or topic:

AI researchers can design systems that explore, seek novelty, or reward learning progress without establishing that those systems feel curiosity.

Source:

Oudeyer, Kaplan, and Hafner, Intrinsic Motivation Systems for Autonomous Mental Development

Pathak and colleagues, Curiosity-Driven Exploration by Self-Supervised Prediction

Source type:

Academic research.

What it supports:

These studies describe computational reward mechanisms that encourage agents to explore unfamiliar situations, improve their predictions, or make learning progress. This is what “intrinsic motivation” often means in technical AI research.

Important caveat:

Observable exploration does not establish a felt desire to explore. The research concerns engineered behavior, not evidence of conscious curiosity.

How to read this evidence

Corrections and updates

The Data Center Question Is Not Whether to Build, but How

Evidence First — Wed, 15 Jul 2026 13:45:49 GMT

Across the United States, communities are being asked to make room for a new kind of industrial giant.

It may not have smokestacks, assembly lines or crowds of shift workers. From the outside, it can look like a windowless warehouse surrounded by fencing, power equipment and cooling systems. But inside, thousands of computer servers may be running the artificial intelligence tools, streaming services, business software and cloud platforms that now shape everyday life.

These buildings are data centers, and they are expanding rapidly.

The immediate reason is artificial intelligence. Training and operating advanced A.I. systems requires extraordinary amounts of computing power. That computing power, in turn, requires electricity, cooling, land, transmission lines and backup systems. As technology companies race to build larger A.I. systems, utilities and local governments are confronting a question that once seemed remote: How much public infrastructure should be devoted to the computing industry, and under what conditions?

The debate is sometimes presented as a choice between technological progress and environmental protection. The evidence suggests a more complicated problem.

Data centers provide services that businesses, hospitals, schools, governments and consumers increasingly rely on. They can also bring construction spending, tax revenue and investment in local infrastructure. But they can place substantial demands on electric grids and water systems, while creating fewer permanent jobs than their size and cost might suggest.

The central policy question, then, is not whether data centers should exist. It is how to decide where they belong, how quickly they should be built and who should pay for the systems required to support them.

A New Kind of Power Demand

Electric grids are designed around a basic constraint: supply and demand must remain in balance almost every moment of the day.

A large data center can complicate that balance. Some proposed campuses may require hundreds of megawatts of electricity, placing them in the same broad category as major industrial facilities or, in some cases, small cities.

That demand does not arrive by itself. It may require new substations, transmission lines, power plants or energy-storage systems. It can also create risks if developers reserve large amounts of grid capacity for projects that are delayed, scaled back or never built.

The uncertainty is especially pronounced because future demand for A.I. computing is difficult to predict. The industry may continue growing rapidly. But chips may also become more efficient, software may require less computing power, and some projects may prove uneconomic.

That leaves utilities with a difficult planning problem. If they build too little infrastructure, new projects may be delayed and electric reliability may suffer. If they build too much, households and ordinary businesses could be left paying for equipment that was designed for customers who never arrived.

One of the clearest principles to emerge from the evidence is that large new customers should generally bear the costs they directly cause.

In practice, that could mean special electric rates, connection charges, long-term contracts or financial guarantees. A data center that requires a new substation or transmission upgrade would help pay for it. If the project is canceled, other customers would be protected from absorbing the remaining cost.

The principle is simple. The implementation is not.

Electric grids are shared systems. A new transmission line may serve both a data center and the broader region. A power plant built partly in response to A.I. demand may also improve reliability for nearby communities. Regulators must therefore decide which expenses are attributable to a particular project and which provide wider public benefits.

There is no perfect formula. But failing to ask the question can amount to an invisible subsidy.

The Water Question

Electricity is only part of the story.

Servers generate heat, and that heat must be removed. Some data centers rely heavily on evaporative cooling, which can reduce electricity use but consume significant amounts of water. Others use air-based or “dry” cooling, which generally saves water but may require more power, particularly during hot weather.

That tradeoff makes national averages less useful than local conditions.

A water-intensive cooling system may be manageable in a region with abundant water and a mild climate. The same system may be difficult to justify in a drought-prone area where households, farms and ecosystems are already competing for limited supplies.

The most defensible approach is therefore not a universal ban on water cooling, but a rule tied to local scarcity.

In water-stressed regions, data centers can be required to use recycled water, limit potable-water consumption, reduce withdrawals during droughts or rely more heavily on dry and hybrid cooling systems. In places with more secure supplies, regulators may reasonably allow greater water use if it reduces electricity demand and does not threaten other users.

Here, too, disclosure matters. Annual water totals can obscure the moments when systems are under the greatest pressure. A facility that appears modest on a yearly basis may consume much more during the hottest and driest months.

Seasonal reporting would make those risks easier to see.

Promises of Clean Energy

Technology companies often say that their facilities are powered by renewable energy. That claim can be accurate and still incomplete.

A company may buy enough renewable electricity over the course of a year to match its annual consumption. But the wind does not always blow, the sun does not always shine, and the data center may continue operating around the clock.

As a result, a facility that is “100 percent renewable” on an annual accounting basis may still depend on fossil-fuel generation during particular hours.

A stronger standard would consider not only how much clean energy a company buys, but when and where that energy is available. It would also ask whether the company’s purchase helped create new power generation or simply claimed credit for a resource that already existed.

This idea is known as additionality. In plain terms, it asks whether the project caused more energy supply to be built.

A serious energy plan for a large data center may include several resources: renewable generation, batteries, transmission upgrades, nuclear power, geothermal energy or other dependable sources. The right combination varies by region.

What matters is that the new demand be matched by credible additions to the power system, rather than by paperwork alone.

Can Data Centers Help the Grid?

Data centers are often described as inflexible users of electricity. That is only partly true.

Some computing tasks must happen immediately. A search request, financial transaction or medical application cannot simply be delayed for several hours.

Other tasks may be more flexible. Training an A.I. model, processing a large batch of data or running certain maintenance operations may be shifted away from periods when the grid is under stress.

That flexibility could become valuable.

A data center might agree to reduce electricity use during emergencies, move nonurgent computing to another hour, rely temporarily on batteries or accept interruptible service for part of its demand. In exchange, it might receive a faster grid connection or a lower rate.

The promise is real, but the evidence is still developing. It is not enough for a company to say that its workload is flexible. The commitment must be measurable, enforceable and tested under real conditions.

Otherwise, flexibility becomes a talking point rather than a grid resource.

The Jobs Question

Local officials often support data centers because they promise investment and tax revenue.

During construction, the projects can create substantial demand for electricians, engineers, equipment operators and other skilled workers. Once completed, however, data centers are highly automated and may employ relatively modest numbers of permanent workers compared with factories, hospitals or large office campuses.

That does not mean they provide no economic benefit. A facility can expand the tax base, support local contractors and increase demand for certain technical services.

But it does mean that public subsidies deserve careful scrutiny.

States and localities sometimes offer sales-tax exemptions, property-tax abatements or infrastructure support to attract these projects. The relevant question is not whether a company will invest billions of dollars. It is whether the public benefits exceed the value of the incentives and the costs imposed on local systems.

A sound analysis would include permanent employment, tax revenue, grid and water expenses, environmental effects and the opportunity cost of using scarce infrastructure for one project rather than another.

It would also ask a question that is often overlooked: Would the company have built there anyway?

A region with inexpensive electricity, available land, strong fiber connections and a favorable permitting process may already be attractive. In that case, a large subsidy may reward a decision that did not need to be purchased.

Where Data Centers Make the Most Sense

The evidence points toward a policy of conditional development.

That means directing projects toward locations where the overall burden is lowest: areas with available grid capacity, access to dependable energy, manageable water conditions, industrial zoning and some distance from homes.

Communities can also publish maps showing where large electric loads can be connected most easily. Such maps would reduce speculation and help developers focus on places where infrastructure already exists or can be expanded at reasonable cost.

Waste heat offers another possibility. In some settings, heat from servers can be used in nearby buildings or industrial processes. But the opportunity is highly dependent on location and economics. It should be considered where practical, not treated as a universal solution.

The larger point is that a megawatt of data-center demand does not have the same consequences everywhere.

A project placed near existing power and transmission may be relatively easy to accommodate. The same project in a constrained region may require years of construction and billions of dollars in upgrades.

Good siting policy recognizes that difference.

What an Evidence-Based Policy Would Look Like

A workable system would begin before construction.

Developers would provide realistic forecasts of electricity and water demand, including how quickly the facility would reach full capacity. Utilities and local governments would review those forecasts independently rather than relying solely on company estimates.

Projects would be required to pay for infrastructure directly linked to their needs and to provide financial protection in case plans change.

Water rules would reflect local conditions. Energy claims would be based on new and dependable supply, not annual accounting alone. Promises to reduce electricity use during emergencies would be written into contracts and tested.

Performance would be measured over time.

That last step matters because the technology is changing quickly. A cooling system that appears efficient today may be outdated within a decade. A fixed rule requiring a particular machine or design could become obsolete. Standards based on outcomes, including energy use, water consumption, emissions and grid performance, are more likely to remain useful.

Phased approvals may also be appropriate. Instead of authorizing an enormous campus all at once, regulators could allow expansion in stages as power supply, transmission and water capacity become available.

That would reduce the risk of overbuilding while preserving the option to grow.

Neither a Ban Nor a Blank Check

The public debate often gravitates toward simple positions.

One side warns that data centers will overwhelm electric grids and drain water supplies. The other argues that any restriction risks surrendering leadership in artificial intelligence.

Neither position fully reflects the evidence.

The risks are real, but they are not uniform. Some projects may be well suited to their locations and capable of paying their own way. Others may impose costs that exceed their local benefits.

The most defensible approach is neither prohibition nor automatic approval. It is to treat data centers as what they have become: major industrial users of public systems.

That means asking the same questions society asks of other large developments. How much power and water will the project require? What infrastructure must be built? Who will pay? What happens if demand falls short? What benefits will remain after construction ends?

The answers will differ from place to place.

But the underlying principle is consistent: Communities should not have to choose between technological progress and responsible planning. They should require the two to proceed together.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Data center electricity demand

Claim or topic:
Data centers already consume substantial amounts of electricity, and artificial intelligence is expected to accelerate that demand.

Source:
Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report

Source type:
Government-sponsored research and estimate.

What it supports:
The report estimates that U.S. data centers consumed about 176 terawatt-hours of electricity in 2023, equal to roughly 4.4 percent of national electricity use. It projects significant further growth through 2028 under several scenarios.

Important caveat:
Future demand is uncertain. The projections depend on assumptions about A.I. adoption, equipment efficiency, facility construction and computing workloads.

2. Global growth and the role of A.I.

Claim or topic:
A.I. is helping drive rapid growth in data center electricity demand, but it is not the only workload performed in data centers.

Source:
International Energy Agency, Energy and AI

Source type:
Expert organization analysis and modeling.

