Jonathan Haidt on Adjusting to Smartphones and Social Media

Jonathan Haidt returns to Conversations with Tyler, timed to The Anxious Generation, for a genuinely adversarial exchange with Tyler Cowen over whether smartphones and social media have caused a distinctive harm to adolescent mental health — or whether it is one more historical mood-shift people will simply adjust to.

Key ideas

1. The 2012 discontinuity is Haidt’s central empirical anchor. Time-use, loneliness, and mental-illness data all show a sharp turn around 2012 — coinciding with the spread of smartphones, the like/retweet button, and threaded comments — with the sharpest effect on secular, liberal, and Anglosphere/Nordic girls, and much smaller effects on conservative, religious, and ‘rooted’ kids.
2. The phone displaces play, not just attention. Haidt frames the smartphone as an ‘experience blocker’: the harm is not screen time as such but the opportunity cost — lost in-person friendship, sleep, outdoor time, and the unsupervised play that mammalian childhood development requires. An AI-mediated digest of the same content would not restore that loss; it would just re-engineer the phone-based childhood.
3. Teenagers are trapped in a collective-action problem, not exercising a preference. Citing a study by Leonardo Bursztyn, Haidt argues students would pay to be collectively freed from TikTok and Instagram even though none will unilaterally quit, since quitting alone means social exclusion — the platforms’ apparent ‘value’ is largely the cost of staying legible to peers who haven’t left either.
4. Cowen’s counter-case is that people adjust to major technological shifts, and AI accelerates the adjustment. He presses Haidt with agriculture, fire, urbanisation, and Aztec dietary workarounds for scurvy, and argues Gen Z’s genuine achievements (chess, Emergent Ventures grantees, small research teams at frontier AI labs) undercut the claim of a damaged generation. Haidt grants the possibility this is ‘early days’ but holds that current mental-health, self-harm, and suicide data describe an ongoing crisis, not a plausible future correction.
5. The remedy is four coordinated norms plus platform accountability, not government prohibition. No smartphones before high school, no social media accounts before 16, phone-free schools, and more independent play (via Let Grow, which Haidt co-founded with Lenore Skenazy and Peter Gray) — backed by raising and enforcing the COPPA data-consent age from 13 to 16, and removing platforms’ Section 230 shield specifically for addictive design choices, not third-party content.

Content

Parenting, politics, and identitarianism

Before turning to the book, Cowen opens with warm-up questions that set up Haidt’s later argument. Asked who makes better parents, Haidt says right-wingers, on the strength of a post-2012 divergence: children rooted in religious or conservative communities were far less likely to be ‘washed out to sea’ by phone-based childhood than secular, liberal peers. Pressed on why he isn’t himself a right-winger if parenting outcomes favour it, Haidt says values don’t simply follow from research findings — his own trajectory ran from a reliable young Democrat, through writing The Righteous Mind and developing sympathy for conservative and libertarian moral foundations, to a self-described centrist alarmed equally by post-2015 illiberalism on the left (the campus ‘cultural revolution’) and on the right (Charlottesville, Trump). On COVID, Haidt argues social conservatives’ resistance to masking and distancing was driven less by lower purity/sanctity concern than by a proxy fight over government control and distrust of a medical establishment coded as aligned with ‘the blue team’. On identitarianism, Haidt distinguishes having an identity — which he says he has no problem with — from a theoretical commitment to putting identity, especially racial identity, first analytically and treating identity groups as morally unequal; on self-labelling with mental-health terms (‘my anxiety’, ADHD, neurodivergence), he is empirically agnostic but suspects, drawing on Stoic and Buddhist ideas about interpretation, that strong identification with a diagnosis is probably counterproductive.

Why AI won’t solve the screen-time problem

Cowen’s first sustained challenge: won’t AI soon let teenagers get a ‘digest’ of their messages and social feeds, cutting screen time by two or three times? Haidt’s answer reframes the problem: the harm was never simply time-on-screen but the opportunity cost — the smartphone is an ‘experience blocker’ that displaces in-person talk, sleep, outdoor time, and above all unsupervised play, which all mammal young need to wire up their brains socially. AI compressing the same content into a shorter format doesn’t restore play; it’s still a virtual substitute for it. Haidt goes further, predicting that AI companions and ‘virtual girlfriends and boyfriends’ will be engineered by market pressure to be effortlessly gratifying — ‘you never have to adjust to them’ — which he calls a likely disaster for social development, since humans need practice with difficult people. Cowen counters that AI could handle the 5–10 hours a day Gen Z spends ‘servicing network connections’, freeing time for real friendship; Haidt responds with the Bursztyn finding on TikTok/Instagram — students would pay to be collectively freed from the platforms even though none would unilaterally quit, because leaving alone means being cut off from what everyone else is discussing — arguing this collective-action trap, not a genuine preference, is what keeps teenagers online, and that AI intensifies rather than escapes it.

The COVID school-closure test

Cowen presses a specific natural-experiment challenge: if in-person school time is what matters, COVID lockdowns removed six to eight hours of it a day, yet youth well-being (per Haidt’s own data) barely moved — doesn’t that undercut the causal story? Haidt’s rebuttal has three parts. First, American Time Use Survey data already showed minutes spent with friends collapsing steeply from 2018 to 2019, continuing at exactly the same rate through the COVID year with no visible inflection — the trend was already close to adult-like lows before lockdowns began. Second, school itself had stopped functioning as ‘time with friends’ once phones entered classrooms: Haidt and Jean Twenge’s analysis of PISA loneliness-at-school items shows loneliness rising across most of the developed world after 2012, alongside falling academic achievement, as phones displaced hallway and lunchtime socialising. Third, American schools are separately ‘depressogenic’ — citing Peter Gray’s finding that suicide rates fall in summer and rise when school resumes — which muddies lockdowns as a clean test, since removing a source of stress (school) and a source of connection (peers) happened simultaneously and in different measure for different kids.

