This study explores the placebo effect of AI in Human-Computer Interaction (HCI). It suggests that even when AI is described negatively, participants perform better if they believe they’re interacting with an AI system, even if there is none. This effect is attributed to the users gathering more information when they think they’re using an AI. The study also indicates that user expectations about AI are generally biased and are not affected by negative descriptions. The findings highlight the impact of user expectations on AI interactions and evaluations.

 

Publication date: 28 Sep 2023
Project Page: https://arxiv.org/abs/2309.16606v1
Paper: https://arxiv.org/pdf/2309.16606