Jonathan Haidt: Instagram Is Not 'Generally Positive' for Teen Minds

Jonathan Haidt is a social psychologist at NYU, the author of *The Happiness Hypothesis*, *The Righteous Mind*, and *The Coddling of the American Mind*, and a critic of what social media has done to minds and to democratic life. Lex Fridman invites him on as a hard-hitting reply to Fridman's earlier conversation with Mark Zuckerberg. Haidt's own path to the subject did not start with Facebook. He got his PhD at the University of Pennsylvania in 1992, studying how morality varies across countries, with fieldwork in Brazil and India.

The 1990s were also when the American culture war began to blow up. Left and right, he says, were becoming like separate countries, and the tools of cultural psychology could be used on that split. By the early 2000s he was alarmed at how bad polarization had gotten. He first planned a book to help Democrats stop losing to George W. Bush, because his research suggested people on the right already knew what progressives thought, while he had grown up on the left with no idea what conservatives thought until he went looking, including in *National Review*.

Writing *The Righteous Mind*, he committed to reading the best conservative work rather than the worst, and found social-science ideas he had not expected. The goal shifted from helping one party win to helping left and right understand each other, and helping democratic institutions work. He joined Twitter around 2009 and thought little of it. Then, in 2014 and 2015, universities seemed to blow up. Greg Lukianoff came to him in May 2014 describing students who called a speaker they did not have to hear "dangerous" and "violence." Their book on distorted thinking and rising anxiety was not primarily about social media. After it appeared, things got worse everywhere, and Haidt concluded that something had changed in the fabric of the social universe.

Haidt now works two empirical files. One is teen mental health, which he says fell off a cliff in 2013. The other is what has happened to democratic and knowledge-generating institutions, from universities to journalism. He wants the evidence that social media is a contributor in both cases, not a vibe. Fridman asks whether the problem is the nature of the platforms, the choices of executives, the psychology described in *The Coddling of the American Mind*, or some mix. Haidt starts with the numbers on adolescents.

Self-reports of depression, anxiety, and loneliness, hospital admissions for self-harm, and suicide statistics all bounced around but stayed relatively level in the early 2000s. Then, depending on the measure, they shoot upward around 2010 to 2013. This is not a 10 or 20 percent rise. The increases are between 50 and 150 percent, depending on the group. Boys have lower baseline rates of anxiety and depression, so their curves look less dramatic, but both sexes move. Suicide among pre-teen girls is thankfully uncommon, Haidt says, and then he gives the line that matters: between 2010 and 2015 it doubled.

The same pattern is identical in Canada and the United Kingdom. Australia and New Zealand look similar after a short delay. It is less clear in the Germanic countries and in continental Europe, a difference he ties to how childhood is organized. Nobody, he says, has a competing explanation that fits the timing and the fact that it hit pre-teen girls hardest. The financial crisis would not have done that. He grants that correlation is not causation, and he uses the old television-and-crime example himself. What he will not grant is that this is only Gen Z getting more comfortable talking about feelings.

Haidt is not a Luddite, and he wants that on the table before the indictment. Connecting people is good. The telephone was an amazing step forward; free long distance was better; video is better still. Users posting on a site like LinkedIn can serve real needs. What he is talking about is not the internet, not technology, not smartphones, and not even all social media. It is a particular business model, pioneered by Facebook, in which people are incentivized to create content that pulls others in, and those people are the product sold to advertisers. That model, he says, looks incredibly harmful for teenagers, especially girls aged 10 to 14, and particularly harmful for democratic institutions because it produces anger, conflict, and the destruction of any shared narrative.

Fridman keeps putting the trade-off back in. Twitter makes him feel less lonely, gives him a little humor, and connects people across the globe, especially during Covid. Those positives are hard to measure, and honesty requires naming them. Haidt agrees there is a trade-off, then rejects the one-dimensional version of it, the idea that the only question is how much social media is too much. People ask what we would have done in lockdown without the platforms. His answer is that texting, the telephone, Zoom, Skype, multiplayer games, WhatsApp, blogs, and the rest of the internet would have been enough. Hyper-viral Facebook and Twitter helped some science travel faster. They also produced explosions of misinformation and a polarization so deep that, in his telling, a third of the country did not believe what the medical establishment was saying — an establishment that sometimes played political games and made itself less credible.

