Collaborative Causal Discovery in a Synthetic Environment: Empirical Evaluation of FLODO

André Meyer-Vitali, Matteo Gregorini

Keywords: Causal Discovery, Collaborative Planning, Agentic AI

TL;DR: Collaborative causal discovery empirical analysis in a synthetic environment

Abstract: Causal discovery aims to uncover the underlying cause-and-effect relationships from observational and interventional data, a task historically dominated by centralised statistical methods. However, traditional centralized approaches are often impractical in real-world applications, as the inherently distributed nature of modern data architectures imposes strict privacy constraints that prohibit the pooling required by single-agent systems. In this paper, we empirically test FLODO (Flock of Dodos), a collaborative multi-agent algorithm designed for decentralised causal structure learning. Inspired by distributed problem-solving, FLODO deploys a ``flock'' of autonomous agents, where each agent independently explores subsets of variables using the causal discovery algorithm DODO and proposes local structural priors. Through a consensus-driven negotiation protocol, agents debate the global causal structure, merging their localised findings into a globally consistent Directed Acyclic Graph (DAG). We evaluate FLODO in a rigorous synthetic setting, systematically varying node counts, edge densities, and noise distributions. Our experimental results highlight the difference in performance between various scenarios, and we speculate on the areas of improvement of the collaborative protocol.

Paper Type: Short (exactly 6 pages excluding references)

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: 5