Candidate spaces grow quickly, and verifying a real-world plan may itself be incomplete or costly.
A possibility worth exploring ●
Verifier-Driven AI Reasoning
AI systems that propose plans, proofs or programs and use deterministic, executable or formal verifiers to reject invalid candidates.
The possibility
AI systems that propose plans, proofs or programs and use deterministic, executable or formal verifiers to reject invalid candidates.
Verification can make generated reasoning more reliable even when discovery remains expensive.
Stronger general search or synthesis algorithms could expand planning horizons and improve completeness.
What to keep in mind
What stands between here and there?
Do not equate an efficient verifier with guaranteed efficient discovery, and do not claim that resolving P versus NP alone yields general AI reasoning.
P versus NP
P versus NP motivates the find-versus-check distinction, but practical reasoning systems involve richer complexity and representation issues.
Understand the question →Ideas it sparks
Ways to start thinking today.
These are prompts for curiosity—not roadmaps or promises.
Constraint-Aware Reasoning Copilot
Generate candidate plans, configurations or code, then validate them through executable checks, constraints or formal solvers before presentation.
Why it is thinkable now
Explore the idea →Yes for bounded domains with clear verifiers.
Sources and uncertainty 3
The Complexity of Theorem-Proving ProceduresStephen A. CookSource verifiedZ3: An Efficient SMT SolverLeonardo de Moura, Nikolaj BjørnerSource verifiedFormal Conjectures: An Open and Evolving Benchmark for Verified Discovery in MathematicsMoritz Firsching, Paul Lezeau, Salvatore Mercuri, Miklós Z. Horváth, Yaël Dillies, Calle Sönne, Eric Wieser, Fred Zhang, Thomas Hubert, Blaise Agüera y Arcas, Pushmeet KohliSource verified
AI summarized and source-backed · Qualified expert review pending.