When AI Does the Work, What Is Learning For? Post-Instrumental Learning and the Risk of Capacity Dissolution

Learning must serve purposes beyond just task completion. The paper argues that as AI systems become capable of producing essays, code, reports, and decisions, we need to rethink why learning matters. If we only value learning as a way to compensate for AI failures, then every technical improvement to AI undermines the case for learning itself—a problematic foundation for education and professional development.

Five critical human capacities are at risk of "capacity dissolution." The authors identify specific abilities that could erode if we over-delegate to AI: end-setting (deciding what goals matter), reason-giving (explaining decisions), contestability (challenging outcomes), refusal/revision (saying no or changing course), and participation (having a voice in decisions). These capacities are essential for accountability and democratic governance, independent of whether AI systems work well.

Assessment practices must change fundamentally in the age of generative AI. Since polished AI-generated artifacts no longer reliably show what a person actually understands, institutions cannot simply evaluate the final product. Instead, they must assess "the learner's accountable relation to AI-mediated work"—essentially, evaluating whether people understand, can explain, and take responsibility for what AI produces on their behalf.

Good AI governance requires more than technical fixes. The paper argues that responsible AI policy should evaluate not only whether systems perform their tasks well, but also whether deploying them leaves people with the ability to understand, challenge, revise, and share responsibility for the practices those systems mediate. This shifts focus from AI capability to human agency.