Worst-First Thinking

In the AI Story

The pattern's signature is its indifference to base rates. Crime rates had been declining for fifteen years when Skenazy's nine-year-old son made his 2008 subway ride; the probability of stranger kidnapping remained approximately one in 1.4 million; the streets he navigated were statistically safer than those Skenazy had roamed as a child. None of this mattered to the reaction. The cultural immune system had decided the world was dangerous, and anyone acting on evidence rather than feeling was committing a form of negligence. The availability heuristic does the cognitive work: vivid catastrophes are easy to imagine, and ease of imagination is mistaken for probability.

The AI discourse of 2025–2026 reproduced the pattern with remarkable fidelity. The vocabulary shifted — nobody was talking about playgrounds — but the cognitive architecture was identical. Parents imagined children becoming intellectually passive, students losing the ability to write, young people developing parasocial relationships with AI companions that would hollow them out. The scenarios had grounding in real research on automation dependence, depth atrophy, and artificial intimacy. The fear was not irrational. The question was what to do with well-founded fear, and worst-first thinking was the wrong answer for the same reason it had always been wrong.

Skenazy distinguishes between dismissing fear and dismissing the evidence that sometimes supports it. Her argument is not that risks are imaginary. It is that prohibition is the wrong instrument for managing them, because prohibition prevents the development of the only thing that can actually mitigate the risk: the child's own growing competence. Every protective measure has a developmental cost. That cost is not a philosophical abstraction — it is measurable, documented in the self-efficacy research and the longitudinal studies of iGen — and it is systematically ignored by institutions optimizing for liability.

The twelve-year-old whose parent discovers her using Claude for a school essay is the diagnostic scene. In the worst-first framework, this is intellectual fraud. The parent confiscates the device, delivers a lecture, contacts the school. The parent never asks what the child actually did — whether Claude helped her see the concept from a different angle, catch a mistake the textbook made, follow her curiosity past the curriculum's limit. The parent skipped the actual child and responded to the imagined catastrophe. This is the pattern Skenazy has been fighting since the subway ride.

Origin

Skenazy developed the concept through the response to her 2008 column about letting her nine-year-old son ride the New York City subway home alone. Dubbed "America's Worst Mom" by national media, she noticed that the outrage had no relationship to the statistical reality of child safety. The cultural response was calibrated to a feeling, not evidence. Over the next decade, she catalogued hundreds of analogous cases — running banned at recess, CPS investigations for unsupervised park visits, helmets for tetherball — and identified the common cognitive mechanism.

The framework acquired wider purchase through Skenazy's collaboration with Jonathan Haidt and Greg Lukianoff on the safetyism critique. When AI discourse erupted in 2025–2026, the framework migrated almost unchanged into debates about children, schools, and generative systems — the same cognitive error wearing different vocabulary.

Key Ideas

Base rate neglect. Worst-first thinking treats the worst outcome as the most likely one, ignoring actual probability distributions and substituting vivid imagination for statistical evidence.

Protection has cost. Every protective measure prevents the development of capacities that can only be built through encounter; the cost is invisible because it is an absence rather than an event.

Fear is not data. That adults are afraid of AI's effects on children does not by itself establish that prohibition is the correct response — the same fear has produced destructive policy in every previous generation.

The pattern migrates. Worst-first thinking is not about any specific technology; it is a reflexive cognitive architecture that finds new domains with each generation, and the AI discourse is its current home.

Debates & Critiques

Critics argue that AI presents genuinely novel risks — artificial intimacy, fluent fabrication, depth atrophy — that the playground framework cannot accommodate. Skenazy concedes that AI's failure modes are invisible in ways physical risks are not. Her response is that the invisibility changes the design of the response, not its underlying logic: the child still learns through encounter, the competence still develops through practice, and the protection-versus-development tradeoff still applies.