Author: Nita Farahany Source: https://nitafarahany.substack.com/p/the-attention-evidence-gap Published: January 28, 2026
This essay is the 2.2 installment of Nita Farahany's "Advanced Topics in AI Law and Policy" course (Class 2.2 of approximately 30), published on her Substack on January 28, 2026. Where the earlier Class 2 framed attention as a candidate legal interest, this installment examines whether the empirical evidence on attention-capture harms can support legal regulation. Farahany characterizes the evidence base as substantial but methodologically contested, and argues that the form the science produces does not match the form courts require.
Summary of argument
Farahany frames the class as supplying the methodological predicate for any attention-rights doctrine, asking whether the evidence can actually support legal regulation. Her stated answer is "yes, with significant qualification."
The argument turns on a mismatch between two kinds of evidence. Courts asked to accept "compelling state interest" justifications for regulation require evidence that is causal, quantified, and generalizable. The science on attention and adolescent mental health instead produces evidence that is probabilistic, population-level, and context-dependent. Farahany argues that this mismatch is not a fault of the science, but that it tilts the legal field against regulation.
Key claims
The central illustration is the debate between Jonathan Haidt and the researchers Amy Orben and Andrew Przybylski. The Anxious Generation (Haidt, 2024) argues that smartphones and social media drove documented adolescent mental-health declines beginning around 2012. Orben and Przybylski (2018, Nature Human Behaviour) re-analyzed the same data and reported effect sizes that they compared to "wearing glasses or eating potatoes." Farahany notes that both sides have replicated their results and that the debate is not resolved.
On why the evidence is difficult to settle, Farahany observes that most studies are correlational, and that randomized experiments in which researchers control who uses social media show smaller and more mixed effects than survey work. Causation can run in multiple directions: depressed teens may use more social media; social media use may make teens more depressed; or both may respond to a third cause, such as the economic precarity following 2008.
Farahany also argues that subgroup effects are the most consequential part of the picture. Even where population-level effect sizes are small, she contends that the most vulnerable subgroups — girls aged 11–13, LGBTQ+ youth in unsupportive households, and children with prior mental-health conditions — experience effects orders of magnitude larger. The legal question she poses is whether population-level evidence is the right yardstick.
Drawing the threads together for the remainder of Week 2 and for Week 9 of the course, Farahany argues that if the evidence will not carry strict-scrutiny weight on its own, regulators must either accept the constitutional headwind or shift to architectural or data-minimization approaches that do not require proving content-specific harm.
Provenance
Published January 28, 2026 on Farahany's Substack as part of her "Advanced Topics in AI Law and Policy" course. Classified as a foundational essay.
Relationships
- part-of: Nita Farahany Advanced Topics course (Class 2.2 of ~30)
- related: AI Mental Health and Psychological Harm, Regulating Under Uncertainty
- previous: Inside My Advanced Topics Class 2: Can You Pay Attention? (Farahany, January 2026) next: Inside My Advanced Topics Class 2.3: The Laws That Miss the Point (Farahany, January 2026)