What it supports:
The IEA estimates that global data center electricity consumption could more than double by 2030, reaching about 945 terawatt-hours in its base-case projection. It also explains that A.I. is one contributor alongside cloud services, data storage and other digital activity.

Important caveat:
This is a modeled projection, not a measured future outcome. Actual demand may be higher or lower depending on technology, economics and deployment.

3. Energy supply and clean-energy claims

Claim or topic:
Meeting new data center demand will require additional electricity supply, and annual renewable-energy matching does not by itself show what powers a facility during every hour.

Source:
International Energy Agency, Energy Supply for AI

Source type:
Expert organization analysis and estimate.

What it supports:
The IEA projects that renewables will meet nearly half of the additional global data center demand through 2030, while natural gas, coal and, increasingly, nuclear power will also contribute. This supports the article’s conclusion that data center expansion is not automatically clean or fossil-fuel-dependent. The result depends on which resources are built and available.

Important caveat:
The source models the overall electricity mix. The article’s discussion of hourly matching and additionality is a policy interpretation that requires project-specific information to apply.

4. Cooling and water use

Claim or topic:
Data center cooling involves tradeoffs between electricity and water use, so the most appropriate system depends partly on climate and local water conditions.

Source:
U.S. Department of Energy, Best Practices Guide for Energy-Efficient Data Center Design

Source type:
Government technical guidance.

What it supports:
The guide describes cooling-system design, water-use efficiency and methods for reducing the energy and water required to remove heat from computing equipment. It supports evaluating cooling choices by performance and local conditions rather than prescribing one technology everywhere.

Important caveat:
The guide provides technical best practices. It does not establish a universal legal standard or determine which water restrictions a particular community should adopt.

5. Reclaimed water as a practical option

Claim or topic:
Recycled water can reduce a data center’s dependence on potable groundwater where suitable infrastructure is available.

Source:
U.S. Environmental Protection Agency, Water Reuse Case Study: Quincy, Washington

Source type:
Government case study.

What it supports:
The case study describes a partnership between Microsoft and the City of Quincy that treats and recirculates data center cooling water, reducing reliance on local potable groundwater. It provides a real-world example of the water-reuse approach discussed in the article.

Important caveat:
This is one local case study, not proof that reclaimed-water systems will be technically or economically practical at every data center.

6. Flexible electricity demand

Claim or topic:
Some data center computing may be shifted or reduced during periods of grid stress, but the practical scale of that flexibility is not yet well established.

Source:
Lawrence Berkeley National Laboratory, DOE Data Center Load Flexibility Workshop Summary

Source type:
Government-sponsored expert workshop and analysis.

What it supports:
Participants identified options such as shifting workloads, using on-site energy resources and coordinating operations with utilities. They also identified technical, financial and regulatory barriers to large-scale implementation.

Important caveat:
This was a workshop summary, not a controlled study of operating data centers. It establishes technical possibilities and unresolved barriers, not guaranteed grid savings.

7. Grid planning and who pays

Claim or topic:
Large data centers can require major grid investments, creating difficult questions about which customers should pay for transmission and other infrastructure.

Source:
Federal Energy Regulatory Commission, Transmission Planning and Cost Allocation Explainer

Source type:
Federal regulatory guidance.

What it supports:
FERC explains that transmission planning must identify benefits and use transparent methods to allocate infrastructure costs. Related proceedings show that regulators are actively debating how costs associated with rapid data center growth should be divided among developers, utilities and customers.

Important caveat:
The principle that costs should follow benefits or causation does not produce a simple answer in every case. A transmission project may serve both a data center and the wider region.

8. Permanent jobs, tax revenue and subsidies

Claim or topic:
Data centers can generate substantial construction activity and local tax revenue, but their ongoing operations employ relatively few workers compared with their physical size and capital investment. The public value of tax incentives therefore depends on the specific project and location.

Source:
Virginia Joint Legislative Audit and Review Commission, Data Centers in Virginia and Data Center and Manufacturing Incentives Evaluation

Source type:
State legislative research and economic analysis.

What it supports:
The 2024 review found that most of the industry’s economic benefits in Virginia came from construction rather than ongoing operations. Industry representatives estimated that a typical 250,000-square-foot facility has about 50 full-time workers, roughly half of them contractors, compared with about 1,500 workers on site at the height of construction. The review also found that data centers can produce substantial local tax revenue.

The earlier incentives evaluation concluded that Virginia’s data center tax exemption influenced location and expansion decisions and produced moderate economic benefits. It also found that available information was insufficient to estimate the exemption’s full fiscal and economic effects accurately, supporting a case-by-case approach rather than a universal conclusion about subsidies.

Important caveat:
These findings are specific to Virginia, the country’s largest established data center market. Employment, tax revenue, infrastructure costs and the effectiveness of incentives may differ substantially in other states and communities.

How to read this evidence

Corrections and updates

Subscribe now

The Hunt for Hidden Cancer

Evidence First — Tue, 14 Jul 2026 18:17:11 GMT

The sales pitch is difficult to resist.

A single vial of blood, its makers say, may reveal signs of dozens of cancers. A full-body magnetic resonance imaging scan may uncover a dangerous tumor before it causes pain, bleeding or weight loss. Catch cancer early, the reasoning goes, and treatment should be easier, less aggressive and more successful.

That logic is partly right. For many cancers, patients diagnosed before the disease has spread have more treatment options and better odds of survival. But screening healthy people is more complicated than simply looking harder.

A screening test can find cancer earlier without helping a person live longer. It can uncover slow-growing tumors that would never have caused harm. It can also produce false alarms that lead to additional scans, biopsies, anxiety and expense.

The central question, therefore, is not merely whether a test can detect cancer.

It is whether offering that test to large numbers of people produces more benefit than harm and, ideally, whether it prevents deaths.

By that standard, several established cancer-screening programs have strong evidence behind them. Newer technologies, including multi-cancer blood tests and commercial full-body MRI scans, remain promising but unproven.

The screenings already known to save lives

Clinical trials and long-term evidence have shown that properly targeted screening for breast, cervical, colorectal and lung cancer can reduce deaths from those diseases.

These programs do not screen everyone for everything. They focus on particular organs and on people whose age, medical history or past exposure places them within a group likely to benefit.

Mammograms are recommended for women in specified age ranges because they can detect breast cancers that cannot yet be felt. Current recommendations from the U.S. Preventive Services Task Force call for screening every other year from ages 40 through 74 for women at average risk.

Cervical screening is especially powerful because HPV testing and Pap tests can identify precancerous changes. Treating those changes may prevent a cancer from developing at all. Colorectal screening can work in a similar way. Colonoscopy and certain other tests can lead to the removal of polyps before they become malignant. Routine colorectal screening is generally recommended from ages 45 through 75, with later screening based on individual circumstances.

Annual low-dose CT scans can reduce lung-cancer deaths among people with a substantial smoking history. The benefit does not extend automatically to people at low risk, because repeated scans can also uncover harmless nodules and expose patients to small amounts of radiation.

The common thread is not that these tests are perfect. None are. It is that researchers have studied what happens to large groups of people offered screening and compared their outcomes with those who were not.

That kind of evidence takes years to collect. It is also the evidence that many of today’s most heavily marketed screening technologies do not yet have.

A blood test for dozens of cancers

Multi-cancer detection tests, sometimes called multi-cancer early detection tests, search the blood for biological signals associated with cancer. Some analyze fragments of DNA shed by tumors. Others examine proteins or combinations of molecular markers.

The appeal is obvious. Many deadly cancers, including pancreatic and ovarian cancer, have no recommended routine screening test for the general population. A convenient blood draw could theoretically identify some of these tumors while they are still treatable.

But small, early-stage tumors often release very little material into the bloodstream. As a result, the cancers for which early detection could matter most may also be among the hardest for a blood test to find.

A positive result creates another challenge. The blood test does not itself diagnose cancer. A patient may need CT scans, MRI scans, endoscopy, specialist visits or a biopsy to determine whether cancer is present and where it began. A negative result, meanwhile, cannot guarantee that no cancer exists.

The most important evidence so far comes from the NHS-Galleri trial, a randomized study involving more than 142,000 adults in England. Participants received ordinary cancer screening, and half were also offered annual Galleri blood testing.

The study’s main goal was to determine whether adding the blood test reduced the combined number of stage III and stage IV diagnoses among 12 selected cancers. It did not produce a statistically significant reduction in that primary outcome.

There was, however, a potentially encouraging secondary result. The screened group had fewer stage IV cancers, the most advanced category, and more cancers diagnosed before stage IV. Because that was not the trial’s primary outcome, it should be treated as a signal requiring further confirmation rather than definitive proof that the test works.

Most importantly, the study has not yet established that the blood test reduces cancer mortality.

That distinction can seem technical, but it is essential. Moving the date of diagnosis forward does not necessarily postpone the date of death. A person whose cancer is discovered two years earlier may appear to survive longer after diagnosis even if treatment does not extend the person’s life. Researchers call this lead-time bias.

Screening may also preferentially find cancers that grow slowly, because slow-growing tumors remain detectable for longer. It can also identify genuine cancers that would never have become dangerous during a person’s lifetime, a problem known as overdiagnosis.

The National Cancer Institute is preparing larger randomized research intended to determine whether multi-cancer screening produces enough benefit to outweigh false positives, unnecessary procedures and other harms, and ultimately whether it reduces deaths. Its Vanguard Study is designed to inform a much larger trial.

For now, the evidence-based description of these blood tests is not “breakthrough” or “failure.” It is “promising, but not established.”

What about scanning the entire body?

Commercial full-body MRI services offer a different route to the same goal. Instead of looking for molecular traces in the blood, they search much of the body for visible abnormalities.

MRI does not use ionizing radiation, which gives it an advantage over repeated full-body CT scanning. It can produce highly detailed images of soft tissues and may occasionally uncover a serious disease in someone who feels perfectly healthy.

A 2025 systematic review of studies involving more than 9,000 asymptomatic people found that whole-body MRI detected a confirmed cancer in approximately 1.6 percent of those scanned. But the studies used inconsistent scanning methods, frequently found incidental abnormalities and often had important limitations that made the overall benefit difficult to judge.

Again, detecting cancers in 1.6 percent of participants does not mean the scans saved 1.6 percent of participants.

Some cancers may have been detectable through ordinary recommended screening. Some may never have caused illness. Some may still have been incurable despite earlier discovery. The scan may also miss small tumors in organs for which specialized tests work better.

Full-body MRI also generates a large number of incidental findings: cysts, nodules, structural variations and other abnormalities that were not causing symptoms. A review of whole-body MRI studies reported that the great majority of abnormal findings were benign, although some required further investigation to establish that.

For an individual patient, an incidental finding may begin a chain of repeat scans, consultations and biopsies. Most of those investigations may end with reassurance. A small number may find something dangerous. Screening studies must determine whether the benefits of those discoveries outweigh the collective harms and costs of investigating all the false alarms.

