Designing Artificial Reasoners for Communication

Project leader is IRIT.

Modern intelligent systems, notably large language models (LLM), are able to accomplish tasks that were considered to be out of the reach of machines some years ago. However, they still present several limitations on the ability to realize complex deductive tasks, such as moral reasoning and reasoning about human emotions. The absence of a theory of mind (ToM) yields an impossibility to represent its interlocutor’s mental state as well as beliefs, desires and intentions.

The aim of the DesiRes project is to equip LLM with ToM. The project is theoretical and also practical, going from the formal conception of algorithms to the implementation of solutions combining logical reasoning and LLM in conversational agents, as well as the evaluation of agents in their interactions with humans.


Scientific Responsible for CRIL :
Participants :
Duration :
2026-2029