This paper reports six longitudinal studies (each with three sessions) using a purpose-built corporate credit-rating AI that outperformed human users.
Key findings: as users gain experience their initial estimates move closer to AI advice, yet larger initial gaps from AI prompt more revision but lower relative dependence on AI (an effect shaped by decision-making experience). Adding extra information generally reduced alignment with AI, though when that information was close to AI advice it helped users adjust toward the AI.
The authors discuss implications for AI adoption, decision support design, and training in professional settings.
Authors & affiliations:
- Kyootai Lee: Sogang University
- Wooje Cho: Seoul National University
- Han-Gyun Woo: Sogang University
- Simon B. de Jong: Maastricht University