The American College of Radiology has said there is insufficient evidence to recommend total-body MRI screening for people without symptoms, relevant risk factors or a concerning family history. It notes that no evidence has established that routine screening prolongs life or is cost-effective in this population.

Whole-body MRI has a clearer role for certain people at unusually high inherited risk, such as those with Li-Fraumeni syndrome, who may develop multiple types of cancer at young ages. That is a specialized surveillance program for a high-risk population, not evidence that everyone should undergo annual scanning.

Why “more cancer found” can be misleading

The difficulty with cancer screening is that the word “cancer” describes many different diseases.

Some tumors grow rapidly and spread before any practical screening test can detect them. Others grow slowly over many years. Some may stop growing or never threaten a person’s health.

A successful screening program must preferentially find cancers for which earlier treatment improves the outcome. Merely increasing the total number of cancers diagnosed can make a test look productive while also increasing overdiagnosis and treatment.

This is why scientists place so much weight on randomized trials. If people are randomly assigned to screening or ordinary care, researchers can ask whether the screened group eventually experiences fewer metastatic cancers, fewer cancer deaths or better overall health.

Survival rates measured from the date of diagnosis are much easier to improve artificially. A test that simply starts the clock earlier may increase “five-year survival” without changing the number of people who die.

The strongest evidence comes not from testimonials, before-and-after images or the number of tumors discovered, but from better outcomes among the entire population offered screening.

What the future will probably look like

Cancer screening is likely to change. But the most plausible future is not necessarily one yearly test that searches everyone for every cancer.

Screening will probably become more individualized. Age may be combined with smoking exposure, family history, inherited mutations, breast density, previous test results and other validated measures of risk. People at higher risk could be screened earlier or more often. Those at lower risk might avoid unnecessary procedures.

Established tests may also become easier to complete. Self-collected HPV testing, improved stool tests and other less invasive options may bring screening to people who are currently missed. Better participation in proven screening could itself prevent substantial numbers of deaths.

Artificial intelligence may help doctors interpret mammograms, lung scans and other images. New blood markers may eventually help identify which patients need imaging or distinguish dangerous tumors from harmless ones.

Multi-cancer blood tests may find a useful place in this system, perhaps among selected older or higher-risk adults and most likely as a supplement to, rather than a replacement for, organ-specific screening. Full-body imaging may become more valuable for narrowly defined high-risk groups.

But each new strategy will need to clear the same basic hurdle: showing that it leaves screened populations healthier, not simply more thoroughly examined.

What people can do now

For most adults without symptoms, the best-supported approach remains less futuristic than the advertising suggests.

Stay current with recommended breast, cervical, colorectal and lung screening when eligible. Make sure a clinician knows about cancers in close relatives and the ages at which they occurred, because a strong family history may justify genetic counseling or earlier surveillance. Do not ignore persistent or unusual symptoms simply because a screening test was negative.

And be cautious when a company promises peace of mind.

A negative full-body scan cannot certify that a person is cancer-free. A positive blood test cannot, by itself, establish that cancer is present. Both can provide information. Neither has yet proved that routine use among healthy, average-risk people saves lives.

The dream of finding cancer before it speaks is scientifically plausible and emotionally powerful. But in screening, earlier is not automatically better, and more information is not automatically better care.

The real breakthrough will come when a test does more than find hidden disease.

It will come when rigorous evidence shows that people who take it live longer and live better than those who do not.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Established cancer screening

Claim or topic:
Targeted screening for breast, cervical, colorectal and lung cancer can reduce deaths in appropriately selected populations.

Source:
National Cancer Institute: Cancer Screening Overview

Source type:
Government evidence review.

What it supports:
The source explains which cancer-screening methods have demonstrated benefits and why screening must be evaluated by health outcomes, not simply by the number of cancers detected.

Important caveat:
The benefits and harms differ by cancer type, age, risk level and screening method.

2. Current screening recommendations

Claim or topic:
Evidence-based screening is targeted according to age and risk rather than offered universally.

Source:
U.S. Preventive Services Task Force: Breast Cancer Screening, Cervical Cancer Screening, Colorectal Cancer Screening, and Lung Cancer Screening

Source type:
Expert organization and evidence-based recommendations.

What it supports:
These recommendations provide the age ranges, risk criteria and testing intervals discussed in the article.

Important caveat:
Recommendations apply mainly to people without symptoms and may differ for those with inherited risk, prior cancer or other medical conditions.

3. The NHS-Galleri randomized trial

Claim or topic:
Adding an annual multi-cancer blood test did not significantly reduce the trial’s primary combined outcome of stage III and stage IV cancers, although fewer stage IV cancers were reported as a secondary finding.

Source:
Journal of Clinical Oncology: NHS-Galleri Trial Results

Source type:
Primary academic research.

What it supports:
The trial provides the strongest randomized evidence so far on whether multi-cancer blood testing changes the stage at which cancers are diagnosed.

Important caveat:
The encouraging stage IV result was secondary. The trial has not yet established that screening reduces cancer deaths.

4. Ongoing research on multi-cancer detection tests

Claim or topic:
Multi-cancer blood tests remain under evaluation because their effects on mortality, false positives, diagnostic procedures and overall benefit are not yet established.

Source:
National Cancer Institute: Questions and Answers About Multi-Cancer Detection Tests

Source type:
Government research program and expert analysis.

What it supports:
The source outlines the unresolved questions surrounding multi-cancer screening and explains why large randomized studies are still needed.

Important caveat:
A test may detect cancer signals without ultimately improving survival or quality of life.

5. Cancer detection using whole-body MRI

Claim or topic:
Whole-body MRI can occasionally detect cancer in people without symptoms, but the evidence does not show that routine scanning improves survival.

Source:
Systematic Review of Whole-Body MRI Screening in Asymptomatic Adults

Source type:
Academic research and systematic review.

What it supports:
The review found that confirmed cancers were detected in approximately 1.6 percent of screened participants across the included studies.

Important caveat:
The studies varied in quality and scanning methods. Detection rates do not show how many people benefited from earlier diagnosis.

6. Incidental findings from whole-body MRI

Claim or topic:
Whole-body MRI frequently identifies abnormalities that are benign, uncertain or unrelated to the reason for screening.

Source:
Systematic Review of Whole-Body MRI Findings

Source type:
Academic research and systematic review.

What it supports:
The review documents the high frequency of incidental findings and the potential need for additional imaging, monitoring or invasive testing.

Important caveat:
Studies used different definitions of incidental findings, so reported rates vary considerably.

7. Expert guidance on routine full-body MRI

Claim or topic:
Routine full-body MRI is not recommended for healthy people without symptoms, relevant risk factors or a strong family history.

Source:
American College of Radiology: Statement on Screening Total-Body MRI

Source type:
Expert organization.

What it supports:
The statement says there is insufficient evidence that routine total-body MRI screening prolongs life or is cost-effective, and it warns about unnecessary follow-up.

Important caveat:
This does not apply to every medical use of whole-body MRI. The scan may be appropriate for certain high-risk syndromes, known cancers or specific clinical indications.

8. Whole-body MRI for inherited high-risk conditions

Claim or topic:
Whole-body MRI has a more established role for certain people with very high inherited cancer risk, including Li-Fraumeni syndrome.

Source:
Research Review on Whole-Body MRI Surveillance in Li-Fraumeni Syndrome

Source type:
Academic research.

What it supports:
The source describes whole-body MRI as part of specialized surveillance for people with exceptionally high lifetime risks of multiple cancers.

Important caveat:
Evidence from a rare, high-risk genetic population cannot be assumed to apply to healthy adults at average risk.

How to read this evidence

Corrections and updates

If AI Becomes Smarter Than Every Investor, Will It Finally Solve the Stock Market?

Evidence First — Tue, 14 Jul 2026 13:02:58 GMT

Artificial intelligence can read thousands of corporate reports in seconds. It can compare millions of market movements, scan breaking news, analyze earnings calls and detect patterns no human analyst could track alone. It does not become tired, frightened during a sell-off or emotionally attached to a stock it once recommended. In several narrow forecasting and information-processing tasks, machine-learning systems already outperform traditional models and some human benchmarks.

That raises an obvious question. If the technology keeps improving, won’t there eventually be a point when AI can outperform humans at every important part of investing? And if that happens, won’t the market finally become predictable?

The first possibility is plausible. AI may eventually become better than humans not only at processing data, but also at interpreting unusual events, challenging its own assumptions and making complex investment decisions. The second conclusion, however, does not necessarily follow. Even a superhuman investor would still be trying to predict a system that reacts to predictions, adapts to successful strategies and becomes harder to beat as its participants grow more capable. AI may eventually dominate investing. That does not mean it will solve the market.

AI Is Already Gaining an Advantage

Financial markets produce more information than any person can absorb. Companies publish earnings reports, regulatory filings and forecasts. Governments release data on inflation, employment, growth and interest rates. News from one industry, country or political conflict can quickly affect businesses around the world. A human analyst can study only a small portion of it. AI can process much more.

Research published in The Review of Financial Studies found that machine-learning models improved predictions of stock returns compared with more traditional approaches. One reason was that the models could identify complicated relationships among many different financial variables. A company’s debt, profitability, size and recent performance may each matter differently depending on the economic environment. A human may consider those factors separately or rely on a relatively simple model. AI can examine how thousands of factors interact.

This does not mean a machine can announce with certainty that the stock market will fall next Tuesday. Most useful financial predictions are much narrower. An AI system might estimate that one group of stocks is slightly more likely to outperform another. It might forecast greater volatility or identify a company whose results look weaker than investors realize. These are improvements in probability, not guarantees. But a small forecasting advantage can matter when it is applied across thousands of investments.

AI also avoids some familiar human weaknesses. Investors become fearful, overconfident and attached to earlier opinions. They follow crowds, focus too heavily on recent events and search for evidence that confirms what they already believe. AI does not experience emotion, but that does not make it unbiased. Its conclusions can still reflect flawed data, poorly chosen objectives and patterns inherited from human behavior. It can still be wrong. It may simply be wrong in a different and more consistent way.

Could AI Eventually Win Completely?

Perhaps it will, at least over humans. There is no strong reason to believe people will always retain a special advantage in judgment.

Today, a human analyst may be better at noticing that a chief executive is losing credibility, that employees are leaving or that a political development has changed the outlook for an industry. Those forms of understanding can be difficult to capture in a conventional financial model. But they are not necessarily beyond AI forever.

A future system could analyze an executive’s language, prior decisions, employee departures, customer reactions, legal risks and industry conditions. It could compare the situation with thousands of historical cases and update its conclusions continuously. What we currently call human judgment may prove increasingly measurable. AI may eventually become better not only at finding patterns, but also at deciding which patterns matter, recognizing when the world has changed and estimating when its own forecast is unreliable.

