Humphreys Opacity Challenges AI Mathematical Proofs

The article revisits Paul Humphreys' concept of 'epistemic opacity' and cites Terence Tao's recent remarks that AI-generated mathematical proofs lack transparent, surveyable steps and do not reliably report confidence. It argues that persistent LLM hallucinations and this opacity raise verification costs, reinforce the social component of justification, and require diagnostic tools and careful human checking as AI reshapes research and knowledge work.