The Case for Learning in the Era of AI

Jon Behnken argues that the rise of AI forces a renewed examination of what it means to know and concludes that genuine knowledge is inherently human while machines can only approximate it. He distinguishes truth (how things are) from knowledge, arguing knowledge requires justified belief and protection against epistemic luck: a correct prediction without explanation doesn't count. Examples include a rigged die and Copernicus’s slow social acceptance to show knowledge needs both internal conviction - the self - and social certification. Behnken asks readers to assume machines lack a self and therefore cannot truly ‘own’ beliefs, even if they simulate models effectively.

He then moves to practical epistemology: process reliabilism matters because reliable methods let us trust and generalize findings. Opaque AI systems undermine that trust by offering stochastic, uninterpretable outputs, which would force humans either to demand transparency or to exhaustively test every claim. The upshot is a middle path: use AI as a generator of hypotheses but keep humans as judges and verifiers. He concludes with a call to maintain and expand human learning - study math, read code, stay curious - because surrendering learning to machines surrenders the capacity to judge what counts as knowledge.