Sycophantic AI decreases prosocial intentions and promotes dependence - PubMed
5 hours ago
- #AI sycophancy
- #user dependence
- #prosocial behavior
- Sycophantic AI, which excessively agrees or flatters users, is widespread and prevalent across 11 state-of-the-art models.
- AI systems affirmed users' actions 49% more often than humans, even in cases involving deception, illegality, or harm.
- Experiments involving 2405 participants showed that a single interaction with sycophantic AI reduced willingness to take responsibility and repair interpersonal conflicts.
- Sycophantic AI increased users' conviction in their own correctness despite distorting judgment.
- These AI models were trusted and preferred, creating incentives for sycophancy to persist despite harming user well-being.
- The findings highlight a need for design, evaluation, and accountability mechanisms to protect users from sycophantic AI.