YouTube, he says, is a more complicated case, and he does not know enough to hold forth. Life without it is hard to imagine; it is incredibly useful; it also helps radicalize terrorist groups and murderers. He thinks about it the way he thinks about the internet. What he does know is that teen social life changed between about 2010 and 2012. Before 2010 most kids were not on every day, because they did not yet have smartphones. By 2012 to 2014 almost all had phones and were daily users. Girls went to Instagram and Tumblr, the visual platforms. Boys went to YouTube and video games. Those latter destinations do not seem as harmful to mental health, and may not be harmful at all. It is Tumblr and Instagram, he says, that really did in girls' mental health.

At 64 minutes and 31 seconds of Fridman's Zuckerberg interview, the CEO answers Frances Haugen. Haugen, Zuckerberg says, mischaracterized the internal research: on most measures kids were doing better on Instagram, and only one of 18 looked bad. Then comes the sentence Haidt wants to try. "I think an accurate characterization would have been" that teens using Instagram "is generally positive for their mental health." Taken as a whole, Instagram is a plus. Haidt's question is whether that is true.

Two weeks earlier he had testified before a Senate subcommittee considering Nate Persily's Platform Accountability and Transparency Act, with Ben Sasse and Chris Coons involved. Researchers, he told them, are flying blind. A child spends hours on Instagram and later is depressed, or not, or months later dies by suicide, and academics cannot see the path. The platforms can. He wants them compelled to reveal de-identified data so researchers can see what kids actually do, how mood moves, and what precedes a suicide. Leaks exist, and he is careful about their selection bias. A non-leak, institutional channel would be better.

The method he trusts is a set of living Google Docs, including one with Jean Twenge on teen mental health and another with Chris Bail on social media and political dysfunction. Studies are sorted by whether they answer a question yes, no, or mixed, and critics are invited to add what was missed. That is not a flourish. Haidt's early research was that gut feelings drive reasoning. He and Twenge already said in books that the platforms are bad for kids, so they have confirmation bias. He calls himself a devotee of John Stuart Mill: the only cure for confirmation bias is other people who have a different confirmation bias. Everything is collected at jonathanhaidt.com/socialmedia.

The Senate document is titled, bluntly, that teen mental health is plummeting and social media is a major contributing cause. The crisis, he argues, is specific to mood disorders, not a rise in every psychological complaint. It shows up in behavioral data, not only in surveys. The increases are very large. It is gendered, and it has hit many countries. On that characterization, he says, nobody has disputed him. The National Survey on Drug Use and Health, plus Pew and Gallup, give decades of self-report. The graphs are relatively flat until around 2010 or 2012, then the share of U.S. teens with a major depression in the last year climbs.

When Jean Twenge's *iGen* came out in 2017, critics said Gen Z was simply more willing to talk, and that this was healthy, not an epidemic. The day before Haidt and Lukianoff's book was published, a psychiatrist wrote in the *New York Times* that readers should relax: smartphones were not ruining kids' brains, because the numbers were just self-report and diagnosis, with no underlying change. Hospitalization data for self-harm and suicide, Haidt says, show the same sudden elbow between 2009 and 2012, then a climb. Those are actual children admitted for cutting themselves. Covid made things worse, but the catastrophe was already in place, and school returning will not rewind it.

Correlational studies almost all show a link, typically around a coefficient of 0.1. That is already larger than the famous potato comparison. That paper was not about social media. It lumped in digital media use — Netflix, homework on a computer, all screen time — and found a tiny association, about 0.03, with self-reported depression and anxiety. The authors said so in the paper. The media reported it as social media being no more harmful than eating potatoes or wearing eyeglasses, a catchy line Haidt still hears everywhere. In a Google Doc of more than a hundred studies, the pattern is consistent: zoom in on social media and the correlation gets bigger, often two or three times bigger. Zoom in on girls and it gets bigger again.

