Causal Learning and Reasoning in Multi-Agent Systems: Goals, Issues, and Taxonomy

Stefano Mariani, Giovanni Briglia, André Meyer-Vitali, Matthias Klusch, Franco Zambonelli

Keywords: Causal learning, Causal reasoning, AI agents, Multi-agent system, Taxonomy

TL;DR: Causal models have been widely used to improve decision making in AI agents, mostly considering single-agent systems. In this paper we turn attention to the peculiarities of using causal models in multi-agent systems.

Abstract:

*Causal reasoning* is a necessary prerequisite of *agency*, that is, the capability to act purposefully towards goals.
Accordingly, *causal models* became a fundamental research topic for Artificial Intelligence (AI) agents and agentic AI systems, lately, as they provide for mathematically sound approaches to support correctness, adaptability, robustness, transparency, trustworthiness, and accountability of agents' decision making.
However, most of the research agenda is focussed on *single-agent* systems, where the causal model either captures the agent inner reasoning, or its interactions with the operational environment---while assuming the single agent to be the only source of actions.
Instead, the role of causal reasoning and learning in *Multi-Agent Systems* (MAS), i.e. to support both situated (agents-to-environment) and social interactions (agents-to-agents), is still under-explored.
In this paper, we motivate that causal reasoning and learning in MAS is not a simple extension of the single-agent case, shed light on the many reasons why causal models -- and especially *explicit* structural causal models -- are necessary in MAS, propose a taxonomy to classify existing (and future) approaches to learn and exploit causal models for reasoning in MAS, and propose a MAS-centred research agenda in terms of open challenges.

Paper Type: Full (minimum of 10 pages and a maximum of 16 excluding references)

Poster Opt In: Yes, I'm open to having my submission accepted as a poster (leave blank if you are submitting a poster, or if you DON'T want your submission to be accepted as a poster instead of a full or short paper)

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 10