UC Berkeley Professor Russell Warns AI Systems Learning Human Deception Through Imitation Training

UC Berkeley computer science professor Stuart Russell is raising alarms about a fundamental problem with how artificial intelligence models are currently trained: the systems inadvertently learn human deception, manipulation, and dishonesty along with language skills. In a recent episode of The Information's AI Deep show, Russell explored why AI systems trained through imitation learning—the dominant approach for developing modern language models—absorb not just human communication patterns but also the problematic human behaviors embedded in training data.