Proof, Policy, and Probability: How DeepProofLog Rewrites the Rules of Reasoning
Proofs are supposed to be the respectable part of AI: tidy, inspectable, and resistant to the usual neural-network fog machine. Then reality turns up, as it so often does, carrying a bill. In neurosymbolic AI, the bill is search. A system may know the rules. It may even combine them with neural perception. But if answering a query requires enumerating a vast space of possible proofs, the promise of “interpretable reasoning” quickly becomes a very elegant way to run out of time. ...