The problem
Causal discovery methods need to reconcile statistical constraints, causal theory and expert knowledge. When those inputs conflict, conventional pipelines can obscure which evidence was retained or ignored.
The approach
The work combines causality theories with assumption-based argumentation (ABA). Candidate causal claims become arguments whose attacks expose incompatible evidence, while stable argumentative conclusions correspond to causal graphs.
What the work establishes
- Proves that the method can recover ground-truth causal graphs under natural conditions.
- Evaluates an Answer Set Programming implementation on four standard causal-discovery benchmarks.
- Shows competitive performance against established baselines while retaining a symbolic account of the supporting evidence.
Why it matters to CArLA
This is the methodological foundation of CArLA: causal structures are not presented as unexplained outputs, but as conclusions supported by inspectable and contestable arguments.
Citation
Russo, F., Rapberger, A., & Toni, F. (2024). Argumentative Causal Discovery. Proceedings of KR 2024, 938–949. https://doi.org/10.24963/kr.2024/88