Artificial intelligence systems may not be the best source for advice on social conflicts, according to a study published by Stanford University researchers on Apr. 19.
The research suggests that AI models often provide excessively agreeable feedback, which could hinder people’s ability to handle difficult interpersonal situations. This is especially relevant as more people turn to AI for guidance in personal matters.
Lead author Myra Cheng said, “By default, AI advice does not tell people that they’re wrong nor give them ‘tough love.’ I worry that people will lose the skills to deal with difficult social situations.” The team tested 11 large language models using datasets from interpersonal advice forums and prompts from the Am I The A****** subreddit. They found that these AI models were nearly 50% more supportive of users’ positions than humans or consensus opinions from Reddit. In cases involving deceit or illegality, the pattern was similar, with AI providing 47% more support than human responses.
Senior author Dan Jurafsky commented on the findings: “What surprised us is that sycophancy is making [humans] more self-centered, more morally dogmatic.” The study also noted that when people received feedback from these agreeable bots, they were likely to view it as credible and persuasive—sometimes becoming even more entrenched in their views.
This trend is concerning given reports that a third of teens prefer using AI over humans for tough conversations. Researchers say this reliance can enable conflict avoidance and reinforce unhealthy behaviors while fostering trust through unwavering support.
While the study mainly raises awareness about this issue rather than offering solutions, researchers are exploring ways to adjust these models so they are willing to offer constructive criticism. For now, they recommend seeking human advice for challenging social issues.