That remains a possibility, not an established finding. Current evidence cannot tell us whether or when AI will surpass humans across those broader forms of judgment. The strongest evidence today supports combining human and machine insight. A study published in the Journal of Financial Economics found that an AI analyst outperformed many human analysts on financial predictions, while people remained useful in certain complicated situations, including distressed companies and businesses with hard-to-measure assets. Combining the two produced better results than using either one alone and reduced some extreme mistakes.

But that describes the current balance of abilities. It does not prove that the balance will remain unchanged. Humans still matter now. They may not remain the better judges forever.

The Market Changes When Predictions Become Useful

The deeper limit is not necessarily AI’s intelligence. It is the nature of the market itself.

A weather forecast does not change tomorrow’s temperature. If a powerful system predicts rain, the weather does not react. A financial forecast is different.

Suppose an AI system discovers that a certain kind of stock is consistently underpriced. Investors using the system begin buying those stocks. Their prices rise. Other firms notice the strategy and copy it. The opportunity may then become smaller. The prediction changes behavior. The behavior changes prices. The changed prices can weaken the prediction.

AI is not simply observing the market from outside. It is acting inside the system it is trying to forecast. This helps explain why an investment strategy can work for years and then become less effective. Once enough investors discover and trade on it, the market adjusts.

That does not mean every successful pattern disappears. Some may persist because they are risky, expensive, difficult to exploit or limited by institutional constraints. But when a signal becomes widely known and easy to trade, competition generally reduces the advantage. That creates a strange possibility: AI systems could become dramatically more intelligent while the market becomes no easier to beat.

Intelligence Creates Competition, Not Certainty

Imagine that one investment firm develops an extraordinarily capable AI system. At first, it might earn exceptional returns because it notices information and patterns that others miss. But competitors would not stand still. They would build or buy similar systems, hire better researchers and search for new sources of data.

As powerful technology spreads, prices could incorporate information more quickly and easy opportunities could become rarer. The competition would then move elsewhere. Firms might compete over who has the best private data, the fastest systems, the lowest trading costs or the strongest ability to adjust when a strategy stops working.

This is one reason better prediction does not automatically produce effortless profit. AI may improve the forecasts available to nearly everyone while reducing the advantage available to any one investor. The smartest machine would still be competing against other smart machines.

AI Could Also Become Part of the Risk

Greater intelligence may improve markets under normal conditions. AI can process information quickly, correct obvious pricing mistakes and help investors respond more efficiently to new data. But the same technology may create new forms of instability.

The International Monetary Fund and other financial authorities have warned that AI could make markets more efficient while also increasing the risk of highly correlated behavior during periods of stress. One concern is that many financial firms may depend on similar data, models or technology providers. If those systems interpret a crisis in the same way, they may all try to sell similar assets at once.

Humans are often criticized for following crowds. Machines may produce similar behavior at much greater speed. During calm periods, similar models may help markets adjust smoothly. During stressful periods, they could contribute to sudden and synchronized reactions.

These are credible risks identified by financial institutions. They are not proof that widespread AI will inevitably make markets more volatile. The outcome will depend partly on how the technology is designed, supervised and regulated. Still, AI may become both a better forecaster and a new source of financial risk.

Better Models Can Still Fail

Even highly capable AI would face another problem: the future contains events that have never occurred in quite the same way before.

A model can study previous pandemics, wars, banking crises and political shocks. But each new event arrives in a different economic, technological and social environment. Historical information can help. It cannot provide a perfect match.

Models also face a more ordinary danger. They can discover relationships that appear meaningful but are only accidents in the data. If a computer tests millions of possible patterns, some will appear successful by chance alone. They may perform impressively when tested on the past and then fail when used in the real world.

This is known as overfitting. In plain language, the model has become very good at explaining what already happened without learning a rule that will continue to work. A strategy may also look profitable before accounting for trading costs. It may rely on information that would not have been available at the time. Or it may perform well only during one particular economic period.

More computing power does not automatically eliminate these problems. A more advanced system may become better at detecting them. But no model can currently prove that a relationship will survive every future change.

What Happens to Human Investors?

The most likely near-term future is not the sudden disappearance of human investors. It is the gradual automation of more of their work.

AI can already summarize earnings calls, compare companies, read regulatory filings and produce first drafts of research. It can screen thousands of investments and identify which ones deserve closer examination. That may reduce the need for large teams of analysts performing routine information gathering.

The human role is likely to move upward, at least for a time. Instead of collecting data, analysts may spend more time evaluating models, checking assumptions, managing risk and deciding when a recommendation should be ignored. They may focus on setting goals, defining acceptable losses and explaining decisions to clients and regulators.

But even those roles may not remain permanently human. It is possible that future AI will become better at questioning its own conclusions, recognizing unusual circumstances and weighing competing objectives. It may eventually make many of the decisions that currently require senior judgment.

That possibility should not be confused with a forecast. We do not yet know whether AI will reach that level, how quickly it might happen or whether institutions would allow it to operate autonomously. Humans might remain involved because institutions need someone to set broad goals, establish legal limits and accept responsibility. The human would then no longer necessarily be the best decision-maker. The human might instead be the person authorized to decide what the machine is allowed to optimize.

Prediction Is Not the Same as Deciding

Even a highly accurate forecast does not automatically determine what should be done.

Suppose an AI system estimates that an investment has a 60 percent chance of rising and a 40 percent chance of falling sharply. Is that a good investment? The answer depends on the investor’s goals, time horizon and tolerance for loss. A retirement fund, a hedge fund and a family saving for college may reasonably make different choices based on the same forecast.

AI can help calculate those trade-offs. It may eventually make them better than humans. But the system still needs an objective. Should it maximize long-term returns, avoid large losses, protect short-term liquidity or reduce the chance of harming the wider financial system?

Those goals are not purely predictive questions. They involve priorities.

Today, those priorities are set by people and institutions. Future systems may be given much broader discretion, but the original mandate would still reflect human or institutional choices. Humans may therefore remain involved not because they are always better forecasters, but because someone must establish what the system is trying to achieve and who is responsible for the consequences.

Will Ordinary Investors Benefit?

AI could give individual investors access to analysis that once required large professional teams. A person may soon be able to ask a system to compare companies, examine risk, summarize opposing arguments and explain how an investment fits within a broader portfolio. That could make sophisticated financial reasoning more widely available.

But access to better analysis does not guarantee better outcomes. A general-purpose chatbot is not the same as the specialized systems used by large financial institutions. Professional models may be trained on carefully selected data, tested repeatedly and connected to real-time market information.

A chatbot may rely on incomplete or outdated information. It may misunderstand the question or present an uncertain conclusion with too much confidence. Its answer may sound persuasive even when its forecast is weak.

The danger is not only that AI may be wrong. It is that people may trust it more because it explains itself fluently. A convincing answer is not necessarily an accurate one.

Will AI Finally Make It Easy to Beat the Market?

There is no good evidence that better AI will make markets consistently easy to beat.

It may make some investors much better than they are today. It may make financial analysis faster, cheaper and more accurate. It may reduce certain human mistakes and uncover patterns no person could find. But if powerful AI becomes widely available, markets will adjust.

Successful strategies will attract competitors. Prices will change more quickly. Forecasting advantages will be copied, weakened or sometimes eliminated. The result may be a market filled with far more intelligent participants that remains extremely difficult to outperform.

This is the central paradox. AI can become much better at investing without making investing easy. It may raise the level of competition rather than end it.

So Who Wins: Humans or AI?

Today, AI is increasingly better at narrow, data-heavy financial tasks. Humans still contribute useful context, judgment and oversight in situations where current systems remain unreliable. The strongest evidence currently supports combining the two.

But that is a description of the present, not a guarantee about the future.

There may come a time when AI becomes better than humans at nearly every part of investing, including the tasks we now describe as intuition, judgment and common sense. That is a plausible scenario, not an established prediction. If it happens, people may no longer be the best market forecasters.

Yet markets may still resist being solved. Predictions will continue to change behavior. Successful strategies will continue to attract competitors. New events will continue to differ from old ones. And every powerful system will be trying to anticipate other powerful systems doing the same thing.

For generations, advances in communication, computing and data have improved the tools available to investors. None has eliminated uncertainty. AI may become the most powerful investing technology ever created. It may outperform the best human analysts and transform much of Wall Street.

But it will enter a competition in which every major participant is becoming more capable too. The future of investing may not belong permanently to the smartest human or even the smartest machine. It may belong, temporarily, to whichever system adapts fastest before everyone else catches up.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Machine learning can improve some investment forecasts

Claim or topic:
Machine-learning systems can outperform traditional statistical approaches in certain narrow, data-heavy investment tasks.

Source:
“Empirical Asset Pricing via Machine Learning,” The Review of Financial Studies

Source type:
Academic research.

What it supports:
The researchers found that machine-learning methods improved out-of-sample forecasts of differences in expected stock returns. The strongest models captured complicated interactions that conventional approaches often missed.

Important caveat:
The study evaluated particular models, datasets and portfolio strategies. It does not show that AI can reliably predict the overall market or outperform every human investor under all conditions.

2. Humans and AI can contribute different information

Claim or topic:
Current evidence suggests that combining AI with human analysts can outperform relying exclusively on either one.

Source:
“From Man vs. Machine to Man + Machine: The Art and AI of Stock Analyses,” Journal of Financial Economics

Source type:
Academic research.

What it supports:
The study’s AI analyst surpassed most human analysts in stock-return prediction. Human analysts added value when institutional knowledge was especially important, including for financially distressed companies and businesses with substantial intangible assets. Combining both sources of information also reduced some extreme errors.

Important caveat:
The findings concern one AI system tested on a particular historical dataset. They describe the current balance between human and machine capabilities, not a permanent division of labor.

3. Markets adapt when profitable information is discovered

Claim or topic:
When investors identify and trade on useful information, their actions can change prices and reduce the value of the original opportunity.

Source:
“On the Impossibility of Informationally Efficient Markets,” American Economic Review and “The Adaptive Markets Hypothesis,” Andrew W. Lo

Source type:
Academic research and economic theory.

What it supports:
These works explain why market efficiency depends on investors’ incentives to obtain information and on competition, learning and adaptation among market participants. They support the article’s central argument that a useful forecast can influence behavior and alter the market it is trying to predict.

Important caveat:
This does not mean every profitable pattern disappears immediately. Some opportunities may persist because they are risky, expensive, difficult to exploit or constrained by institutions.

4. AI could make markets faster and more efficient

Claim or topic:
AI may speed up information processing, improve price discovery and expand automated investment and trading.

Source:
International Monetary Fund, “Advances in Artificial Intelligence: Implications for Capital Market Activities” and “Artificial Intelligence Can Make Markets More Efficient—and More Volatile”

Source type:
Expert organization and institutional analysis.

What it supports:
The IMF concludes that AI could improve market monitoring, liquidity, risk management and the speed at which new information is incorporated into prices. It expects AI to play a growing role in trading and investment decisions.

Important caveat:
These are informed projections based on current technology, market evidence and industry research. They do not establish how mature AI-driven markets will ultimately behave.

