The paper “Nicer Than Humans: How do Large Language Models Behave in the Prisoner's Dilemma?”, led by research assistant Nicoló Fontana and supervised by Prof. Francesco Pierri of the Data Science Lab at the Department of Electronics, Information and Bioengineering – Politecnico di Milano, in collaboration with Prof. Luca Maria Aiello from the IT University of Copenhagen, has been cited in the 2025 United Nations Human Development Report, “A matter of choice: People and possibilities in the age of AI”. The paper will also be presented at the 19th AAAI International Conference on Web and Social Media (ICWSM) that will be held in Copenhagen between June 23-26, 2025.
The research systematically investigates how Large Language Models (LLMs) behave as artificial social agents when faced with strategic dilemmas. By simulating Iterated Prisoner’s Dilemma games against adversaries of varying hostility, the study evaluates the cooperative tendencies of state-of-the-art open and closed LLMs. The results show that LLMs generally act at least as cooperatively as human players, displaying particularly forgiving and non-retaliatory behaviours when faced with low levels of opponent defection. The findings highlight notable differences in model behaviour and introduce a systematic methodology to assess LLMs in archetypal social scenarios, contributing to emerging efforts in LLM auditing and alignment.
