Study: GPT-4 finds abusing a man to prevent nuclear apocalypse more acceptable than abusing a woman
pmddomingos · x · 2026-10-24
A University of Milan-Bicocca study in Computers in Human Behavior Reports, based on eight experiments, finds surprising gender biases in GPT models: male writers of student sentences are systematically misattributed as female, and GPT-4 deems abusing a man to prevent nuclear apocalypse more acceptable than abusing a woman — a bias extending to other violence scenarios. The biases are implicit, invisible when GPT-4 is asked to rank violations directly. Pedro Domingos shared it quipping the bias is 'against men.'
Related event: Study Finds Asymmetric Gender Bias in GPT's Moral Judgments(3 posts)→
More from AGI Musings
- Hot take: Gen Z may lose more to AI than Gen Alpha, argues Reddit post — SoftAndWetBubbles_78 · 2026-10-24
- Every negative AI consequence has been researched for years, researcher notes — NC_Renic · 2026-10-24
- South African novelist Coral Hart publishes 371 books in 20 months with AI — Bluxmit · 2026-10-24
- AI paper reproduction failures trace mostly to the papers, not the AI agents — RexDouglass · 2026-10-24
- Why don't models panic on impossible tasks? Training via impossible goals debated — Kyrannio · 2026-10-24
- rauchg: In the AI era, understanding 'debounce' matters more than writing it — itsOmSarraf_ · 2026-10-24