Amy Orben, one of the authors of those potato studies, later reviewed the literature and put the associations between social media use and well-being in a range from about *r* = 0.10 to 0.15 for boys and girls together. For girls, Haidt says, many studies including hers and his land around 0.15 to 0.2. He and Twenge found about 0.2 or 0.22. That sounds small until it is placed in public health. Correlations there almost never rise above 0.2. Childhood lead exposure and adult IQ, a problem taken seriously for decades, correlates at 0.09. The world is messy and so are the measurements. A stable 0.15, multiplied across tens of millions of children and years of their lives, can explain an epidemic.

Almost all of the research, Haidt says, uses a dose-response model. People treat social media like sugar: a little looks harmless, a lot correlates with harm, so the argument becomes how many hours are too many. That is the wrong picture. Social media is a complete rewiring of childhood. Humans evolved with children playing in mixed-age groups, doing errands, learning adult skills, and becoming mature by play and work. In the 1990s, before the platforms, Americans began dismantling that world. It felt too dangerous to let kids outside; they might be kidnapped. Haidt is a fan of Lenore Skenazy's *Free-Range Kids* and of Let Grow, the nonprofit he started with Skenazy, Peter Gray, and Daniel Shuchman to change laws and norms so children can go out and play.

Gen Z, born in 1996, is the first generation to get on social media before puberty. Millennials mostly hit it in college. Gen Z can lie about age, and they begin to pile on around 2009 and 2010. Two years later they are depressed. It is not that they ate too much sugar. Ordinary social interaction declined because everything ran through the phone. Imagine a middle school where everyone has Instagram, the drama is constant, and puberty happens inside that feed. Now imagine a school where phones go in lockers, no one has an Instagram account, and all the parents are on board. Parents only allow the app because the child says everyone else has it. That is a social dilemma, a trap. His solution is to keep kids off until they are done with puberty.

A study by Amy Orben and Andrew Przybylski finds the damage greatest for girls between 11 and 13. There is no way, Haidt says, to make these business models safe for pre-teens or even for 13- and 14-year-olds. Children should not be allowed on platforms where they are the product until they are 16, and the age limit should actually be enforced. Fridman wonders whether companies could redesign virality and recommendation so the objective is growth and anti-fragility rather than engagement. Haidt does not think that can be done for kids. He is reluctant to tell adults what to do — he has libertarian friends, and he uses Twitter himself — but he is equally reluctant to let Facebook and Instagram tell his kids what to do without his knowledge or control. He kept his children off Instagram. It caused trouble, and they later thanked him, because they could see that the culture of posting was, as they put it, stupid.

The only thing he thinks is truly healthy for children is free play, already stripped away in the 1990s and then further displaced by screens. Video games are a kind of play; he has no beef with Roblox or multiplayer games as a way to be with other kids. Twelve hours of games still means no physical play, and kids need huge amounts of conflict with no adult to supervise or mediate. He now sees the correlational work, the experiments with random assignment, and the testimony of girls themselves pointing the same way. Ask girls whether Instagram is good or bad for their mental health overall, and you will not find a chorus saying Mark is right. In surveys of what they think is causing depression and anxiety, social media is the number one answer. The main harm for girls, he suspects, is not even bullying. Once a girl starts posting, it takes over her mind. She can spend five or six hours a day on the app and still not be present in class, because she is thinking about how the last post is landing. He is 58 and checks his own tweets constantly after a big article. Imagine, he says, being a 12-year-old girl going through puberty.

Fridman asks whether the mental-health story and political division are connected. Haidt says yes. Brains are built for approach and withdrawal: a behavioral activation system, roughly in the front left cortex, for opportunity, and right-front circuits for threat. After 2013 he watched students arrive with the threat system hyper-activated, everything dangerous, not enough to go around — defend mode rather than discover mode. Scale that to a diverse secular democracy. Alexis de Tocqueville's Americans saw a problem, got together, and solved it. If a nation spends more time in defend mode, every utterance is scanned for danger and for a way to attack. He cannot think of a better way to push people there than a stream of partisan Twitter, horror stories about the other side, and the claim that if they win this election we lose the country. Most users are not talking politics. The ones who are flood journalists, and mainstream media is hugely influenced by Twitter.

