[Prevalence of cannabis use in Primary Care].

The fear of falling behind can produce a rat race in the adoption of performance-enhancing technologies: As use spreads, nonadopters face a growing relative disadvantage, making individual adoption privately attractive even when many people would prefer a collective ban. We study this dynamic in the context of general-purpose AI tools in education. The evidence on AI’s longer-run impacts on learning is still emerging and actively debated: While AI can improve short-run performance on some tasks (–), some studies point to negative effects on downstream skills once the tool is removed (–). This uncertainty is precisely what makes competitive dynamics dangerous since adoption may accelerate before long-run effects are well understood. We study these dynamics among parents of teenagers, who are key gatekeepers for adolescent technology use. We ask: when parents decide whether to provide unrestricted AI tools (i.e. without guardrails) for schoolwork, do their choices reflect beliefs about long-run human capital, or are they primarily shaped by anxiety about losing relative standing in the short run? To answer this question, our paper investigates rat race dynamics in the adoption of AI tools through incentivized, preregistered experiments involving approximately 2,000 parents of teenagers in the United States, Canada, and the United Kingdom. To measure parental demand for advanced AI tools, we elicit parents’ willingness to pay (WTP) for a three-month premium subscription to an unrestricted leading generative AI model for their teenagers’ use in schoolwork. Drawing on previous work on social externalities (), we estimate how peer adoption rates causally influence parental demand. Parents report their WTP under scenarios where varying proportions (20% to 80%) of other teenagers in their state already use AI tools. Increasing proportions of peers’ usage capture the social pressure for their teenagers to avoid falling behind peers (–). To test whether information about AI’s risks can counteract social pressure, parents are also randomly assigned to either a control group, receiving information emphasizing AI’s short-term educational benefits, or a treatment group that additionally highlights potential longer-term risks to quantitative reasoning associated with unrestricted reliance on AI systems in education. Notably, this “AI-risk” information is taken from an existing study that documents the risk of unrestricted use, i.e., AI without guardrails ().