{"slug": "the-debate-over-math-and-ai-why-mathematicians-are-still-needed-and-valued", "title": "The debate over math and AI–why mathematicians are still needed and valued", "summary": "Mathematicians remain essential to AI-assisted mathematics because AI-generated proofs require expert validation, according to a commentary by Grey Enlightenment. The author argues that AI cannot determine proof correctness and that human mathematicians are the final arbiters, citing an endless cycle of AI-generated proofs being critiqued by other AI models. The piece counters the notion that AI will render mathematicians obsolete, emphasizing that serious mathematical training is still required for prompt crafting and output evaluation.", "body_md": "I saw these articles going viral about the “AI crisis in math,” those being “[Mathematicians need to act](https://tasmin.substack.com/p/mathematicians-need-to-act)” and “[The Dark Night of Mathematics](https://kirwinhampshire.substack.com/p/the-dark-night-of-mathematics).”\n\nIf the math career doesn’t work out or if AI takes over, they certainly have a career in writing. This also agrees with the literature that verbal IQ and math/quantitative IQ are highly correlated, as also indicated by SAT and GRE scores. It’s very uncommon for someone to get a near-ceiling math score and a mediocre verbal score. That correlation is also reflected in writing ability, which these authors clearly possess in abundance.\n\nThe second author concedes towards the end of his massive post, that AI still needs mathematicians:\n\nMoreover, these AI companies need us. They need us to validate their proofs; to certify their write-ups; to use their free subscriptions; to work for their companies. They need us to imbue their products with soft power and cultural prestige. We are just one of many industries that they are prepared to disrupt. I am sure that many users are out there prompting an LLM: “Do a breakthrough! Do a breakthrough!” This strategy is unlikely to lead to success without serious mathematical training; indeed, even if a breakthrough resulted, the reader is unlikely to be able to understand the proof. As of this moment, mathematicians are still required for the production of AI-assisted mathematics.\n\nSo I’m not even sure what the problem is? He says unequivocally that AI is no substitute for serious mathematical training, so this runs counter to the popular notion of “AI rendering mathematicians obsolete.” This has been my [argument](https://greyenlightenment.com/2026/06/27/ai-is-not-going-to-make-mathematicians-obsolete/) from day one, so there is no disagreement here. Mathematicians are needed for crafting prompts and evaluating outputs for correctness. Knowing what to input or how to interpret the output requires specialized math knowledge. This is why, AFIK, no laypeople have made contributions to math with AI, and it’s reasonable to assume none will for the foreseeable future.\n\nWhat makes math uniquely hard to fully automate with AI is it has to be 100-percent correct, especially for proposed solutions for important problems. A considerable amount of vetting by humans goes into checking AI-generated proofs. As I said, AI cannot tall you if a proof is correct.\n\nAI can generate proofs that, on the surface, appear convincing but contain subtle mistakes that only a trained expert would catch. Of course, AI has also succeeded in proving many difficult problems, but it can fail in spectacular ways. Using AI to solve mathematics is like hiring a surgeon with perfectly steady hands who is also liable to remove the wrong organ or operate on the wrong side of the body. It is extraordinarily capable, but it also requires considerable supervision. Due to survivorship bias, you don’t see all the outright incorrect proofs or near-misses. You’re not going to see headlines like, “AI Generates a Near-Proof of a Famous Problem.”\n\nI can confirm this from my own experience. I can generate a proof with a model like Claude, then ask ChatGPT to evaluate it, which often finds errors. I send that critique back to Claude, which revises the proof. I then paste the revised version back into ChatGPT, which identifies new issues. The process never ends. It becomes an endless cycle of revisions and fresh objections. Who is right? Ultimately, humans remain the final arbiter or “consensus mechanism” for the correctness of proofs. A proof is considered correct because expert mathematicians judge it to be correct, not because an AI declares it so.\n\n[Formalization is often touted as the solution to the limitations of human proof checking, but it introduces bigger complications. Formalizing a mathematical proof is like rewriting the entire argument in a specialized language, making the process many orders of magnitude more difficult and far more error-prone.]\n\nAI is a major time saver. I can give it some ideas and have it check/validate them or to run calculations. I want to know as soon as possible if something does or doesn’t work, ideally without having to do all the calculations myself. AI saves time here, which is great. Does this diminish some of the sense or awe of discovery that comes from finding results on one’s own? Perhaps. But if it means producing more results and quickly pruning bad ones, I think it’s a worthwhile tradeoff.\n\nThat said, you have to push back when AI goes over its skis. Accepting its outputs blindly will inevitably lead to mistakes. There have been times when it generated proofs or made claims that, upon closer inspection, contained gaps or relied on unsubstantiated assertions. When I challenged those claims, it conceded that it had jumped to conclusions or that the proof was circular.\n\nA chat console is a forgiving environment, but in the setting of peer review, such mistakes can lead to instant rejection of the paper. Every step of a proof must be justified, even if the details are not spelled out explicitly. Referees expect every assertion to be supported by a valid argument, not merely to sound plausible. Many are in for a rude awakening when their AI-generated proofs are met not with viral praise on Twitter, but are ignominiously rejected without even being closely reviewed. At some point this means having to actually learn math to diagnose what went wrong.\n\nDoes this contradict my earlier posts arguing that AI progress in writing has largely stalled compared to progress in mathematics? Not really. AI gives mathematicians far more powerful tools, but it also introduces more failure modes if one is not careful. Writing, by contrast, has much less at stake because correctness is not an absolute requirement.\n\nClunky or formulaic writing is usually not fatal, because readers’ expectations for correctness are much less demanding to begin with vs math, and even clunky writing can still convey the author’s intended message. But a wrong math result can invalidate an entire proof at the ground level, as later results build on earlier ones, so there is much less room for error. Or in other words, math mistakes can lead to a cascade failure, whereas isolated grammatical mistakes for writing doesn’t (usually) invalidate the entire piece.\n\nRegarding careers, mathematicians will still come out on top. Some of the commentary about “mathematics being in crisis” seems less like an objective forecast, and more like a projection of the commentators’ own desire to see mathematicians struggle, especially given that they have long enjoyed strong career prospects in the private sector, such as technology or finance.\n\nNow, similar to the arguments about AI replacing white-collar jobs (a prediction that, years later, has also failed to come true), mathematicians are next on the chopping block. But mathematical knowledge is highly specialized and technical, and hence the underlying skills transfer well to finance and technology. Even if AI replaces many mathematicians, the analytical abilities they develop remain valuable, and those skills are still highly sought after by many high-paying employers.", "url": "https://wpnews.pro/news/the-debate-over-math-and-ai-why-mathematicians-are-still-needed-and-valued", "canonical_source": "https://greyenlightenment.com/2026/08/05/the-debate-over-math-and-ai-why-mathematicians-are-still-needed-and-valued/", "published_at": "2026-08-05 20:23:36+00:00", "updated_at": "2026-08-09 13:47:47.447323+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-research", "ai-ethics"], "entities": ["Grey Enlightenment", "Claude", "ChatGPT"], "alternates": {"html": "https://wpnews.pro/news/the-debate-over-math-and-ai-why-mathematicians-are-still-needed-and-valued", "markdown": "https://wpnews.pro/news/the-debate-over-math-and-ai-why-mathematicians-are-still-needed-and-valued.md", "text": "https://wpnews.pro/news/the-debate-over-math-and-ai-why-mathematicians-are-still-needed-and-valued.txt", "jsonld": "https://wpnews.pro/news/the-debate-over-math-and-ai-why-mathematicians-are-still-needed-and-valued.jsonld"}}