The paper 'GENESIS: Towards Explainable Causal Discovery' was published on arXiv on 2026-08-04. It addresses two challenges in causal discovery from observational data: statistical methods lack power in low-sample regimes, and LLM-assisted hybrid methods lack transparency in edge decisions. The authors formalize 'decision traceability' requiring every inferred edge to be supported by auditable statistical evidence, Markov Blanket consistency, or explicit domain reasoning. They propose GENESIS, an explainable hybrid CD framework that decomposes graph construction into interpretable decision points, first identifying and scoring three-node structural motifs (chains, forks, colliders) to establish transparent structural priors.
A new research paper introduces GENESIS, an explainable hybrid causal discovery framework that makes every edge decision in a learned directed acyclic graph auditable by combining statistical evidence, Markov Blanket consistency, and domain reasoning. It addresses the opacity of LLM-assisted methods by decomposing graph construction into interpretable steps, starting with scoring of three-node structural motifs.
GENESIS introduces decision traceability as a formal requirement for causal discovery, ensuring each edge is justified by auditable evidence. The framework uses three-node structural motifs (chains, forks, colliders) to create transparent structural priors, then progressively refines the graph. This hybrid approach integrates statistical methods with LLM reasoning in a way that makes the influence of semantic reasoning on individual edge decisions explicit and verifiable.
In high-stakes domains like healthcare or finance where causal models inform decisions, the lack of explainability in current hybrid causal discovery methods limits adoption. GENESIS's decision traceability could enable regulatory compliance and trust by providing auditable justifications for each causal link, potentially accelerating deployment of causal AI in regulated industries.
Explainable causal discovery can reduce risk in applications where causal understanding drives critical decisions, such as drug discovery, supply chain optimization, and policy evaluation. By making causal models auditable, GENESIS could lower barriers to enterprise adoption and create opportunities for causal AI tools in compliance-heavy sectors.
If validated, GENESIS could become a standard for explainable causal discovery, influencing future hybrid methods to incorporate decision traceability. Next signals include empirical benchmarks comparing GENESIS against existing methods on real-world datasets, open-source code releases, and adoption by applied fields requiring auditable causal models.