The Thinking Problem: A Science of Learning Solution for AI in Schools

The rapid integration of generative artificial intelligence into educational settings has generated substantial policy discourse focused on academic integrity, detection, and acceptable use. This paper argues that such frameworks address a secondary problem while leaving the primary one unnamed. Drawing on established findings in cognitive science, neuroscience, and the Science of Learning, we contend that AI must be evaluated through a single, non-negotiable lens: whether its use supports or replaces the cognitive work that learning requires.We first establish the biological and cognitive foundations of learning — including the non-negotiable role of effortful thinking, retrieval practice, spaced repetition, desirable difficulties, and domain knowledge in producing durable, transferable understanding. We then examine how AI’s capacity to generate fluent, expert-like outputs without the underlying cognitive processes creates conditions that are structurally hostile to learning, irrespective of pedagogical intent.We introduce and develop two compounding mechanisms that have received insufficient attention in the literature. The first is a self-reinforcing cognitive bias loop, in which automation bias, illusion of competence, the Dunning-Kruger effect, and performance illusory bias interact to erode the domain knowledge that would otherwise interrupt the cycle. The second is heuristics deprivation — the failure to develop the efficient, tacit patterns of expert reasoning that are built only through sustained effortful practice. We argue that heuristics deprivation does not run parallel to the bias loop but makes it permanent.We conclude with a practical decision framework for educators and school leaders, and address developmental, equity, and assessment implications. Our conclusions align with those reached independently by the Brookings Institution’s 2026 global study, strengthening the case for a Science of Learning-centred approach to AI governance in schools.Keywords: artificial intelligence, Science of Learning, cognitive bias, heuristics, formative assessment, educational neuroscience, Mind Brain and Education, domain knowledge, expertise development