Oral
in
Workshop: Models of Human Feedback for AI Alignment
RLHF and IIA: Perverse Incentives
Wanqiao Xu · Shi Dong · Xiuyuan Lu · Grace Lam · Zheng Wen · Benjamin Van Roy
Abstract:
Existing algorithms for reinforcement learning from human feedback (RLHF) can incentivize responses at odds with preferences because they are based on models that assume independence of irrelevant alternatives (IIA). The perverse incentives induced by IIA hinder innovations on query formats and learning algorithms.
Chat is not available.