Comparing the effects of dark and light personality traits in humans and AI on moral dilemma judgments

This study examined relationships between dark/light personality traits and moral dilemma judgment, comparing humans and large language models (LLMs). Using the CNI model, we analyzed 404 Chinese participants and 2,092 LLM-generated responses. In humans, superordinate dark traits negatively predicted moral norm sensitivity, whereas superordinate light traits positively predicted it. At the subordinate level, sadism and Machiavellianism negatively predicted moral norm sensitivity, Kantianism positively predicted it, and faith in humanity predicted stronger general action preference. For LLMs, only the direction of effects on moral norm sensitivity was consistent with humans; substantial differences appeared in consequence sensitivity and general action preference. LLM-generated dark personality showed lower moral norm sensitivity and stronger action preference than LLM-generated light personality, whereas consequence sensitivity showed no stable pattern. The results reveal that LLMs only show a preliminary simulation capacity and fail to reflect the complex features of personality and moral judgment.