Large language models can decipher human decisions

Explanations Provide Important Clues

The combination of decisions, participants’ own explanations, and AI analysis showed that the participants’ statements provide valuable insights into their thought processes. Another finding of the study: People do not always make decisions according to the same pattern. The considerations that guide them depend on the situation at hand. Sometimes the prospect of the best possible outcome plays a greater role, while at other times it is the desire to avoid a potential loss.

Participants’ own explanations thus become a valuable source of information. “Many important decisions – from financial planning and medical decisions to social dilemmas, the use of technology, and public policy – involve complex trade-offs that cannot be fully understood simply by observing decisions,” explains Fuławka. He emphasizes that such explanations are particularly valuable in precisely these areas.

The study also shows how LLMs can help researchers evaluate such explanations on a large scale in the future. This opens up new avenues for more closely examining human decisions in more realistic and complex situations.


Original publication:
K. Fuławka, R. Hertwig, & D.U. Wulff, Large language models accurately identify decision reasons in verbal reports, Proc. Natl. Acad. Sci. U.S.A. 123 (27) e2526798123