Generative AI Triggers Welfare-Reducing Decisions in Humans

Knowing you're playing against AI makes people less fair and cooperative: When study participants (N=3,552) were explicitly told their partner was ChatGPT in economic games, they showed "diminished fairness, trust, trustworthiness, cooperation, and coordination" compared to when they thought they were interacting with humans.

Transparency about AI may backfire on human welfare: Counterintuitively, the study found that making AI involvement explicit had "detrimental effects on welfare," suggesting that well-intentioned transparency policies could inadvertently harm social outcomes by triggering less cooperative human behavior.

People readily delegate decisions to AI, especially when it's hidden: Participants "frequently delegate decisions to ChatGPT, particularly when the AI's involvement is undisclosed," raising concerns about how AI might be used in social contexts without people's knowledge or consent.

Humans struggle to distinguish AI from human decisions: The research found that "individuals struggle to discern between AI and human decisions," meaning transparency alone may not solve the problem if people can't reliably identify when they're interacting with AI anyway.

The welfare impact depends on whether people know about the AI: Importantly, the study found "no adverse welfare effects when individuals are uncertain about whether they are interacting with a human or generative AI," suggesting the negative effects are specifically triggered by explicit knowledge of AI involvement rather than AI use itself.