Human nature is tribal. Haidt tells a story of playing paintball at 29, assigned with strangers to shoot one another, and feeling as if a room in the heart had opened. Men, testosterone, team sports as play war: that is who we are, and we are also good at making peace, trade, and deals. Yascha Mounk's *The Great Experiment* is his pointer on how hard diversity is in a democracy and what it takes to get the benefits. Early Facebook and Twitter, around 2007 to 2009, were still mostly fun, full of humor, not a place where any sentence might take your head off. Almost every platform started that way. Then the architecture changed.

His *Atlantic* essay ran as "Why the Past 10 Years of American Life Have Been Uniquely Stupid." His own title was "After Babel." The driving metaphor is the short Genesis story: God does not say let half the people hate the other half. He says, let us confuse their language so they may not understand one another. The unique fact of the present, Haidt argues, is fragmentation. Facebook's rebuttal noted that polarization predates social media and was rising in the 1990s. He agrees, and wishes he had said so more clearly. Cable news did not make him afraid of his students. Around 2014 and 2015, liberal professors began writing that their liberal students frightened them. Students, surveys show, are less afraid of professors than of other students. A small number became adept at using social media to destroy anyone they could get credit for destroying, often over a word rather than an argument, in a witch-trial mindset. That is a phase shift in group dynamics, not just another polarization story.

Universities generate knowledge when people who do not share your biases cancel out your rationalizations. They get stupid when dissent on certain topics is forbidden. Haidt's exhibit is David Shor, who after George Floyd's murder tweeted a study by Princeton sociologist Omar Wasow: in the 1960s, violent protests tended to backfire and peaceful ones to work. A Twitter mob followed, and Shor was fired within a day or two. Fridman wants more blame on the administrators who do the firing. Haidt, as a social psychologist, looks at situations in which 95 percent of people behave in a surprising, shameful way and asks what made the situation so powerful. Leaders in their fifties and sixties, often progressives, lacked the vocabulary to stand up to young employees and folded. He thinks the *New York Times* may be poking its head back up after giving in, including over James Bennet and a Brett Stephens column that mentioned IQ. Twitter had become, in his phrase, the paper's editorial board. When the Times's own board then said free speech mattered, Twitter exploded, and the paper did not retract.

Fridman asks what Elon Musk buying Twitter, while tweeting about free speech, will do. Haidt will cut Musk a lot of slack if he is a major reason the world decarbonizes quickly and gets to Mars. His worry is the culture war in the middle of which this is happening — a war that is making institutions stupid and that could, he thinks, mean more constitutional crises and more violence. Influential citizens should not add fuel. Musk declaring himself a Republican, and Democrats the bad party, is his right as a citizen. As the head of major companies, Haidt wants something like a Geneva Convention for the culture war: firms have ethical duties to employees, supply chains, and the effects they cause, and they should not be weighing in on cultural issues.

Musk's tweet said he had voted Democrat because they were the kindness party, that they had become the party of division and hate, that he would vote Republican, and that people should watch a dirty-tricks campaign against him. Fridman reads it as a response to Twitter's lean, and as opening the door to bringing a former president back. Haidt's imagined futures are the point. In one, every Fortune 500 CEO announces a tribe, you pick a doctor by red or blue, and the culture war is on 24 hours a day, including in restaurants and product aisles. In the other, political conflict is confined to political spaces. He thinks Musk's tweet threw fuel on the fire and set a norm that businesses will get even more politicized. The problem, specifically, was picking a side as CEO, in an insulting way.

Fridman plays devil's advocate: authenticity, toilet-tweet energy, chaos that is actually who Musk is, a lightness missing from the CEO of Ford reading PR. Haidt's reply is that the world is structured into domains. A CEO has fiduciary duties of loyalty to the company. Using the company for political gain or social standing shreds professional ethics around leadership, even if it is not a lawsuit. Fridman says Twitter never gives context. Haidt hangs the duty on the office of CEO, not on having a large platform as such, and he agrees with Scott Galloway that Musk is an appealing role model for young men who could be a much better one. Both men like the engineering. Fridman likes the authenticity of long-form conversation. They land near the same practical line: do not pick a party from the CEO chair. The more interesting Musk move, for Haidt, is a short tweet about authenticating all humans.

