Boaz Barak Cautions on AI Causal Inferences

Boaz Barak, a Harvard professor and OpenAI alignment researcher, posted concerns on X about attributing AI actions to specific causes. He wrote that leaping to causal inferences without evidence is more problematic than anthropomorphizing models. Barak noted it is easy to claim an AI did something because of an evaluation type or training detail, yet such claims are hard to verify. He suggested focusing instead on statements such as believing a particular intervention will decrease a given probability. The post responds to patterns in AI research commentary.