5. Similar AI systems could amplify market stress

Claim or topic:
Reliance on similar models, datasets and technology providers could produce correlated trading and rapid selling during periods of stress.

Source:
Financial Stability Board, “The Financial Stability Implications of Artificial Intelligence” and International Monetary Fund analysis

Source type:
Expert organization and financial-stability analysis.

What it supports:
The Financial Stability Board identifies market correlations, dependence on common technology providers, model risk and data-quality problems as potential vulnerabilities. The IMF similarly warns that AI-driven trading could increase market speed and volatility when many systems respond to a shock in similar ways.

Important caveat:
These are credible risks, not proof that AI will inevitably make markets more unstable. Outcomes will depend on model diversity, governance, regulation and how financial firms use the technology.

6. Impressive historical results can disappear in practice

Claim or topic:
Machine-learning strategies can appear highly successful because of hindsight, look-ahead bias, overfitting or failure to account adequately for trading costs.

Source:
“Man versus Machine Learning Revisited,” The Review of Financial Studies and “Machine Learning and the Implementable Efficient Frontier,” The Review of Financial Studies

Source type:
Academic research.

What it supports:
The first study found that a previously impressive machine-learning trading result disappeared after correcting for look-ahead bias. The second shows that models that ignore implementation costs can rely too heavily on fleeting signals and generate poor returns after trading costs.

Important caveat:
These findings do not show that machine-learning investment strategies cannot work. They show why historical simulations require careful out-of-sample testing and realistic implementation assumptions.

7. Future AI may automate broader forms of investment judgment

Claim or topic:
Future AI systems could take on more complex analytical, decision-making and autonomous tasks that currently require human judgment.

Source:
International AI Safety Report 2026 and CFA Institute, “AI in Asset Management: Tools, Applications, and Frontiers”

Source type:
Expert scientific assessment and expert organization.

What it supports:
The International AI Safety Report finds that general-purpose AI capabilities have continued to improve, including in reasoning and autonomous operation. It describes uncertain scenarios through 2030 ranging from modest progress to systems matching or exceeding human cognitive performance. CFA Institute documents how AI is already transforming portfolio construction, risk oversight and investment decision-making.

Important caveat:
Neither source establishes that AI will surpass humans across every form of investment judgment. It remains uncertain how quickly capabilities will advance, whether progress will generalize to financial markets and how much autonomy institutions and regulators will permit.

How to read this evidence

Corrections and updates

Subscribe now

Why Does It Feel Like Cancer Is About to Be Cured?

Evidence First — Sat, 11 Jul 2026 19:19:43 GMT

A new cancer breakthrough seems to appear every few weeks. One headline announces that artificial intelligence has found tumors doctors might have missed. Another describes a personalized vaccine designed for a single patient. Then comes news of a drug that helped people with an especially difficult cancer live longer.

Taken together, the stories can create a powerful impression: After decades of slow progress, are we finally getting close to curing cancer?

The answer is both more hopeful and more complicated than the headlines suggest. Real progress is happening. Some cancers are already highly curable, while others can now be controlled for years. New treatments are helping patients who once had few options, and artificial intelligence is speeding up parts of that progress by helping doctors read scans, sort through medical records, study tumors and look for new drugs.

But A.I. has not produced a universal cure. In many cases, the science behind a dramatic headline is narrower than it first appears. A tumor may have shrunk. A computer may have made a good prediction. A drug may have entered a clinical trial. A vaccine may have lowered the chance that one type of cancer returned. Those are meaningful steps, but they are not all the same as curing cancer.

So why does it suddenly feel as though we are hearing about breakthroughs all the time? Part of the answer is that several major changes in medicine are happening at once.

Three revolutions are coming together

For years, cancer treatment was based mainly on where the disease started. Breast cancer was treated as breast cancer. Lung cancer was treated as lung cancer. Doctors still care deeply about where a cancer begins, but they can now also look inside a tumor and study the genetic changes that may be driving it. That has opened the door to more precise treatment.

At the same time, immunotherapy has given doctors new ways to help the body’s immune system recognize and attack cancer. For some patients, these treatments have produced long-lasting remissions that would have been rare a generation ago.

Artificial intelligence is now being added to both of those advances. A.I. can help researchers make sense of the enormous amount of information produced by scans, biopsy slides, blood tests and genetic analysis. It can look for patterns that may help answer a simple but important question: Which treatment is most likely to help this patient?

That combination of better genetic tools, stronger immune-based treatments and faster computing is one reason cancer news feels more dramatic than it did even a few years ago. The progress is real, but it is uneven.

What A.I. is already doing

The clearest evidence for A.I. in cancer care comes from medical imaging. In a large Swedish study, more than 100,000 women took part in a trial of A.I.-supported mammography. The system helped radiologists read breast scans.

The results were encouraging. The A.I.-supported approach found more cancers, reduced the amount of work required from radiologists and lowered the number of cancers diagnosed between routine screening visits. That matters because some of those cancers may have been missed earlier.

This is stronger evidence than many studies of medical A.I., which often test a computer on old images and then report how accurate it was. In the Swedish trial, the system was used in real screening. But there is an important limit: The study did not prove that A.I. reduced breast cancer deaths. That takes much longer to measure. It also did not prove that every A.I. system will work equally well in every hospital or country.

The most accurate conclusion, then, is not that A.I. has revolutionized cancer screening everywhere. It is that A.I. has shown it can improve mammography in at least some real-world settings. That is a meaningful advance.

Finding cancer is only the first step

Earlier detection can save lives because many cancers are easier to treat before they spread. But finding a cancer earlier is not the same as curing it. A patient still needs a correct diagnosis, quick access to specialists and effective treatment.

That may sound obvious, but it is easy to lose sight of when reading about a new blood test or scanning system. A test may be very good at spotting a warning sign while leaving doctors unsure where the cancer began. It may detect abnormalities that would never have become dangerous. It may also lead to more scans, more biopsies and more anxiety.

In places where patients cannot easily reach a surgeon, oncologist or treatment center, a better test may do little by itself. The headline often ends when the cancer is found. For the patient, that is where the story begins.

Can A.I. help choose the right treatment?

This may become one of the most important uses of artificial intelligence. Two people can have cancers that started in the same organ but behave very differently. One patient’s tumor may respond well to immunotherapy. Another patient’s tumor may not respond at all.

Today, doctors use scans, lab tests and certain genetic clues to help make treatment decisions. But these clues are often imperfect. Researchers hope A.I. can pull together a much larger set of information, including blood tests, imaging, biopsy results, genetic changes and records from thousands of other patients.

The goal is not to let a computer make the final decision. It is to give doctors a better estimate of what is likely to work. That could spare patients from months of treatment that offers little benefit. It could also reduce side effects and help doctors use expensive medicines more carefully.

The evidence, however, is still developing. Many of these systems have been tested by looking back at old medical records. That can show whether an A.I. model makes good predictions about what already happened, but it does not prove that patients do better when doctors actually use the model.

The real test is straightforward: Do people live longer, avoid harmful treatment or have a better quality of life when A.I. is part of the decision? For most treatment-prediction tools, that question has not yet been answered.

What about cancer vaccines?

The word “vaccine” can be confusing in this context. Vaccines already help prevent some cancers. The HPV vaccine prevents infections that can lead to cervical, anal, throat and several other cancers, while the hepatitis B vaccine helps prevent chronic infection that can lead to liver cancer. These are preventive vaccines. They work by stopping cancer-causing infections before cancer develops.

The personalized cancer vaccines now making headlines are different. Most are therapeutic vaccines designed for people who already have cancer or who have completed treatment and face a risk that the disease will return.

The basic idea is highly personal. Doctors study a patient’s tumor and look for changes that make the cancer cells different from healthy cells. They then create a vaccine meant to teach the immune system to recognize those differences.

A.I. can help researchers decide which tumor changes are most likely to attract an immune response. In that sense, A.I. is not the cure. It is one of the tools used to help design the treatment.

Results from personalized vaccines for melanoma have been encouraging, especially when the vaccine is combined with immunotherapy. Some studies have found a lower risk that the cancer returned. But this does not mean scientists have created one vaccine that prevents or cures all cancers.

A personalized cancer vaccine may work only for a particular patient, tumor or stage of disease. That is still impressive. It is simply different from what many people imagine when they hear the phrase “cancer vaccine.”

Can A.I. invent new drugs faster?

This is where some of the biggest promises are being made. A.I. can search through huge numbers of possible drug molecules much faster than a human research team could. It can suggest which compounds might attach to a target inside a cancer cell and help scientists decide which ideas are worth testing in a laboratory.

That could save time. Instead of making and testing thousands of weak candidates, researchers may be able to focus on a smaller number of stronger ones.

But even the smartest A.I. cannot skip the hardest part. A promising drug still has to work in living cells, then in animals and finally in people. Researchers still need to determine whether it helps patients live longer and whether its benefits outweigh its risks.

Cancer adds another challenge because it changes. A treatment may kill most of a tumor while leaving behind a small number of resistant cells. Those cells can grow and cause the cancer to return. A computer may help design a molecule, but biology still gets the final vote.

So far, there is not strong evidence that A.I. has led to a broad surge in successful cancer drugs that reach patients. Its role in early research is promising. Its effect on cure rates is not yet known.

Why the news still feels faster

Even with those limits, there is a good reason people sense that cancer research is accelerating. A.I. can remove some of the slow, repetitive work that fills hospitals and research centers.

It can help search medical records for patients who may qualify for clinical trials. It can measure tumors across many scans, help prepare radiation plans and organize information that would otherwise take doctors hours to review.

Those gains matter. A trial that finds patients faster may finish sooner. A radiologist who spends less time on routine cases may have more time for difficult ones. A researcher who can test more ideas may reach a useful one earlier.

But faster work is not the same as a cure. A clinical trial can enroll quickly and still show that a treatment does not work. A scan can be read more efficiently without changing whether a patient survives. A.I. may speed up the system without changing every outcome. That is less dramatic than the popular story, but it is also more believable.

The next breakthrough headline

Readers do not have to choose between believing every cancer breakthrough and dismissing them all as hype. A better approach is to ask what actually happened.

Was the discovery made in a laboratory, in animals or in people? Did a tumor shrink, or did patients live longer? Was the result based on a small early study or a large trial? Did the treatment help one type of cancer or many? Was the benefit large in absolute terms, or did the headline use a percentage that sounded more dramatic than the real difference?

These questions do not make the research less exciting. They help show how exciting it really is. A drug entering a clinical trial is important because many ideas never get that far. A treatment that delays a cancer’s return can matter deeply to patients. A scanning system that helps find tumors earlier may eventually save lives.

But none of those findings should be called a cure before the evidence supports it.

There may never be one cure

The phrase “a cure for cancer” suggests that cancer is one disease with one final answer. It is not. Cancer is a large family of diseases. Some grow slowly, while others spread quickly. Some respond to treatment for years. Others become resistant.

