Sepehr Ilami

Nunzio Lore, Sepehr Ilami, Babak Heydari

As the performance of larger, newer Large Language Models continues to improve for strategic Theory of Mind (ToM) tasks, the demand for these state-of-the-art models increases commensurately. However, their deployment is costly both in terms of processing power and time. In this paper, we investigate the feasibility of creating smaller, highly-performing specialized algorithms by way of fine-tuning. To do this, we first present a large pre-trained model with 20 unique scenarios that combine different social contexts with games of varying social dilemmas, record its answers, and use them for Q&A fine-tuning on a smaller model of the same family. Our focus is on in-context game-theoretic decision-making, the same domain within which human interaction occurs and that requires both a theory of mind (or a semblance thereof) and an understanding of social dynamics. The smaller model is therefore trained not just on the answers provided, but also on the motivations provided by the larger model, which should contain advice and guidelines to navigate both strategic dilemmas and social cues. We find that the fine-tuned smaller language model consistently bridged the gap in performance between the smaller pre-trained version of the model and its larger relative and that its improvements extended in areas and contexts beyond the ones provided in the training examples, including on out-of-sample scenarios that include completely different game structures. On average for all games, through fine-tuning, the smaller model showed a 46% improvement measured as alignment towards the behavior of the larger model, with 100% representing indistinguishable behavior. When presented with out-of-sample social contexts and games, the fine-tuned model still displays remarkable levels of alignment, reaching an improvement of 18% and 28% respectively.

Qiliang Chen, Sepehr Ilami, Nunzio Lore et al.

Modern engineered systems increasingly involve complex sociotechnical environments where multiple agents, including humans and the emerging paradigm of agentic AI powered by large language models, must navigate social dilemmas that pit individual interests against collective welfare. As engineered systems evolve toward multi-agent architectures with autonomous LLM-based agents, traditional governance approaches using static rules or fixed network structures fail to address the dynamic uncertainties inherent in real-world operations. This paper presents a novel framework that integrates adaptive governance mechanisms directly into the design of sociotechnical systems through a unique separation of agent interaction networks from information flow networks. We introduce a system comprising strategic LLM-based system agents that engage in repeated interactions and a reinforcement learning-based governing agent that dynamically modulates information transparency. Unlike conventional approaches that require direct structural interventions or payoff modifications, our framework preserves agent autonomy while promoting cooperation through adaptive information governance. The governing agent learns to strategically adjust information disclosure at each timestep, determining what contextual or historical information each system agent can access. Experimental results demonstrate that this RL-based governance significantly enhances cooperation compared to static information-sharing baselines.