The Internalization Boundary: Human Capability Formation and Assurance in the Age of Artificial Intelligence

Artificial intelligence is weakening a relationship on which modern education has longrelied: the connection between competent task performance and capability possessed bythe person performing the task. Historically, producing a sophisticated analysis, proof,diagnosis, program, or argument generally required substantial prior internalizationof the knowledge and cognitive processes involved. Education could therefore useperformance both as a means of forming capability and, imperfectly, as evidence thatcapability had been formed.Generative artificial intelligence disrupts both relationships. Increasingly sophisticated cognitive work can now be delegated to external systems, allowing high-qualityperformance without equivalent individually possessed capability. Existing research hasresponded by examining cognitive offloading, human–AI hybrid intelligence, epistemicagency, selective delegation, and the conditions under which AI assistance augmentsor erodes learning. These approaches substantially clarify how learners should dividecognitive labor with AI. A different institutional question nevertheless remains: whenexecution can be delegated, what capabilities should educational institutions still requirehumans to possess, and what capabilities should degrees, qualifications, and professionalcredentials continue to warrant as belonging to the human agent?This paper develops the internalization boundary as a framework for answeringthat question. It distinguishes three decisions that AI increasingly separates: taskdelegation, concerning who or what performs an operation; capability internalization,concerning what the human must still understand or be able to do; and capabilityassurance, concerning what an educational or professional institution must be ableto establish and credibly certify about the individual. The paper conceptualizes theinternalization boundary as an institutionally selected human capability floor ratherthan as a defense of maximal internalization. Four considerations justify retaining capabilities within that boundary even whenexecution is technologically delegable: developmental necessity, epistemic control,accountable agency, and resilience. The boundary is therefore dynamic and developmentally asymmetric. Experts may safely externalize operations that novices still need toperform because those operations participate in the formation of later judgment. Likewise, AI-assisted performance cannot by itself establish that durable human capabilityexists.The argument does not oppose extended or distributed cognition. It begins fromtheir premise. The educational question is what the human component of increasinglydistributed cognitive systems must still be capable of, and what educational institutionscan legitimately claim to have formed and verified. AI therefore transforms educationnot only by changing how tasks are performed, but by forcing societies to make explicita previously less visible institutional choice: what must remain a capability of thehuman?Keywords: artificial intelligence; education; internalization boundary; human capability;cognitive offloading; capability assurance; assessment; human–AI collaboration; expertise