Story perspectives

AI Misjudges Human Rationality in Strategic Games, Study Reveals

12/26/2025

35 2

Summary

Story summary

Researchers at HSE University found that AI models, including ChatGPT and Claude, overestimate human rationality in strategic games like the Keynesian beauty contest.
The study, led by Dmitry Dagaev, replicated classic experiments where AI played Guess the Number against human opponents.
Results indicated that AI adapted its choices based on opponents' traits but struggled to identify dominant strategies in two-player games.
Dagaev emphasized the importance of ensuring AI mimics human decision-making as it increasingly replaces humans in operations.
The study underscores the necessity for ongoing research into AI behavior relative to human behavior.

1 of 1

Story summary

Researchers at HSE University found that AI models, including ChatGPT and Claude, overestimate human rationality in strategic games like the Keynesian beauty contest.
The study, led by Dmitry Dagaev, replicated classic experiments where AI played Guess the Number against human opponents.
Results indicated that AI adapted its choices based on opponents' traits but struggled to identify dominant strategies in two-player games.
Dagaev emphasized the importance of ensuring AI mimics human decision-making as it increasingly replaces humans in operations.
The study underscores the necessity for ongoing research into AI behavior relative to human behavior.