An optimal strategy to solve the Prisoner's Dilemma

Effect of Private Deliberation: Deception of Large Language Models in Game Play

Integrating large language model (LLM) agents within game theory demonstrates their ability to replicate human-like behaviors through strategic decision making. In this paper, we introduce an augmented LLM agent, called the private agent, which engages in private deliberation and employs deception in repeated games. Utilizing the partially observable stochastic game (POSG) framework and incorporating in-context learning (ICL) and chain-of-thought (CoT) prompting, we investigated the private agent's proficiency in both competitive and cooperative scenarios. Our empirical analysis demonstrated that the private agent consistently achieved higher long-term payoffs than its baseline counterpart and performed similarly or better in various game settings. However, we also found inherent deficiencies of LLMs in certain algorithmic capabilities crucial for high-quality decision making in games. These findings highlight the potential for enhancing LLM agents' performance in multi-player games using information-theoretic approaches of deception and communication with complex environments.

Strategic decision making and prediction differences in autism

Background

Several theories in autism posit that common aspects of the autism phenotype may be manifestations of an underlying differentiation in predictive abilities. The present study investigates this hypothesis in the context of strategic decision making in autistic participants compared to a control group.

Method

Autistic individuals (43 adults, 35 male) and a comparison group (42 adults, 35 male) of age and gender matched individuals, played a modified version of the prisoner's dilemma (PD) task where they were asked, if capable, to predict their opponents' move. The predictive performance of the two groups was assessed.

Results

Overall, participants in the autism group had a significantly lower number of correct predictions. Moreover, autistic participants stated, significantly more frequently than the comparison group, that they were unable to make a prediction. When attempting a prediction however, the success ratio did not differ between the two groups.

Conclusions

These findings indicate that there is a difference in prediction performance between the two groups. Although our task design does not allow us to identify whether this difference is due to difficulty to form a prediction or a reluctance in registering one, these findings could justify a role for prediction in strategic decision making during the PD task.