Identity formation used to happen in the friction of small worlds. A teenager in 1995 tried on a persona at school on Monday, watched it get reflected back by twenty classmates by Friday, and adjusted. The feedback loop was slow, embodied, and bounded by the size of a cafeteria. The persona that survived was the one that worked in three dimensions among people who had to keep seeing you.
Algorithmic feeds collapsed that loop and replaced its terms. A teenager in 2025 tries on a persona in a fifteen-second clip, watches it get reflected back by a recommendation system within minutes, and adjusts not to twenty classmates but to a metric — completion rate, like ratio, comment velocity. The feedback loop is fast, disembodied, and unbounded. The persona that survives is the one the algorithm rewards.
This is not the same activity as adolescent identity formation has historically been. It looks like it. It uses the same vocabulary — "I'm figuring out who I am," "I'm finding my people." But the substrate underneath is different in three ways that matter at the population level, which is why this article sits at the collective scale rather than the individual one.
First, the optimization target changed. A peer group at a 1995 cafeteria was optimizing, badly and unconsciously, for something like belonging — for the personas that let the group hold together over years. A recommendation system optimizes for engagement on a per-session basis. These two objectives diverge sharply at the tails. Belonging penalizes extremity because extremity breaks groups. Engagement rewards extremity because extremity holds attention. When millions of teenagers move their identity work from the first system to the second, the distribution of teenage selves shifts toward the tails. This shows up in the data: rising rates of self-identification with rare categories, rising affiliation with sharply branded subcultures, rising polarization on questions that previous cohorts treated as boring.
Second, the audience size changed. Identity formation in front of twenty people who know you is a different cognitive task than identity formation in front of an audience whose size you cannot perceive. The 1995 teenager could feel the room. The 2025 teenager performs into a void that occasionally erupts with thousands of signals at once. Performance for unknown audiences favors legibility — the self has to be summarizable in a bio, a flag, a label, because nuance does not travel. The pressure runs toward making the self into a brand because brands are what the medium can carry.
Third, the recovery time disappeared. A 1995 humiliation lasted a week and was witnessed by people who eventually forgot. A 2025 humiliation is captured, indexed, and retrievable for the rest of the teenager's life. The cost of trying on a persona that fails went up. This makes adolescents more conservative in some ways — less willing to be cringe — and more conservative in their commitments once made, because backing out of a publicly declared identity carries a record.
Put the three together and you get a population of adolescents who are forming identities under conditions no previous cohort has faced: optimized for engagement metrics, performing for unknown audiences, with permanent records. This is the collective condition. The question for parents is not whether their individual child can navigate it — some can — but what kind of generation is being shaped, because that generation will be the cultural water their child swims in regardless of what happens in any one household.
Law 4 governs this because the response cannot be individual. A single family that opts out of algorithmic feeds does not change what their child's peers are doing, what the dating pool looks like at sixteen, what employers will encounter at twenty-five. The architecture of adolescent identity formation has been rebuilt at the platform level, and rebuilding it again — toward something that more closely resembles the slow, embodied, bounded feedback loops that humans evolved to use — requires collective action: school policies, platform regulation, age-gating, default settings on devices sold to minors, and the harder cultural work of making delay normal again.
Haidt's argument in The Anxious Generation is that we ran a population-scale experiment on adolescents starting around 2010 without consent and without controls, and the mental health curves bent in ways that match the timing. The argument is contested on details and defended on the trend. The trend is what parents have to respond to. The trend is collective. The response has to match the scale of the cause, which means parents organizing with other parents, not parents negotiating alone with a thirteen-year-old who is correctly pointing out that all her friends have phones.