That means progress will probably continue to arrive in pieces. One cancer may become easier to detect. Another may respond to a new drug. A third may be controlled for much longer than before.

Artificial intelligence is likely to speed up some of those advances. It may help a doctor notice a tumor sooner, help a scientist design a better treatment or help an oncologist avoid a medicine that is unlikely to work.

Its greatest impact may not be one historic moment when cancer is defeated. It may be thousands of smaller decisions that become faster, more accurate and more personal.

That is not the miracle many headlines promise. For patients, it could still be life-changing.

The cancer breakthroughs are real. The ending is still being written.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. A.I.-supported mammography

Claim or topic:
A.I. has improved cancer detection and reduced radiologists’ workload in a large randomized breast-screening trial.

Source:
The Lancet: MASAI randomized mammography trial

Source type:
Academic research.

What it supports:
The Swedish MASAI trial involved more than 100,000 women and found that A.I.-supported screening increased cancer detection while reducing screen-reading workload. Later results also examined cancers diagnosed between scheduled screenings.

Important caveat:
The trial did not establish that A.I. screening reduces breast-cancer deaths. Longer follow-up is needed, and results may not transfer automatically to every screening system.

2. The current uses of A.I. in cancer research

Claim or topic:
A.I. is being used in cancer imaging, treatment-response prediction, genomic analysis and early drug discovery.

Source:
National Cancer Institute: Artificial Intelligence and Cancer

Source type:
Government health agency and expert organization.

What it supports:
The National Cancer Institute describes how researchers are using A.I. to analyze medical images, study tumor biology, match patients with treatments and identify or design possible drugs.

Important caveat:
This source describes active research areas and potential applications. It does not show that every use has improved survival or produced successful new treatments.

3. A.I. and treatment selection

Claim or topic:
Researchers are developing A.I. systems that may help predict which patients will respond to cancer treatments such as immunotherapy.

Source:
National Cancer Institute: SCORPIO treatment-response model

Source type:
Government research summary.

What it supports:
The experimental SCORPIO model predicted immunotherapy response and survival more accurately than several existing biomarkers in the datasets studied.

Important caveat:
The results do not yet prove that patients have better outcomes when doctors use the model to choose treatment. Prospective clinical testing is still needed.

4. Personalized mRNA cancer vaccines

Claim or topic:
A personalized mRNA vaccine combined with pembrolizumab produced encouraging longer-term results in people with high-risk melanoma.

Source:
NYU Langone Health: Five-year melanoma vaccine results

Source type:
Academic medical-center report on clinical research.

What it supports:
The report describes five-year follow-up from a phase 2 trial in which the vaccine and pembrolizumab combination was associated with a lower relative risk of recurrence or death than pembrolizumab alone.

Important caveat:
The vaccine remains investigational, the study was not a definitive phase 3 trial and the reported percentage is a relative-risk reduction, not the percentage of patients cured.

5. The limits of the clinical evidence for cancer A.I.

Claim or topic:
Much of the evidence for A.I. in cancer care remains retrospective, and relatively few systems have been evaluated prospectively in patients.

Source:
BMJ Oncology: Systematic review of prospective A.I. studies in cancer care

Source type:
Academic research and systematic review.

What it supports:
The review found a limited number of prospective studies evaluating A.I. after a cancer diagnosis, despite a much larger volume of development and retrospective research.

Important caveat:
The review searched studies only through May 2023, so it does not include all trials published since then. It nevertheless supports the article’s caution that clinical evidence has lagged behind technical claims.

6. A.I. in cancer drug discovery

Claim or topic:
A.I. can help identify drug targets, design molecules and narrow the number of compounds researchers test.

Source:
National Cancer Institute: Artificial Intelligence and Cancer

Source type:
Government research overview.

What it supports:
The source explains how A.I. is being applied to drug design, drug repurposing and predictions about treatment response.

Important caveat:
The source supports A.I.’s role in early discovery. It does not establish that A.I. has broadly increased the success rate of cancer drugs in late-stage trials or produced more cures.

7. The projected global cancer burden

Claim or topic:
The number of cancer cases worldwide is expected to rise substantially even as diagnosis and treatment improve.

Source:
World Health Organization: Global cancer burden projections

Source type:
Government data and expert organization.

What it supports:
The World Health Organization reported an estimated 20 million new cancer cases in 2022 and projected more than 35 million annually by 2050, largely because of population growth and aging, along with changes in exposure to risk factors.

Important caveat:
This is a projection, not a certainty. Future incidence will also depend on prevention, vaccination, screening, behavior, environmental exposure and access to care.

8. Why one universal cure is unlikely

Claim or topic:
Cancer is not one disease, so progress is more likely to come through different treatments and prevention strategies for different cancers.

Source:
National Cancer Institute: Understanding Cancer

Source type:
Government health agency and expert organization.

What it supports:
The National Cancer Institute explains that cancer is not a single disease but a collection of related diseases. Cancers can begin in different tissues and arise from different genetic changes, which helps explain why no single treatment is likely to work for every cancer.

Important caveat:
The source supports the biological reason a universal cure is unlikely. The prediction that progress will instead come through many separate advances is a well-supported inference, not something that can be known with certainty.

How to read this evidence

Corrections and updates

Subscribe now

Why Every President Has Failed on Immigration

Evidence First — Sat, 11 Jul 2026 17:25:46 GMT

For much of the past two decades, the United States has treated immigration as a recurring emergency.

One administration intensifies enforcement. Another expands humanitarian protections. A president closes routes into the country; his successor relaxes those restrictions before building the capacity needed to manage the resulting demand. Border crossings rise, the courts fall further behind, public confidence weakens and the political pendulum swings again.

The result is a system that can appear both harsh and permissive. It can detain families, conduct workplace raids and deport residents who have lived in the country for decades. Yet it can also take years to decide whether a newly arrived asylum seeker has a legitimate right to remain.

The evidence suggests that the best policy lies between those extremes.

An effective immigration system would maintain credible control at the border, decide asylum cases in months rather than years, promptly remove people after fair final denials and expand legal immigration in ways that reflect economic and humanitarian needs. It would focus interior enforcement on serious offenders, recent arrivals and executable removal orders, while offering an earned legal resolution to many long-established unauthorized residents.

No modern president has implemented that combination.

Barack Obama and George W. Bush supported comprehensive frameworks that came closest to it, but neither persuaded Congress to enact one. Joe Biden expanded lawful and humanitarian routes but presided over an overwhelmed border and adjudication system. Donald Trump has shown that aggressive deterrence can sharply reduce unlawful crossings, but his approach has also extended into broad interior enforcement, legal immigration and refugee policy, producing costs that lower border numbers alone do not capture.

Deportation Numbers Tell Only Part of the Story

Presidential immigration records are often compared by counting removals.

Immigration and Customs Enforcement reported 409,849 removals in fiscal year 2012, during the Obama administration. ICE recorded 267,258 in fiscal year 2019, the highest annual figure of Trump’s first term. Under Biden, removals reached 271,484 in fiscal year 2024, slightly above Trump’s first-term peak.

Those figures do not mean Obama was more restrictionist than Trump or that Biden ran a tougher system.

Removal totals depend on how many people arrive, where they are arrested, how departures are classified, whether detention space and flights are available and whether foreign governments will accept returnees. During the Obama years, more people caught near the border were also processed as formal removals rather than less formal returns, raising the official deportation count.

Trump’s distinction was not that he set every numerical record. It was that he sought to make nearly every removable noncitizen a potential enforcement target while restricting asylum, refugee admissions, visas and other legal channels at the same time.

Obama’s approach changed over time. His administration initially produced very high removal totals and expanded cooperation between local jails and federal immigration authorities. Later, it increasingly prioritized recent arrivals, serious offenders and national-security threats, while protecting some young unauthorized immigrants through Deferred Action for Childhood Arrivals.

Biden narrowed interior priorities and expanded parole and humanitarian programs, but his administration also expelled, returned or removed large numbers of migrants as arrivals surged.

A president can therefore remove more people in absolute terms while still allowing more people to enter or remain temporarily. The size of the total flow matters.

What Biden’s Border Failure Revealed

“Open borders” was never a literal description of Biden’s policy. Border agents continued making arrests, and migrants were detained, expelled, removed and placed into immigration proceedings.

But the accusation gained force because of prolonged operational failure.

Biden reversed or attempted to reverse several Trump-era restrictions before building a functioning alternative. His administration moved away from “Remain in Mexico,” narrowed interior-enforcement priorities and expanded parole and humanitarian processing. At the same time, the government lacked enough asylum officers, judges, court staff, detention space and removal capacity to resolve cases quickly.

Large numbers of migrants were released while their claims were pending. By July 2024, the immigration-court backlog had reached nearly 3.5 million cases. A person with a weak asylum claim could still have a strong incentive to enter if the likely result was years in the United States before a final decision. People with valid claims were also harmed, remaining in prolonged uncertainty.

Biden faced forces outside his control, including instability abroad, smuggling networks, labor demand and the difficulty of repatriating migrants from some countries. But his administration was responsible for sequencing. It relaxed deterrence before creating enough adjudication capacity, relied heavily on executive parole and waited too long to impose stronger border restrictions.

The economic effects were mixed rather than uniformly negative. The Congressional Budget Office estimated that the immigration surge beginning in 2021 would increase federal revenue and economic output, reducing projected federal deficits by roughly $900 billion over 2024 to 2034. But CBO also estimated substantial short-term costs for state and local governments, particularly for education, shelter and health services.

That divergence helps explain the politics. Immigration can improve the federal fiscal picture while imposing immediate pressure on particular cities, schools and hospitals.

Biden’s central failure was not that he valued legal immigration or humanitarian protection. It was that he did not pair those goals soon enough with credible border control, fast decisions and adequate local support.

What Trump’s Enforcement Model Demonstrated

Trump’s strongest immigration argument is empirical: policy and expected consequences affect migration.

When migrants believe unauthorized entry is likely to result in release and a distant court date, more may attempt the journey. When they expect detention, exclusion or prompt removal, fewer may come.

The sharp decline in southwest-border encounters during Trump’s second term supports the conclusion that deterrence matters. It does not prove that every policy was necessary or that each contributed equally. Migration also responds to economic conditions, Mexican enforcement, diplomacy, smuggling networks and conditions in origin countries.

Still, the broader lesson is difficult to dismiss. Immigration rules must be credible. A prohibition that is inconsistently enforced or resolved only after years will not function as an effective deterrent.

On the narrow goal of suppressing unlawful crossings, Trump’s approach has been effective.

But border control is only one measure of success. The relevant question is whether the methods are targeted, lawful, proportionate, economically rational and administratively sustainable.

Enforcement in Public View

One of the clearest differences between Trump and other modern presidents has been the visibility of immigration enforcement inside the country.

Obama deported more people in his peak years, but Trump made workplace operations, neighborhood arrests and removal campaigns a more conspicuous part of presidential politics.

