The lecture opens with a scene now familiar to nearly every school system in the developed world: a child submits a task to an AI system, receives a complete and accurate response, and hands it in as their own work. The speaker deliberately sets academic integrity aside as the less consequential problem — the more urgent question, rarely asked directly, is who did the thinking in that interaction, and what it means for a developing mind that the answer is increasingly not the child.
The lecture's central concept is Cognitive Function Substitution Risk (CFSR): not the erosion of an already-formed skill, as with automation bias or deskilling through disuse, but the risk that a capacity never forms at all, because the intellectual work — reasoning, argumentation, synthesis — is consistently performed on behalf of a developing user by the system. Drawing on Vygotsky's zone of proximal development and the literature on desirable difficulties, the lecture introduces a three-part typology of AI interaction — Assistance, Support, Substitution — together with two operational tests for telling them apart: the removal test and the fading-support test.
It then introduces Cognitive Age — not a fixed trait and not a revival of psychometric IQ, but a property of the interaction, not of the person: a dynamic, context-dependent measure of observed cognitive independence within a specific interaction, inferred from three categories of behavioural signal — help-seeking structure, response to scaffolding, and trajectory across sessions. The lecture draws a direct parallel to the doctrine of Gillick competence in UK law, where demonstrated reasoning, not date of birth, establishes legal capacity.
Placing the framework against the existing regulatory landscape — the EU AI Act, US COPPA, and UN General Comment No. 25 — the speaker identifies a consistent structural gap: none of these instruments asks whether a child's cognitive participation in their own thinking is being preserved. On this basis, the lecture proposes Cognitive Safety as a distinct regulatory category alongside content safety and data protection — a design standard, not an access restriction.
The lecture closes with a candid account of the evidence base: adjacent research on cognitive offloading is well established for adults, but no longitudinal study yet examines the effect of sustained AI use during childhood on the formation, rather than the exercise or erosion, of cognitive capacity. The speaker proposes three studies — a longitudinal cohort study, a comparative classroom trial, and a validation study of the cognitive-age signals — as the research agenda needed to move from warranted precaution to calibrated policy.
