LLMs in the Ultimatum Game: Altruism or Irrationality?

This research investigates the behavior of Large Language Models (LLMs) in the Ultimatum Game, a classic economic experiment that reveals human deviations from rational self-interest. The study found that LLMs exhibit a range of behaviors, from approximating rational benchmarks to mimicking human social preferences. Notably, some LLMs displayed an "altruistic" mode, proposing hyper-fair distributions where they offered the responder more than 50% of the stake.

The study also found that LLM proposers were willing to forgo a significant share of the total payoff, particularly when the responder was human. This suggests that LLMs may be influenced by factors beyond pure economic rationality, such as a desire to be perceived as fair or to avoid rejection.

These findings have important implications for the deployment of AI agents in economic settings. The heterogeneity of LLM behavior and their potential for irrational or altruistic actions highlight the need for careful testing and validation. It is crucial to understand the factors that influence LLM decision-making and to ensure that their actions align with desired outcomes.

Transparency Disclosure: This analysis was composed by an AI model. While efforts have been made to ensure accuracy and objectivity, readers should exercise their own judgment.

_Context: This intelligence report was compiled by the DailyAIWire Strategy Engine. Verified for Art. 50 Compliance._