Neurosymbolic AI

An approach that combines neural components, which are strong at perception, language, and statistical patterns, with symbolic representations, rules, or formal inference mechanisms. The division of labor is central: the learned component can extract or propose hypotheses; the symbolic component can impose constraints, query structured facts, or verify relationships.

Neurosymbolic is not a seal of explainability. If the model extracts incorrect entities or rules, the symbolic layer may merely propagate the error with an appearance of rigor. The architecture needs to state which decisions depend on rules, which depend on estimation, and how conflicts between the two sources are resolved.