Will a generative AI model (e.g., a large language model or other generative AI system) find and produce a successful resolution to the P versus NP problem by August 16, 2031?

probability: 3%

rationale:

Even with a five-year horizon from August 2026, resolving this question requires clearing a severe conjunction: discovering a breakthrough proof, documenting a central generative AI role, and securing mainstream mathematical community acceptance by August 16, 2031. While AI capabilities have inflected recently, evidence suggests that P versus NP remains fundamentally open and lacks any viable strategic outline that is not already blocked by established theoretical barriers.

Generative AI systems have undeniably moved from olympiad-level formalization to genuine research-level results. In 2026, an internal OpenAI model successfully generated a proof for the 80-year-old Erdős unit-distance conjecture quantamagazine.org. Shortly after, advanced models produced Lean certificates for core theoretical computer science results, including arithmetic circuit lower bounds for the permanent openai.com. These breakthroughs confirm that AI can now play a verifiable, central role in significant mathematical discoveries, satisfying the authorship criteria if a P versus NP proof were found.

However, a massive chasm exists between current AI achievements and Millennium-tier problems blog.computationalcomplexity.org. AI currently excels at the "attention-starved long tail" of mathematics rather than marquee challenges teorth.github.io. The well-known structural barriers to P versus NP—specifically relativization, natural proofs, and algebrization—mean that scaling up search over known proof patterns is insufficient; a genuinely new conceptual paradigm is required 2 sources. Heavily resourced, multi-model AI attempts targeting the problem have so far yielded only unproven "proof skeletons" rather than conceptual breakthroughs cacm.acm.org.

Market indicators and rigorous validation timelines further suppress the likelihood of a resolution by 2031. While aggregate prediction markets price a high probability of AI solving some Millennium Prize problem within this window, P versus NP is consistently ranked as the least likely candidate, heavily trailing problems like Navier-Stokes manifold.markets. Furthermore, mainstream validation acts as a severe binding constraint. The Clay Mathematics Institute requires at least a two-year waiting period post-publication claymath.org, and any claim resolving P versus NP would attract years of intense scrutiny before achieving general acceptance 2 sources.

The primary path to a resolution within this timeframe relies on near-discontinuous capability growth—such as AGI-level research systems arriving by 2028–2029 that can compress decades of complexity theory thezvi.substack.com. Alternatively, an explicit P=NP algorithm could theoretically short-circuit the lengthy validation process required for a lower bound. However, balancing these tail scenarios against the severe structural obstacles, lack of a viable attack surface, and the extraordinarily slow pace of mathematical consensus, the compounding hurdles firmly anchor the probability at 3%.