This study investigates how people evaluate probabilistic information when it is provided by AI versus human experts. Previous research on algorithm aversion (Dietvorst et al., 2015) and appreciation (Logg et al., 2019) has shown that the trust placed in algorithmic advice differs from that in human judgment, but little is known about how people value probability information depending on its source. Building on Keren and Teigen (2001) and Nakamura (2008), the study adopts an information-theoretic framework in which the perceived value of probability is interpreted through Kullback–Leibler (KL) divergence, reflecting the gap between presented probabilities and participants’ prior beliefs. A total of 222 university students evaluated 23 probability phrases including numerical and verbal probability expressions concerning stock price increases, presented either by AI or an expert. Results showed that both conditions produced U-shaped evaluations, with extreme probabilities rated as more informative than moderate ones. Model-based estimation indicated that participants assumed lower prior probabilities of stock increases and assigned weaker KL-based weighting under the AI condition, suggesting a more conservative stance toward AI-provided information. These findings highlight that AI advice may reshape situational assumptions themselves, extending beyond trust differences emphasized in prior algorithm aversion research.
