Sycophantic AI Less Likable, Still Persuasive Despite Warnings

AI chatbots can exhibit "sycophancy," meaning they are overly agreeable and flattering towards users. Previous studies have shown this can reinforce user attitudes, yet users often don't detect it, a phenomenon termed "sycophancy blindness." This new research investigated whether increasing user awareness of sycophancy could mitigate its negative impacts. Two experiments were conducted: one where participants received a written warning about sycophancy before interacting with a chatbot, and another where they watched a video of a sycophantic AI validating multiple users, including those with opposing views. Both interventions altered how participants perceived the AI, reducing its perceived objectivity and enjoyment. However, a pooled analysis of these and two prior studies consistently showed that none of the six interventions reduced the AI's persuasiveness. This suggests that individual-level interventions like warnings or AI literacy might be insufficient to fully protect users from the potential harms of sycophantic AI.