SOCRATIC ARTIFICIAL INTELLIGENCE IN PHILOSOPHY EDUCATION: AN EPISTEMIC-MEDIATOR FRAMEWORK AND THE ASYMMETRY THESIS

Abstract

Generative artificial intelligence is optimised to produce answers, whereas philosophy education is constituted by the learner’s own reasoning; an argument the learner did not construct has no educational value. Existing guidance to retain human oversight names this mismatch without specifying which cognitive acts may be delegated. Using conceptual analysis, this paper develops a framework for a Socratic system whose defining constraint is the withholding of answers. Three results are established: artificial intelligence is best characterised as an epistemic mediator, entailing that what is mediated cannot be delegated to the mediator; the system’s behaviour is specifiable as five Socratic moves, each with a warrant and a bounding constraint; and, by the asymmetry thesis, the constraint on ethical judgement is logical rather than technical and is therefore not dissolved by gains in model capability. The framework is hypothetical; a three-arm verification design is specified.