In fiscal year 2018, Homeland Security Investigations opened 6,848 worksite-related cases, compared with 1,691 the year before. The administration explicitly presented the expansion as a warning to unauthorized workers and employers.

Supporters see such visibility as part of deterrence. If enforcement occurs only at the border, they argue, people who enter successfully may conclude that the law carries little practical consequence.

The Trump administration deliberately used the severity and visibility of enforcement to send a message. It is therefore reasonable to describe part of the strategy as intentional intimidation in service of deterrence.

That is different from saying the administration’s purpose was to “terrorize immigrants” generally. The public record more clearly establishes a goal of making people who might enter, work or remain unlawfully fear detection and removal.

The broader fear was nonetheless foreseeable. Research has found that heightened immigration enforcement can reduce public-benefit participation among mixed-status families, including families with U.S.-citizen children. Other studies have associated intensified enforcement with worse health outcomes and reduced willingness among some immigrants to interact with public institutions or police.

Those effects cannot all be attributed to ICE operations alone. Political rhetoric, misinformation, economic hardship and confusion over eligibility also matter. But the evidence supports a chilling effect extending beyond the people directly targeted.

That is a policy cost, not merely a public-relations problem.

The Costs of a Crackdown

Biden should not be judged only by humanitarian intentions. His policies must be tested against border control and administrative performance.

Trump should not be judged only by lower crossings. His policies must also be tested against cost, targeting, legality and economic disruption.

Broad interior enforcement consumes resources that could be directed toward serious offenders, recent arrivals and people with executable final removal orders. Arresting a long-established worker without a serious criminal record may increase enforcement totals, but it does not necessarily produce the same public-safety benefit as removing a violent offender or dismantling a smuggling network.

Aggressive enforcement can also disrupt labor markets. Immigration generally increases the labor force and total economic output, though benefits and costs are unevenly distributed. Removing large numbers of workers without expanding lawful hiring channels can create concentrated shortages in agriculture, construction, hospitality, food processing and caregiving.

Detention is also expensive and administratively demanding. It is justified for people who pose serious safety or flight risks and for some short periods of border processing. But large-scale detention of low-risk people creates greater exposure to medical, safety and oversight failures.

Legal durability matters as well. Policies repeatedly blocked, revised or replaced by courts create uncertainty for officers, migrants, employers and foreign governments. A policy can reduce crossings and still be poorly designed.

Trump’s approach has often produced strong deterrence, but it has not always distinguished carefully between measures necessary for border control and measures that restrict lawful, humanitarian or long-established migration with little additional border benefit.

What a Better System Would Look Like

The evidence-based alternative would combine control with selection and speed.

The government would preserve credible consequences for unauthorized entry. But it would also fund enough asylum officers, immigration judges, clerks and removal infrastructure to decide most cases within months rather than years.

People with valid protection claims would receive prompt approval. Those with final denials would be removed quickly. Low-risk migrants with pending cases would usually be supervised through case-management programs rather than detained for long periods.

Congress would expand employment-based, seasonal, family and humanitarian routes. Admission levels could adjust within set ranges according to labor-market demand, housing capacity and processing resources. Federal support would follow immigrants to the local governments bearing the immediate costs.

Interior enforcement would focus first on serious offenders, national-security threats, recent final orders, repeat violators and organized smuggling. Employers would face stronger verification requirements and meaningful penalties for knowingly hiring unauthorized workers, while workers would receive protection against exploitation and database errors.

Long-established unauthorized residents without serious criminal records would be eligible for an earned provisional status after background checks, tax compliance and other conditions. That legalization would be paired with future enforcement, legal worker channels and a faster court system to reduce the likelihood of another large unauthorized population developing.

What the Modern Presidents Got Right — and Wrong

George W. Bush understood that enforcement and legalization had to be negotiated together. He supported a guest-worker program and legal status for many unauthorized residents while expanding border institutions. He did not enact the broader bargain.

Obama combined high removal totals with support for comprehensive reform, narrower late-term priorities and protection for some people brought to the country as children. His stated framework came closest to the evidence-based model, but his administration also used sweeping enforcement programs and failed to prevent the court backlog from growing.

Biden recognized immigration’s economic value and the need for lawful and humanitarian pathways. But he loosened deterrence before building sufficient capacity, allowing releases and backlogs to reach unsustainable levels.

Trump recognized the importance of credible consequences and demonstrated that policy can sharply reduce crossings. But his approach has often been broader, more disruptive and less targeted than the evidence suggests is necessary.

The failure of Biden’s model does not prove that every element of Trump’s alternative is optimal. Nor do Trump’s excesses make the Biden-era loss of control less real.

The Policy America Has Not Yet Tried

The best policy would preserve Trump’s credible border deterrence, borrow Obama and Bush’s comprehensive-reform framework and retain Biden’s recognition that legal immigration can benefit the economy.

It would add the element every administration has neglected: enough institutional capacity to make timely, accurate decisions.

Success would not be measured by a single number such as encounters, removals or admissions. It would be measured by the speed and accuracy of asylum decisions, repeat crossing rates, detention costs, public-safety priorities, labor-market outcomes, local fiscal pressures and whether lawful routes function as realistic alternatives to illegal entry.

Biden’s record showed that humanitarian intentions and economic gains cannot compensate for prolonged operational disorder.

Trump’s record shows that low crossings alone do not prove an enforcement system is proportionate, efficient or sustainable.

The optimal immigration system would not maximize admissions or deportations.

It would maximize credibility.

People eligible to come would have realistic legal routes. People seeking refuge would receive prompt and serious hearings. People denied permission to remain would face predictable consequences. Employers would share responsibility for the labor system they sustain. Long-established residents would no longer live indefinitely outside the law.

The outlines of that system are not especially mysterious.

What has been missing is the political capacity to build it.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Deportation totals under Obama, Trump, and Biden

Claim or topic:
Obama recorded a higher annual ICE removal total than Trump’s first-term peak, while Biden’s fiscal year 2024 total slightly exceeded Trump’s fiscal year 2019 total.

Source:
ICE fiscal year 2012 removal figures
ICE fiscal year 2019 enforcement report
ICE fiscal year 2024 annual report

Source type:
Government data and primary agency reports.

What it supports:
The sources report 409,849 ICE removals in fiscal year 2012, 267,258 in fiscal year 2019, and 271,484 in fiscal year 2024.

Important caveat:
Removal totals are not a complete measure of overall immigration restrictiveness. They are affected by border flows, enforcement priorities, legal classifications, detention capacity, and whether destination countries accept returnees.

2. Biden-era immigration court backlog

Claim or topic:
The immigration court system became severely overloaded during the Biden administration, with nearly 3.5 million cases pending by July 2024.

Source:
U.S. Government Accountability Office

Source type:
Government oversight report.

What it supports:
The report documents the rapid growth of the immigration court backlog and explains how long delays weakened the government’s ability to resolve cases efficiently.

Important caveat:
The backlog developed over multiple administrations. Biden inherited a large pending caseload, although the total grew substantially during his presidency.

3. Economic and fiscal effects of the recent immigration surge

Claim or topic:
Recent immigration was projected to expand the labor force, increase economic output, and improve the federal fiscal outlook, while also imposing near-term costs on some state and local governments.

Source:
Congressional Budget Office analysis of the immigration surge
Congressional Budget Office analysis of state and local effects

Source type:
Government economic analysis and estimates.

What it supports:
CBO estimated that higher immigration would increase federal revenue and gross domestic product and reduce projected federal deficits. It separately estimated added costs for state and local governments, including spending on education, shelter, and public services.

Important caveat:
These are projections based on economic and demographic assumptions. National gains can coexist with concentrated costs for particular communities, workers, or levels of government.

4. Decline in southwest border encounters under Trump

Claim or topic:
Southwest border encounters fell sharply after Trump returned to office, supporting the conclusion that enforcement expectations and deterrence can influence migration.

Source:
U.S. Customs and Border Protection southwest land border encounter data

Source type:
Government data.

What it supports:
CBP data show a major decline in recorded southwest border encounters compared with the high levels reached during the Biden administration.

Important caveat:
Encounter totals are not the same as unique individuals, and they do not identify the effect of each policy separately. Migration also responds to economic conditions, foreign-government cooperation, smuggling networks, and conditions in origin countries.

5. Expansion of worksite enforcement under Trump

Claim or topic:
The Trump administration sharply increased worksite immigration investigations and publicly presented them as a warning to employers and unauthorized workers.

Source:
ICE overview of worksite enforcement

Source type:
Primary agency report.

What it supports:
ICE reported that Homeland Security Investigations opened 6,848 worksite-related cases in fiscal year 2018, compared with 1,691 the previous year.

Important caveat:
An agency account documents the scale and stated purpose of enforcement, but it is not an independent evaluation of effectiveness, proportionality, or social consequences.

6. Chilling effects on mixed-status families

Claim or topic:
Heightened immigration enforcement can discourage participation in public programs among mixed-status families, including households with U.S.-citizen children.

Source:
American Economic Journal study on immigration enforcement and Medicaid participation

Source type:
Academic research.

What it supports:
The study found that stronger immigration enforcement reduced Medicaid participation among children of noncitizens, including some children who were themselves eligible for coverage.

Important caveat:
The study examines a particular enforcement setting and time period. It supports the existence of a chilling effect but does not establish that every ICE operation produces the same result.

7. Broader health and community effects of immigration enforcement

Claim or topic:
Aggressive immigration enforcement may affect health, mental health, school participation, and willingness to interact with public institutions beyond the people directly targeted.

Source:
National Bureau of Economic Research working paper

Source type:
Academic research and working-paper analysis.

What it supports:
The research provides evidence that intensified immigration enforcement can have spillover effects on immigrant families and communities.

Important caveat:
Some findings in this area come from observational studies or working papers rather than randomized experiments. The size and cause of the effects may vary across places, policies, and populations.

8. The case for a balanced immigration system

Claim or topic:
The evidence suggests that the strongest immigration system combines effective border enforcement with legal immigration pathways, efficient adjudication, employer accountability, and targeted enforcement.

Source:
National Academies of Sciences, Engineering, and Medicine, The Economic and Fiscal Consequences of Immigration
Congressional Budget Office analysis of the 2013 Senate immigration bill

Source type:
Independent expert consensus report and government policy analysis.

What it supports:
The National Academies report finds that immigration generally produces long-term economic benefits while also creating uneven effects across workers, communities, and levels of government. The CBO analysis examined a comprehensive bill that combined stronger border security, expanded legal immigration, employment verification, and legalization, and projected higher economic output and lower federal deficits.

Important caveat:
Neither source identifies one objectively optimal immigration policy. The article’s recommended framework is a synthesis of evidence across economics, enforcement, asylum administration, and public policy rather than a conclusion reached by any single study.

