Is Big Tech pulling up the draw bridge and breaking history? (Part 4)

Mathematics

Similar to what may happen to cybersecurity, but more profound, is the potential loss of an area of significant human endeavour, the loss of the discipline of mathematics, and its replacement with a limited “AI”-labelled model version. The strongly related field of statistics, and central to the running of these models is also experiencing encroachment and replacement.

Mathematicians, as individuals and collectives, are increasingly expressing concern the use of “AI”-labelled models and similar is actually leading to a reduction in the potential of mathematics to solve problems. This is because, to use a cliche, it is the journey not the destination that is important. Along the way to one result, mathematicians can find other interesting and useful problems. Quickly appearing to solve a mathematical problem may make “AI”-labelled models look very intelligent but it may also curtail the discovery of new mathematical problems and directions (for example see Weinreich, Max, 2026, The crisis of AI-generated mathematics) and, hence, its benefits to society.

Mathematicians are a diverse bunch, like any group, and not everyone is concerned about “AI”-labelled models and similar but enough are and the “Leiden Declaration on Artificial Intelligence and Mathematics” published on 2/6/2026 to help protect the “fundamental values of the discipline.” The declaration was initially signed by a small but international group of research mathematicians with the support of the International Mathematical Union and has since garnered many thousands of signatures including Assistant Professor Weinreich. The assistant professor having recently completed his doctorate is, perhaps, in part, wondering whether his career is evaporating before his eyes due to computer automation.

Michael Harris, Columbia University, in his substack Silicon Reckoner, wrote about an OpenAI announcement regarding the Millennium Prize Navier Stokes problem “extremely damaging to mathematics; it convinces decision makers that human mathematicians are obsolete, and it convinces young people that their passion form mathematics has no future.” This was from a quote that was published in Science magazine.

The declaration encompasses both well-known potential threats and those specific to mathematics. Broader threats include: the still unresolved production of incorrect and unreliable “AI”-labelled model results; not giving credit to people whose published work may have been instrumental in leading to a result; using copyrighted data; and; giving false impressions to the general public by “overestimating the significance of automated tools” via, for example, company press releases. Concerns particular to mathematics include: the difficulty encountered when trying to check mathematical proofs as computer coded mathematical concepts can be hard to relate to human ones, effectively restricting the working out of answers to those able to run commercial models; and; “automated mathematics” may distort the setting of priorities.

The declaration states mathematics is a “human practice” and transparency and accessibility, amongst other values, are important to its conduct.

The recent announcements about “solving” mathematical problems by Big Tech companies, particularly OpenAI, but Google and Anthropic as well, have lead to consternation as accessibility to the model to verify results, the role of humans and/or transparency is often lacking.

OpenAI has been tackling long-standing “Millennium Prize Problems.” These are problems mathematicians found challenging to resolve at the start of the 21st century and are reportedly seen as another, but better, form of benchmarking. OpenAI does this on an internal model that mathematicians outside of the company can not check. In October, 2026, OpenAI, despite the well-published mathematician concerns, went ahead and announced the “sharing” of results of a large number of mathematical problems via GitHub and still from an internal model. It seems OpenAI can’t release the model used to produce these results until the model can be used to “empower scientists” and to reach that level it must keep on internally solving mathematical problems (and other science problems) it wrote on its website on the 6th October 2026.

Reporting of these results lacks details about human involvement such as: a mathematician may have directed the model processes using it somewhat as “a calculator”; the model unintelligently produced reams of results a mathematician needed to search through to find a potential solution; or; the result might have already been in its training data because of use of a mathematician also trying to solve the same problem. A model appears to have arrived at some potential solutions but it is uncertain as the result may have been scanned from mathematical research and put into the training data. Citations are not included in results, hence, leading to a lack of transparency and lack of clarity of its real capability in this area.

There is one thing producing a solution but there is another thing verifying it is correct. There are established ways of doing this but more recently digitised mathematic proofs have become available. Proofs are a way of checking the proposed solution is mathematically sound. When OpenAI believed it had a solution for the Millennium Prize Navier Stokes problem they ran it through the proof checker program Lean. Lean can be found in GitHub. Anthropic produced a proof for an already solved Millennium problem in the Lean programming language also using an unreleased model. Lean is an open source formal maths checker developed by Microsoft in 2013. The original developer is now with Amazon but started a Lean non-profit with another Lean developer that they run after hours. Microsoft sees Lean as “democratising” mathematics which is similar to the “share economy” justification used for UBER – anyone can use it not just mathematicians (and remember UBER is now using “driverless (limited human involvement)” taxis). A problem with Lean is that it can help check if the maths used to solve a problem is correct based on what the user entered but not whether the problem is actually solved.

Big Tech’s mathematical vision appears to be an “AI”-labelled model automatically produces a solution that will almost instantaneously be verified by an automated proof checker.

Tasmin Chu, PhD student at Caltech, USA, wrote in an essay published on her blog that Big Tech AI-labelled companies “misrepresent our fields, the goals of our field, and the scientific training which is core to our work; they devalue key aspects of our labour and present mathematics as a theorem-producing factory. In their campaign of scientific misinformation, they presume to tell the public and ourselves [mathematicians] what the real goals of research mathematics are.”

The Future

We may face a loss of more than jobs if we continue with a, largely, wholesale embrace of Big Tech “AI”-labelled models. We face a world depauperate of the knowledge we have now and, what there is, siloed into summaries – a technique these companies appear to prefer. We may lose the words, the maths, the skills and experience of other ways of doing things and the requisite tools and equipment may be difficult to find or no longer made. The world we live in will be using the words, maths and the tools behind the closed drawbridge of the castle of Big Tech “AI”-labelled models. Finding alternative pathways will be hampered by this but also this loss of history in these areas and the replacement by the history Big Tech computes and Big Tech stores; we won’t even know what we are missing. The higher fixed level we strove to reach to protect ourselves and the environment, a forgotten thing of the past. The problem solving process in maths leading to a new avenue of thought – gone. The ability to program an alternative digital system (if we wanted to) – lost. These areas of human endeavour were built on the sweat and tears of many people often paid but also those who did it all for free.

History Professor Emiritis Odlyzko thought the Big Tech “AI”-labelled models would be useful for historical research. Hopefully, he won’t so much now once he sees the impact it’s already having on mathematics and other areas. Why would any one involved in keeping of human history, from a university, librarian, newspaper or publisher, keep using or embed these “AI”-labelled models within their own private business or publicly-owned site? They could be the last bastions of human history.