Automated theorem proving searches a formal space of definitions, lemmas and inference rules to construct a proof. Systems may use symbolic search, decision procedures, satisfiability solvers, heuristics or learned guidance to decide which steps to attempt.
AI can improve proof search by proposing useful lemmas and tactics. A generated proof should still be checked by a trusted proof system, separating the potentially complex search process from the smaller mechanism that verifies correctness.
ELI5
Automated theorem proving uses software to search for a formal sequence of logical steps that proves a statement. The system works from precise definitions and allowed reasoning rules rather than informal confidence.
For example, AI may suggest promising lemmas or proof steps for a difficult mathematical statement. A separate trusted proof checker can then verify every step, so a creative search process cannot make an invalid proof correct merely by sounding convincing.
