Project GO-Y : Unifying Graph Databases and Causal Models.

Causality constitutes a fundamental abstraction for understanding numerous real-world phenomena. It occupies a central role in human intelligence and is increasingly recognized as a key component in artificial intelligence, particularly in the explanation and interpretation of decision-making processes. Cause–effect relationships and conditional probabilities form the backbone of structural causal models, which offer a concise formalism for representing the data-generating process among variables.

A Directed Acyclic Graph (DAG) associates a set of variables with a joint probability distribution. Such graphs admit a probabilistic interpretation, whereby each variable is independent of its non- descendants conditional on its direct parents. They also convey a causal interpretation: directed edges encode causal influences among variables. Causal reasoning, grounded in these structures, enables the derivation of interventional probabilities from observed conditional probabilities, without requiring additional experimentation. Interventional probabilities capture what would occur under active manipulation of a system, as opposed to passive observation of its natural behavior.

In parallel, graph-based representations have become central to large-scale data management systems, providing rich expressiveness and high computational efficiency. At present, however, the fields of causality and graph data management are advancing independently. Research in causality concentrates on analysis and inference from empirically validated graphs, often through purpose-built computational scripts, whereas graph data management focuses on querying and integrating data via declarative query languages.

This separation limits the potential synergies between the two domains. The ERC Advanced Grant project seeks to bridge this divide by exploring a novel research direction: causality-aware data management. The project aims to elevate causal relationships to first-class entities within graph databases, encoding both conditional and interventional probabilities through declarative operations. These operations will constitute the foundation of a rigorous framework for causal analysis, thereby opening new avenues for intelligent data management with broad implications for industry and multiple scientific disciplines.