3 ms·
What I would want to see is a project for automatically converting the math literature to machine verifiable proofs. It would have to do NL to formal notation
by pfdietz 4y ago
What I would want to see is a project for automatically converting the math literature to machine verifiable proofs. It would have to do NL to formal notation translation, as well as filling in the gaps that are left in any math paper.
This is a difficult problem, but if solved it would enable every trick used in every published proof to be internalized into the theorem proving tooling.
- zozbot234 4y agoSomething more likely to be achievable is (semi-automatically) converting the literature to formal proof sketches, leaving some steps unfilled. Then the option of whether to accept such proofs as they are or try to make them totally rigorous can just be addressed on a case-by-case basis. Unfortunately, modern proof assistants like Lean and Coq make this a bit harder than it could be because they're based on procedural proof scripts or complete proof terms, and a proof sketch is closer to a partial declarative proof. Since declarative proofs write down every intermediate goal, the notion of skipping some steps arises very naturally.