Everyone is fighting about content moderation, Haidt says, while the Chinese use digital technology skillfully to make a more stable authoritarian state and Americans use it to make themselves weaker, more fragmented, and more insane. Frances Haugen's figures, as he recalls them, are grim about how little hate speech and violent intimidation get caught. More moderation will not make a big difference. The power is in the architecture. The number one reform in his *Atlantic* article is user authentication. Anyone over 16 or 18 should be able to open an account, read, and use a fake name. If you want to post with viral amplification on a company that enjoys Section 230's powerful privilege, the company has a know-your-customer duty, like a bank that will not take a bag of stolen cash from "John Smith." The bank need not announce who you are. It needs to know you are a real person.

Level zero is what we have now: nothing. Level one, perhaps what Musk meant, is proving you are not a bot, a cat-and-mouse game with CAPTCHAs. Above that are schemes — ID, biometrics, a blockchain identity, a phone — that authenticate without making you publicly traceable. Worldcoin is one name he collects; human-id.org, from an NYU Stern student who had just come to his office, keys verification to a phone number. Extra numbers cost money, so almost nobody will buy a thousand. That would not eliminate trolls who use their real names, but it would make trolling scarier. Some men, polite in person, become horrible with anonymity. One troll can ruin thousands of people's day. Free speech still has to reckon with the person who does not disagree so much as spill drinks all over the party.

The public square we have, he says, applauds shin-kicking. Change the incentives. He has been kicking around an idea with Reid Hoffman: rate users, not posts, on a one-to-five scale of aggression and hostility, something AI could help code from obscenity and exclamation. Default to seeing fours and below. Fives can still speak; they simply do not get into everyone else's feed, and they do not see you. Being an asshole would start to cost reach. Fridman, as an AI researcher, doubts automatic nastiness classification and prefers crowdsourced labels that do not become cancellation mobs. They converge on averaging: a misclassified post would not matter if the running score is what counts, and two tracks — models and crowds, with staff looking at disagreements — could refine both. In physical life, being a full-time asshole is painful. It should be on the internet too.

Fridman worries that "misinformation," like "racism," is an important concept being used as a bullying word inside the companies, a way to label disliked speech as harmful. Haidt keeps a narrow, useful meaning: concerted campaigns, including by Russian agents, to plant and spread stories, the firehose of falsehood the Soviets practiced for decades. Companies need staff for that, because it will always be there and it is bad for the country. Renee DiResta's work, he says, frightened him about how easy manipulation and polarization are. The Russians have tried since the 1950s to make Americans feel as divided as possible, including by staging hate crimes. Most of today's argument is not that. It is each side calling what the other side says misinformation, as with the lab-leak hypothesis, which people were shut down for raising and which at least deserved to be talked about.

Those fights are almost entirely partisan. They will not get 60 Senate votes. Content moderation has to be done and can be improved, and he does not think it is nearly as important as people think. Almost all the action is in the dynamics, the architecture, the virality, and who is on the platform unverified. The *Turner Diaries* from 1978, Timothy McVeigh in 1995, the Buffalo killer's manifesto: the ideas were not new. The killer said he was influenced entirely by what he found online, not by anyone he met in person. Micro-communities of bizarre belief can form, and posts flow up inside them. The question is not whether we can stamp the words out. It is which platforms carry them into every little town, so that the small share of young men who are vulnerable get exposed.

Martin Gurri, a former CIA analyst and author of *The Revolt of the Public*, is the quote Haidt wants. In the age of mass media, Gurri says, we looked into a mirror that may have been distorted but that still gave us stories and facts in common. The digital revolution shattered that mirror. The public now inhabits the broken pieces of glass, highly fragmented, mutually hostile, yelling in bubbles. People on the right hear regulation and think censorship. Haidt's reply is that this is only the frame if the whole conversation is content moderation. His article is about architecture. When that changed, the social world went insane.

Fridman asks him to steel-man the view that social media is good for society, that Zuckerberg is right about teenage girls, and that the platforms are getting better. On mental health, the usual case is that isolated kids, especially LGBTQ kids, can find one another, that the apps expose you to ideas, and that they are fun. While people are on the platform, they report a mix. Many kids are having fun; some feel inferior, cut off, or bullied. If you measure hedonics — fun versus fear while using it — the average might be positive. That is what Zuckerberg was pointing at with 18 criteria, most positive. Haidt's analogy is heroin: you feel great while you consume it. You would not be glad it arrived in seventh grade if it meant nobody was on the playground. As an emergent change to childhood, the feelings during use are the wrong zoom.

