Relational Structural Causal Models

Researchers have extended Pearl's structural causal models to settings where objects and their relations vary, addressing a long-standing gap between causal reasoning and combinatorial generalization in AI. The work derives symbolic identification criteria — including under unobserved confounding — and introduces relational neural causal models that outperform non-relational baselines on simulated multi-agent traffic scenes. This is foundational work: the ability to reason causally about novel combinations of objects is a precondition for AI systems that behave reliably in open-ended environments.