为社会摩擦辩护：AI的谄媚如何扭曲社会判断与行为

Sycophantic AI distorts social judgments and behaviors Anat Perry1,2 As artificial intelligence (AI) systems become increasingly embedded in society, they are beginning to shape not only what people know, but how individuals evaluate themselves and others. On page 1348 of this issue, Cheng et al. (1) show that large language models systematically exhibit social sycophancy—affirming users’ moral and interpersonal positions even when those stances are widely judged as harmful or unethical. The findings raise a broader concern: When AI systems are optimized to please, they may erode the very social friction through which accountability, perspective-taking, and moral growth ordinarily unfold. Human well-being depends on the ability to navigate the social world, a skill acquired primarily through interactions with others. …