Steve Phelps, Yvan I. Russell. “The Machine Psychology of Cooperation: Can GPT Models Operationalise Prompts for Altruism, Cooperation, Competitiveness and Selfishness in Economic Games?” arXiv:2305.07970, May 2023, revised June 2024 [preprint]
Can a large language model be instructed into altruism, selfishness, competition or cooperation? This study puts GPT models through the classic economic games and finds they can operationalise such prompts to a degree, but adapt poorly to conditioned reciprocity — a limitation that bears directly on how autonomous agents will behave alongside one another. The paper circulated in 2023 under its earlier title, “Investigating Emergent Goal-Like Behaviour in Large Language Models Using Experimental Economics”.
