Formal Explanations for Neuro-Symbolic AI

Neuro-symbolic AI combines neural and symbolic reasoning to address key weaknesses: The paper identifies that purely neural AI systems suffer from bias, brittleness, and poor performance on reasoning tasks—problems that neuro-symbolic approaches (combining neural perception with symbolic logic) can help mitigate.

A hierarchical explanation method makes complex AI systems more transparent: Rather than trying to explain an entire neuro-symbolic system at once, the proposed approach first explains the symbolic reasoning component, then focuses explanations only on the specific neural inputs that actually matter for that reasoning, making explanations more concise and understandable.

Formal explanations provide rigorous accountability: By using formal abductive explanations (a mathematical framework), the approach offers a systematic way to justify AI decisions rather than relying on informal or approximate explanations, which is important for high-stakes applications where stakeholders need to verify AI reasoning.

This work bridges a gap in AI explainability: While explainable AI (XAI) has advanced for purely neural systems, neuro-symbolic systems—which combine two different types of AI—create new explanation challenges; this paper directly addresses how to make these hybrid systems interpretable.

The method improves both transparency and performance: The hierarchical approach not only makes neuro-symbolic systems easier to explain but also increases computational efficiency by avoiding unnecessary explanations of irrelevant neural components.