Historical mood-shifts and the cross-country evidence

Cowen widens the lens: European and Anglo culture has swung through unexplained multi-decade or multi-century mood shifts before (the 17th century, the French Revolution) — why isn’t 2012 just another one, mistakenly pinned on smartphones? Haidt distinguishes single-year discontinuities (which he says are rare and demand a proximate cause) from slow-moving cultural drifts, and argues the geographic pattern supports causation rather than coincidence: the effect concentrates in the Anglosphere and Scandinavia — the wealthiest, most individualist, and fastest-smartphone-adopting societies — while Eastern Europe, growing more religious, shows improving youth mental health over the same period, and Catholic and Protestant Europe show a worsening gradient tracking secularisation. Cowen raises Sabine Hossenfelder’s citation of very small variance-explained figures linking screen time to mood; Haidt responds that the debate (himself and Twenge against researchers including Andrew Przybylski, Amy Orben, and Candice Odgers) has converged on a correlation near 0.04–0.1 for screen time generally, but nearer 0.17 once narrowed to girls and social media specifically — a size he says is comparable to other accepted public-health effects and not one parents would tolerate for a daily four-hour exposure. [?] The claimed convergence between Haidt and his research critics on this figure is asserted in conversation, not independently verified here.

Benefits, costs, and where Cowen and Haidt actually agree

Cowen challenges the book’s near-silence on social media’s benefits, citing Emergent Ventures teenagers who use Twitter to find scientific collaborators and are, in his account, ‘phenomenally’ productive. Haidt responds that cited benefits almost always describe older teens (16–18), that the internet broadly — blogs, podcasts, texting, multiplayer games — already supplies the connective benefit without social media’s specific harms, and that he has ‘never heard anyone make a good case’ that 11–13-year-olds benefit from it. The exchange narrows to genuine, explicit agreement: both concede the cost-benefit ratio is probably negative for 11–14-year-old girls on Instagram and TikTok, leaving the harder, unresolved case at 14–16, where Haidt favours a account-opening age of 16.

Gen Z’s productivity, and whether the generation is thriving or damaged

Haidt turns the interview around, asking Cowen to name members of Gen Z with major real-world impact, contrasting them with Steve Jobs- and Bill Gates-era teenage founders (citing a Patrick Collison–Sam Altman exchange on the absence of a major under-30 figure in software). Cowen counters with measurable domains — competitive chess, where young players (including young women) keep improving — and his own experience interviewing hundreds of Emergent Ventures grantees aged 16–30, whom he judges smarter, more disciplined, and more productive than prior generations, attributing the lack of famous teenage founders to a genuine shift in how technology companies are now built (small, senior research teams, not dorm-room solo founders). Haidt grants the internet makes it easier to find top performers, but says anecdotal access to exceptional individuals doesn’t establish that the median outcome has improved, and points to MrBeast as an example that ‘proves the rule’ — success within social media’s own attention economy, not influence beyond it.

Adjustment, collective action, and the four norms

Cowen’s broadest challenge: humans adjusted to agriculture, fire, and urbanisation, all disruptive in their early stages — why assume smartphones are different? Haidt’s response distinguishes technologies from fixed developmental requirements: mammalian childhood needs play, face-to-face interaction, eye contact, and synchrony in ways a virtual substitute structurally cannot supply, unlike, say, the Aztecs’ nixtamalisation workaround for a vitamin-C-poor diet. The two converge on a shared diagnosis — teenagers are stuck in a collective-action problem, mostly saying they are online only because everyone else is — and Haidt sets out the remedy: no smartphones before high school (flip phones permitted, for logistics only), no social media accounts before 16, phone-free schools (physically enforced, e.g. via Yondr pouches), and far more unsupervised free play, through the organisation Let Grow, co-founded with Lenore Skenazy and Peter Gray.

Age-gating, liability, and where government should (and shouldn’t) step in

Cowen tests whether Haidt’s programme amounts to government overreach into parental authority. Haidt says no on direct prohibition — he would not have government stop a parent from putting a 15-year-old on Twitter — but wants the COPPA-era data-consent age (13, set in the 1990s after lobbying watered down Ed Markey’s proposed 16) raised to 16 and enforced, with an optional parental-permission carve-out for younger teens. He opposes ID-based verification methods such as Louisiana’s driver’s-licence requirement for pornography sites, preferring platforms choose their own age-verification method against a target accuracy (he suggests 90–95% would earn meaningful liability protection, against an effective baseline near zero today, since most middle schoolers already hold accounts). On liability, Haidt supports narrowing platforms’ Section 230 shield specifically for design choices engineered to be addictive (infinite scroll, constant refresh), while preserving it for third-party content — arguing government created the current zero-guardrail, zero-liability environment and can therefore fix it. He analogises to shoplifting and Napster-era piracy: a real, non-trivial cost for violating an age rule changes behaviour even without perfect enforcement.

See also

• Jonathan Haidt — speaker
• Tyler Cowen — host
• Jonathan Haidt on Morality, Politics, Disgust, and Intellectual Diversity on Campus — Haidt’s first CWT appearance, on moral foundations theory and campus intellectual diversity
• The Anxious Generation — Haidt’s 2024 book, the occasion for this conversation