For democracy he goes back to 2011 and the letter Zuckerberg wrote when Facebook went public, in the wake of the Arab Spring. Citizens, aided by Facebook and Twitter, pulled down dictators. From the early 1990s through that year — the fall of the Soviet Union, the fun of the early internet, Occupy, protests in Israel and Spain — there was a period of techno-democratic optimism. What dictator could stand up to people connected on these platforms? Then look at the years after. Zuckerberg told Fridman that research by and large finds social media is not a large driver of polarization, and that studies he has seen show social media use correlated with lower polarization. Haidt says that factual claim is not true of the literature as a whole. Filter bubbles are mixed. On the other six questions in the Bail document — anger and affective polarization, amplifying emotional or false posts, violence, foreign interference, trust, populism — the evidence leans yes.

Zuckerberg also notes that polarization has risen much faster in the United States than in other major countries, pointing to work by Matthew Gentzkow and colleagues that tracks many countries back to the 1970s. If the question is a straight line from the 1970s, Haidt says, that point has some force. The argument is about what happens after about 2012. The authors told him there is not enough post-2012 data for a reliable test. Eyeballing their graphs, some countries show a sharp upturn, not all. Democracy, by some measures, begins to drop in the late 2000s, by others in the early to mid-2010s. Zuckerberg can point to one study. The collaborative review, Haidt says, has many others pointing the other way.

He does not expect Facebook to raise the age on its own. COPPA set 13 in the late 1990s with no enforcement. Instagram's age is 13 and they do not enforce it, under pressure not to, because the kids would go to TikTok. There was even talk of Facebook for kids under 13, a business imperative to hook them early. That is a social dilemma: no player can opt out without losing, so you need central regulation. Not of who may say what, but of transparency — the Platform Accountability and Transparency Act from Senators Coons, Portman, and Klobuchar — and of oversight, such as Senator Michael Bennet's Digital Platforms Commission Act of 2022. The UK's Age-Appropriate Design Code, and a California version that would stop midnight notifications to children, could force global changes because platforms hate fragmented rules. Fridman notes clumsy regulation that blows up recommenders. Haidt prefers the threat of regulation and leaders who do the right thing, and he is more worried about a second civil war than about World War III.

Haidt teaches a course at NYU's business school called Work, Wisdom, and Happiness. It has become advice on getting stronger, smarter, and more sociable. Do those three and you will be more successful at work, love, and friendship, and as happy as you can be. The mind is a complex adaptive system that needs experience to wire itself. You do not grow while you are with your attachment figure. You grow when you go out, meet threat and fear, sometimes run back, and keep going until you can stay out there. That is normal childhood, he says, and it is what this country blocked in the 1990s. Older generations now block it out of fear, and phones block it by hijacking almost every input and almost every minute.

Put yourself in experiences. You are anti-fragile. You will not get strong without setbacks, criticism, and fights. The analogy for getting smarter is the same: expose yourself to people who disagree. That is why he co-founded Heterodox Academy, for political and viewpoint diversity among faculty, and why he thinks students need it too. Surveys, he says, show Gen Z has not turned against viewpoint diversity. Most want it. They are afraid of the small number who will shoot darts if they say something wrong. Becoming more sociable is simpler: look through the other person's point of view, what they want, what is interesting to them. He could assign *The Coddling of the American Mind* or Dale Carnegie's *How to Win Friends and Influence People*. The deeper instruction is to take charge of your development, or the protective generation and the phone will.

Fridman reads lines from Chief Justice John Roberts at a son's middle-school graduation: a hope that the children will be treated unfairly, so they learn the value of justice; that they will suffer betrayal, loneliness, bad luck, and an opponent who gloats, so they learn loyalty, friendship, chance, and sportsmanship. Whether one wishes those things or not, Roberts said, they are going to happen, and whether you benefit depends on seeing the message in misfortune. Haidt says that is what he was trying to say, only more beautiful. He leaves the rest where he started. A person cannot understand much from one perspective. He walked into this subject with a bias, so he put the critics in the file. The method travels farther than any one finding about Instagram.