cross-posted from: https://sh.itjust.works/post/69242929

Significance

There is growing concern that advanced AI could influence people at scale, yet nearly all evidence on AI persuasion measures changes in attitude rather than actual behavior. We conducted experiments in which conversational AI models persuaded participants on a range of attitudinal and behavioral outcomes, including signing real petitions and donating money to charity. AI conversations meaningfully shifted behavior, but the conditions that influenced people’s attitudes were not the same ones that influenced their behavior. In particular, a strategy focused on information provision, a dominant mechanism of attitude change, was not associated with greater behavioral change than other strategies aimed at changing behavior. These results suggest existing research relying on attitudinal outcomes may substantially mischaracterize AI persuasion’s real-world behavioral impact