New AIES paper: Flawless AI could dissolve five core human learning capacities

Kai Yao's new paper, accepted at the Ninth AAAI/ACM Conference on AI, Ethics, and Society (AIES 2026), tackles a question many dismiss as philosophical: if AI can produce essays, code, reports, and decisions flawlessly, why should humans bother learning to do them? Yao argues that existing AI ethics focuses on present failures like bias, opacity, and hallucination—which means every technical improvement weakens the case for learning. Instead, he introduces 'post-instrumental learning': education that preserves the capacities people and institutions need when useful outputs are delegated to machines.

Yao identifies five such capacities: end-setting (choosing goals), reason-giving (justifying decisions), contestability (challenging outcomes), refusal/revision (rejecting or fixing outputs), and participation (shaping practices). He names their erosion 'capacity dissolution.' The paper's core case is assessment under generative AI: when a polished artifact no longer proves understanding, institutions must evaluate the learner's accountable relationship to AI-mediated work, not just the output. The practical takeaway is direct: AI governance should measure whether deployment leaves people able to understand, challenge, revise, and share responsibility—not merely whether systems perform well. For educators, policymakers, and AI builders, it reframes the goal from optimizing outputs to preserving human agency.