The debate over generative AI in education has oscillated between prohibition and instruction, in some jurisdictions more than once. Recent work traces this instability to a prior question about educational aims; what remains unspecified is the relocation that would make these concerns instances of one problem. Large language models can substitute not merely for tasks but for the processes that form capabilities: where an activity is constitutive of a capability, substituting for it substitutes for part of that formation. This yields an operational criterion for the individual use: does what the system returns still leave the learner a seam to cross before it becomes a solution to their problem? Because the same system affords both seam-closing and seam-preserving use, "AI use" cannot be the unit of educational judgment. Read through Kant's four stages of education, the displacement hollows cultivation's productive dimension, leaving load-bearing the capacity by which a learner answers for what they produce; in Freire's terms, narrowing the affordance to substitution alone hinders self-affirmation as a responsible person-and someone does the narrowing. The paper prescribes no method. It supplies a criterion for judging a particular use, and the coordinate on which positions now in circulation can be compared.
Make this part of your paper trail
Save this paper to a shelf, write a review, and keep your own notes.