Abstract
As artificial intelligence (AI) technology is introduced into different areas of society, understanding people's willingness to accept AI decisions emerges as a critical scientific and societal issue. It is an open question whether people can accept the judgement of humans or AI in situations where they are unsure of their judgement, as in the trolley problem. Here, we focus on justified defection (non-cooperation with a bad person) in indirect reciprocity because it has been shown that people avoid judging justified defection as good or bad. We explore which decisions people would accept when AI and humans make different decisions in the justified defection. Two experiments revealed that only when AI judged positively and a human judged negatively, people rated the AI's judgement as better. The research suggests that people's acceptance of AI decisions depends on estimating the hidden intentions behind societal norms. It offers new insights into human norm analysis by examining the differences in acceptance between AI and human judgments.