Alexis Drogoul

Patrick Taillandier, Jean Daniel Zucker, Arnaud Grignard et al.

This position paper examines the use of Large Language Models (LLMs) in social simulation, analyzing both their potential and their limitations from a computational social science perspective. The first part reviews recent findings on the ability of LLMs to replicate key aspects of human cognition, including Theory of Mind reasoning and social inference, while also highlighting significant limitations such as cognitive biases, lack of true understanding, and inconsistencies in behavior. The second part surveys emerging applications of LLMs in multi-agent simulation frameworks, focusing on system architectures, scale, and validation strategies. Notable projects such as Generative Agents (Smallville) and AgentSociety are discussed in terms of their design choices, empirical grounding, and methodological innovations. Particular attention is given to the challenges of behavioral fidelity, calibration, and reproducibility in large-scale LLM-driven simulations. The final section distinguishes between contexts where LLMs, like other black-box systems, offer direct value-such as interactive simulations and serious games-and those where their use is more problematic, notably in explanatory or predictive modeling. The paper concludes by advocating for hybrid approaches that integrate LLMs into traditional agent-based modeling platforms (GAMA, Netlogo, etc), enabling modelers to combine the expressive flexibility of language-based reasoning with the transparency and analytical rigor of classical rule-based systems.

Patrick Taillandier, Cécile Duchêne, Alexis Drogoul

Many real world problems can be expressed as optimisation problems. Solving this kind of problems means to find, among all possible solutions, the one that maximises an evaluation function. One approach to solve this kind of problem is to use an informed search strategy. The principle of this kind of strategy is to use problem-specific knowledge beyond the definition of the problem itself to find solutions more efficiently than with an uninformed strategy. This kind of strategy demands to define problem-specific knowledge (heuristics). The efficiency and the effectiveness of systems based on it directly depend on the used knowledge quality. Unfortunately, acquiring and maintaining such knowledge can be fastidious. The objective of the work presented in this paper is to propose an automatic knowledge revision approach for systems based on an informed tree search strategy. Our approach consists in analysing the system execution logs and revising knowledge based on these logs by modelling the revision problem as a knowledge space exploration problem. We present an experiment we carried out in an application domain where informed search strategies are often used: cartographic generalisation.

Patrick Taillandier, Cécile Duchêne, Alexis Drogoul

Both humans and artificial systems frequently use trial and error methods to problem solving. In order to be effective, this type of strategy implies having high quality control knowledge to guide the quest for the optimal solution. Unfortunately, this control knowledge is rarely perfect. Moreover, in artificial systems-as in humans-self-evaluation of one's own knowledge is often difficult. Yet, this self-evaluation can be very useful to manage knowledge and to determine when to revise it. The objective of our work is to propose an automated approach to evaluate the quality of control knowledge in artificial systems based on a specific trial and error strategy, namely the informed tree search strategy. Our revision approach consists in analysing the system's execution logs, and in using the belief theory to evaluate the global quality of the knowledge. We present a real-world industrial application in the form of an experiment using this approach in the domain of cartographic generalisation. Thus far, the results of using our approach have been encouraging.

Patrick Taillandier, Alexis Drogoul

Nowadays, supervised learning is commonly used in many domains. Indeed, many works propose to learn new knowledge from examples that translate the expected behaviour of the considered system. A key issue of supervised learning concerns the description language used to represent the examples. In this paper, we propose a method to evaluate the feature set used to describe them. Our method is based on the computation of the consistency of the example base. We carried out a case study in the domain of geomatic in order to evaluate the sets of measures used to characterise geographic objects. The case study shows that our method allows to give relevant evaluations of measure sets.