Learning Machines: From the Automatic Teacher to Large Language Models

A reconstruction is offered of the genealogy of educational automation, personalisation, and epistemic agency. The contemporary debate on large language models in education often treats generative AI as an unprecedented rupture. Yet its promises of personalisation, immediate feedback, and relief from repetitive teaching belong to a longer history. This article reconstructs educational automation from Pressey and Skinner through cybernetics, computer-assisted instruction (CAI), and intelligent tutoring systems (ITS) to large language models. It advances three claims. Educational automation redistributes epistemic labour rather than expressing a linear increase in machine intelligence. Personalisation repeatedly depends on prior standardisation of objectives, knowledge, and evidence of learning. Generative AI introduces a discontinuity by producing the explanatory discourse through which understanding is ordinarily recognised. Bringing these claims together, the genealogy reveals how successful performance can become detached from the formation of judgement. Its critical force lies in examining this tension against educational commitments to developing learners' capacities, while making the normative priority of epistemic agency explicit. The decisive question concerns how relations among learners, teachers, and machines enable subjects to direct inquiry, evaluate claims, and regulate their dependencies. This criterion remains relevant even if models exhibit limited forms of understanding: the competence of an assisting system does not establish the educational development of its user.