Agency-Preserving AI for Adolescents

The project addresses a neglected human-factors layer of AI safety. As advanced AI systems become better at performing reasoning and judgment, users may achieve higher assisted performance while losing opportunities to exercise, verify, and retain the underlying capacities. In adolescents, this may weaken the human judgment and oversight capabilities on which safe use of advanced systems depends.

The causal chain is as follows. First, the study will measure whether delegating developing cognitive operations to AI changes later independent evidence evaluation, reasoning, and transfer. Second, it will test whether reduced independent capacity or difficulty detecting flawed AI advice is associated with delegation itself rather than only with task difficulty or reduced practice. Third, it will test periodic AI-restricted assessments as a governance signal. Finally, validated findings will be translated into safeguards that guide AI systems to cue, scaffold, co-perform, or refrain according to demonstrated human capacity and uncertainty.

The expected contribution is a measurable safeguard against a pathway by which increasingly capable AI could produce dependent or over-trusting human decision-makers. The project does not claim to solve catastrophic risk directly. It aims to preserve human verification, judgment, transfer, and meaningful oversight, and to publish evidence-bounded guidance for safer adolescent-facing AI systems.