Study Shows AI Counterarguments Overturn 30% of Human Moral Judgments

A Japanese research team has found that large language models possess substantial persuasive power in moral reasoning, with over 30% of study participants reversing their ethical positions after encountering AI-generated counterarguments. The study, led by Kouhei Masumoto, a psychology professor at Kobe University, tested 130 participants—including 56 younger adults and 74 older adults—using variations of the classic trolley problem ethical dilemmas. Participants were first asked to make judgments on two scenarios: a "switch dilemma" (deciding whether to divert a train to kill one person instead of five) and an emotionally charged "footbridge dilemma" (deciding whether to push someone off a bridge to stop the train). After providing their initial responses, participants were presented with ChatGPT-generated counterarguments opposing their choices.