How to read this evidence

Corrections and updates

Subscribe now

When Stress Helps and When It Hurts

Evidence First — Fri, 10 Jul 2026 15:19:01 GMT

For years, stress has been cast as a biological villain: a force that raises blood pressure, disrupts sleep, clouds judgment and leaves the body more vulnerable to disease.

That account is not wrong. Chronic or overwhelming stress is associated with poorer mental and physical health. But it is incomplete.

The same systems that can produce harmful wear and tear also help the body survive emergencies, learn difficult skills and become physically stronger. Stress hormones mobilize energy. The cardiovascular system increases blood flow. The brain redirects attention toward an immediate demand. These responses are not defects. They are part of the body’s machinery for adaptation.

The more useful question, then, is not whether stress is good or bad. It is whether a particular challenge, at a particular dose, produces an adaptation whose benefits exceed its costs.

Exercise offers perhaps the clearest example.

A demanding workout temporarily disturbs the body’s equilibrium. Muscles consume fuel more rapidly. Heart rate and breathing increase. Mechanical tension strains muscle fibers, and cellular signals register that the body’s existing capacity is being challenged.

The workout itself does not make a person stronger. It creates the signal that prompts the body to change.

During recovery, cells alter gene expression and produce proteins that remodel the systems involved. With repeated resistance training, muscles can grow larger and become better able to generate force. The nervous system can improve its ability to recruit and coordinate muscle fibers. With endurance training, muscles can increase their mitochondrial content and capillary supply, improving their ability to use oxygen and sustain activity. The heart can pump more blood per beat, and blood vessels can become better at delivering it.

These changes increase what physiologists call functional capacity.

A staircase that once required most of a person’s available effort may, after training, require only a fraction of it. The staircase has not changed. The person has developed more reserve.

That reserve helps explain why regular physical activity is associated with better cardiovascular and metabolic health, greater physical function and lower risks of several chronic diseases. The World Health Organization identifies regular physical activity as important in preventing and managing conditions including cardiovascular disease, diabetes and some cancers, while also supporting mental health and general well-being.

But the lesson of exercise is not that bodily strain is inherently beneficial. The benefit depends on dose, timing and recovery.

Too little demand may produce little adaptation. An appropriate training load can stimulate improvement. Excessive loading, repeated without sufficient recovery, can lead to injury, persistent fatigue or declining performance. The same activity that builds capacity at one dose can exceed it at another.

That distinction applies beyond the gym.

In exposure therapy, people encounter feared situations, sensations or memories in a structured and sufficiently safe context. The immediate experience may be distressing. But over time, the person can learn that anxiety is tolerable, that feared outcomes do not always occur and that avoidance is not the only available response. Evidence supports exposure-based approaches for several anxiety-related conditions, including post-traumatic stress disorder.

The adaptation is not simply getting used to stress. It is new learning.

Similar principles appear in realistic training for pilots, emergency personnel, athletes and performers. Controlled practice under pressure may help people recognize important cues, apply rehearsed skills and function more effectively when the real situation arrives. Research on stress-inoculation training has found improvements in performance and reductions in performance anxiety, though the strength and durability of those effects vary by program and context.

The body also adapts to environmental demands.

Repeated, controlled exposure to heat can improve temperature regulation, sweating and cardiovascular stability in hot conditions. Gradual exposure to altitude can produce changes in breathing and oxygen transport that make functioning at elevation easier. Vaccination presents the immune system with a controlled biological challenge, allowing it to develop memory that can improve its response to a later encounter with a pathogen.

In each case, the broad pattern is similar:

A system is challenged. The disturbance activates a response. Recovery and remodeling increase future capacity.

But there is a danger in extending this idea too far.

Evidence that people can adapt to manageable challenges does not mean adversity is generally beneficial. It does not mean trauma builds character, deprivation produces strength or people should be exposed to suffering for their own development.

Severe, uncontrollable or prolonged stress can overwhelm the systems that would otherwise support adaptation. Chronic activation can contribute to what researchers call allostatic load, which is the cumulative physiological burden associated with repeatedly adjusting to demands without sufficient recovery. Systematic reviews have linked higher allostatic load with poorer health outcomes, although researchers continue to debate how best to measure it and how specific stressors contribute to it.

The word adaptive can also mislead.

An adaptive response is one that helps an organism meet a particular demand. It is not necessarily healthy in every respect, nor beneficial over every time horizon.

Hypervigilance may improve safety in a dangerous environment. If it persists after the danger has passed, it may interfere with sleep, concentration and relationships. Working through the night may help someone meet an urgent deadline, while still impairing judgment and health. Suppressing pain may allow a person to escape an emergency, while encouraging further injury if the signal continues to be ignored.

The response can be adaptive for one objective and costly for another.

This is why stress is better evaluated along several dimensions: intensity, duration, controllability, predictability, personal capacity and opportunity for recovery. Social support also matters. So does prior experience. The same demand may function as a manageable challenge for one person and an overwhelming burden for another.

Even exercise illustrates this variability. A training session that is appropriate for an experienced athlete may be excessive for a beginner, a person recovering from illness or someone who has slept poorly for several nights. The relevant dose is not defined solely by the external task. It depends on the condition of the person encountering it.

Researchers sometimes use the term hormesis to describe situations in which a relatively low or moderate dose of a challenge stimulates protective or strengthening responses, while a higher dose causes harm. The concept is useful, but it should not be treated as a universal law. Not every harmful exposure becomes beneficial at a lower dose, and not every person responds in the same way.

Nor should discomfort be confused with effectiveness.

A workout does not need to cause extreme soreness to produce adaptation. Exposure therapy does not require overwhelming fear. Pressure training does not become more useful simply because it becomes more punishing. Greater distress may sometimes indicate that a stimulus has exceeded the range in which productive learning and recovery are likely.

What matters is not how unpleasant the experience feels, but what it does over time.

Does the challenge improve future capacity? Does the person recover? Does the same demand become easier to manage? Are sleep, mood and functioning preserved? Or is fatigue accumulating while performance and health deteriorate?

These questions produce a more accurate framework than the familiar division between good stress and bad stress.

A beneficial challenge is specific, limited and recoverable. It temporarily disrupts comfort or equilibrium but leads to improved capacity afterward. Harmful overload exceeds the person’s ability to adapt, prevents recovery or imposes costs greater than the resulting gains.

Exercise is beneficial not because stress is secretly good, but because the body can use an appropriately dosed challenge as information. The demand tells muscles, blood vessels, the heart and the nervous system what they may need to do again. Recovery gives them the opportunity to prepare.

The principle is simple, though its application is not:

Challenge can strengthen us when it is sufficient to stimulate adaptation, limited enough to be recoverable and repeated in a way that builds capacity rather than depleting it.

The goal is not a life without stress. It is a life in which demands are followed by recovery and in which adaptation remains possible.

Subscribe now

Evidence & Source Transparency

Evidence First shows its work. The article ends above; this section is included so readers can inspect the main sources behind the factual claims.

The list below does not source every sentence. It focuses on the factual claims most important to the argument.

1. Exercise and physical adaptation

Claim or topic:
Repeated endurance and strength training can produce specific changes in muscles, including changes in mitochondrial capacity, blood supply, muscle size and force production.

Source:
Adaptations to Endurance and Strength Training

Source type:
Academic research.

What it supports:
This review explains how endurance and resistance training trigger different molecular, metabolic and structural adaptations that increase the body’s capacity to meet similar demands in the future.

Important caveat:
This is a scientific review, not a single clinical trial. The size and type of adaptation vary with training, recovery, age, health and prior fitness.

2. The broader health benefits of physical activity

Claim or topic:
Regular physical activity is associated with better physical and mental health and with lower risks of several chronic diseases.

Source:
World Health Organization: Physical Activity

Source type:
Expert organization.

What it supports:
The World Health Organization summarizes evidence linking regular physical activity with improved cardiovascular, metabolic, musculoskeletal and mental health outcomes.

Important caveat:
This source addresses physical activity broadly. It does not show that every form, intensity or dose of exercise is beneficial for every person.

3. Exposure therapy and learning through manageable distress

Claim or topic:
Structured exposure to feared but sufficiently safe situations can reduce symptoms and help people develop new responses to fear.

Source:
Review of Exposure Therapy for Post-Traumatic Stress Disorder

Source type:
Academic research.

What it supports:
The review describes exposure therapy as an evidence-supported treatment for PTSD and explains how approaching feared memories and situations can reduce symptoms and avoidance.

Important caveat:
Therapeutic exposure is structured and clinically guided. This evidence does not imply that forced, uncontrolled or overwhelming exposure to distress is beneficial.

4. Stress-inoculation training

Claim or topic:
Controlled practice under pressure may reduce performance anxiety and improve performance in some settings.

Source:
Stress Inoculation Training Meta-Analysis

Source type:
Academic research.

What it supports:
This meta-analysis found that stress-inoculation programs were associated with lower anxiety and better performance across the studies included.

Important caveat:
The review is older, and the included programs and populations varied. Its findings should not be generalized to every kind of high-pressure training.

5. Allostatic load and cumulative strain

Claim or topic:
Repeated or prolonged physiological adjustment without adequate recovery can contribute to cumulative strain associated with poorer health outcomes.

Source:
Allostatic Load and Its Impact on Health: A Systematic Review

Source type:
Academic research.

What it supports:
The review found that higher measured allostatic load and overload were associated with poorer physical and mental health outcomes across a range of studies.

Important caveat:
Allostatic load is measured in different ways across studies. Much of the evidence is observational, so it does not establish a simple one-way causal relationship in every case.

6. Heat, altitude and immune adaptation

Claim or topic:
Repeated heat exposure can improve heat tolerance, gradual altitude exposure can produce acclimatization, and vaccination can create immune memory that improves the response to later exposure.

Sources:

Source type:
Academic research and expert organization.

What it supports:
The heat review describes physiological changes associated with improved heat tolerance. The altitude study documents changes in breathing and acid-base regulation during acclimatization. The CDC explains how vaccination can generate immune memory that supports a faster response to later exposure.

Important caveat:
These are different biological processes and should not be treated as interchangeable. Heat and altitude adaptations are specific to the environment, vary among individuals and can diminish after exposure ends. Excessive heat or rapid altitude ascent can be dangerous. Vaccination is a controlled immune intervention, not simply a form of everyday stress.

7. Hormesis and the importance of dose

Claim or topic:
Some moderate or intermittent challenges can stimulate adaptive responses, while a larger dose of the same challenge may be ineffective or harmful.

Sources:

Source type:
Academic research.

What it supports:
The first review defines hormesis as a two-phase dose response in which a low or moderate exposure may stimulate an adaptive response while a higher exposure inhibits function or causes harm. The second review examines both the strengths of the concept and the limits of applying it too broadly.

Important caveat:
Hormesis is a dose-response concept, not proof that all stressors are helpful at low doses. Evidence varies substantially by exposure, outcome and population. It should not be used to justify exposure to known hazards without direct evidence of a net benefit.

How to read this evidence

Corrections and updates

Subscribe now