Helping ‘Bad’ People Drew Disapproval From Some of 21 AI Models
A University of Amsterdam study in PNAS Nexus asked 21 large language models to judge people who helped, or refused to help, others with good or bad reputations. The models agreed on good people and split on bad ones. We explain the four social norms, the model-by-model figures, what the norms would do to cooperation, and what organisations should test.
