AI Evaluation in Non-Verifiable Domains Should Borrow Qualitative Social Science Methods

AI practitioners, citing Ethan Mollick, argue that evaluating non-verifiable outputs like writing and creativity should draw on a century of qualitative social science research, since human subjective judgment remains the de facto benchmark for such hard-to-quantify tasks.

2026-08-17 ~ 2026-08-17 · 2 related posts