Why mathematician Terence Tao thinks AI should spark a fast revolution 

Because the stewards of a physique of labor courting again 1000’s of years, mathematicians are not often in a rush. Issues can lie open for many years and even centuries earlier than they’re cracked, and particular person mathematicians suppose nothing of spending years dedicated to a single query. However now Terence Tao, arguably the world’s biggest dwelling mathematician, says that he and his colleagues have mere months to react to a disaster pushed by the rise of AI. He warns that, for the primary time in additional than a century, mathematicians are dealing with a peril so extreme that they ought to return collectively and reimagine what it means to be a mathematician.

“We don’t should passively settle for adjustments by exterior forces. [We] can’t simply passively show our theorems; we now have to organise, change into activists, get a little bit political,” Tao tells me once we sit down to speak on the Worldwide Congress of Mathematicians in Philadelphia. In his estimation, arithmetic wants an enormous overhaul of its tradition and practices, and it wants it very quickly – by the top of this yr. “I want we had extra time to do that slowly,” he says.

His sense of urgency relies on the gorgeous progress that AI models have made in mathematics this year alone. All through 2026, mathematical conundrums that had gone unanswered for many years are seemingly falling by the hands of AI at a tempo of some every week. Remarkably, a few of these efforts have been helmed by amateurs and hobbyists who’ve merely requested fashions for options. Tao says it’s probably that people are about to lose their monopoly on downside fixing.

Within the face of this upheaval, Tao is maybe uniquely positioned to guide a rethink of what it means to be a mathematician. Rising up, he struggled to reply that query. He cherished arithmetic – actually, he was a veritable baby prodigy – however he had no concept what a mathematician does all day. Does some shadowy council simply hand open issues to every mathematician like homework? Possibly, Tao reasoned, he ought to as an alternative simply be a shopkeeper – his love of arithmetic would make him good at balancing accounts and taking stock.

However it shortly grew to become obvious that Tao was destined for a mathematical life. In a single oft-cited anecdote, he was instructing different kids to rely at age 2. In one other, a 7-year-old Tao is recounted studying a calculus textbook. At age 10, he grew to become the youngest winner of a medal on the Worldwide Mathematical Olympiad. In 2006, he won the Fields medal, usually referred to as the Nobel prize of arithmetic, and the award quotation listed 4 completely different branches of arithmetic he had influenced profoundly, having solely simply entered his 30s.

Now, aged 51, Tao has made the kind of singular mark on arithmetic that may take most within the discipline a number of lifetimes to attain. He has tackled a centuries-old query about patterns inside prime numbers. He additionally chipped away on the Kakeya conjecture, which asks in regards to the path taken by a needle because it rotates by area. Fixing a model of this downside simply earned Hong Wang at New York University the Fields medal. Tao has additionally settled seven issues set by legendary mathematician Paul Erdős, whom he met at age 10 (pictured beneath), and contributed to among the largest unsolved issues in arithmetic, just like the Collatz conjecture and the Riemann speculation.

Then there’s his work associated to physics, corresponding to wave maps, that are linked to Albert Einstein’s concept of normal relativity, or the nonlinear Schrödinger equation, which describes gentle that interacts with itself inside fibre optics. He even developed an algorithm to make MRI scanners extra environment friendly.

As this checklist demonstrates, Tao is each prolific and versatile, and he’s ceaselessly on the innovative of recent methods of doing arithmetic. For instance, he helped lead the Polymath Project, an experimental enterprise in massively parallelised arithmetic the place giant groups of volunteers tackled an open downside by splitting it into many mini issues. He isn’t in opposition to AI, saying he makes use of it frequently – however he’s nonetheless remarkably anxious about what the fast tempo of development means.

To outsiders, this ongoing shift from proof shortage to proof abundance, as Tao describes it, would possibly seem to be a great factor. However Tao says it’s as if mathematicians have been driving their automobiles alongside outdated roads constructed for pedestrians and horses, however now AI has joined in whereas driving bigger, quicker automobiles. Visitors jams and accidents are all however inevitable, and the entire infrastructure is proving to be unstable – AI is revealing and magnifying all its potholes, bumps and cracks. For instance, it was that producing proofs was arduous, however now that we now have massively accelerated that a part of the method with AI, the conventional channels of peer assessment and mathematical acceptance have gotten clogged. “Validation is briefly provide,” warns Tao.

There are some options already. Historically, mathematicians write up their proofs in a mix of mathematical symbols and pure language, and their colleagues learn these papers to observe the argument and confirm its logical reality. Extra not too long ago, some mathematicians, together with Tao, have turned to formalisation, by which the mathematical argument is encoded using a programming language called Lean, permitting a pc to confirm every step. This has now been turbocharged by AI with autoformalisation, which arms the duty of writing a proof in Lean over to AI.

However Tao worries that permitting AI to each produce and formalise a proof might simply lead us into territory dangerously near meaninglessness. We would find yourself with one thing that a pc says is appropriate, however if a human can’t understand the proof, what is the point? Is it even actually a proof? In Tao’s view, it could be dangerous for arithmetic if all that the longer term held was simply increasingly more AI-generated proofs. “There’s a hazard that if all science turns into automated, the following technology of extremely curated data will probably be misplaced,” he says.

The final time arithmetic was in severe disaster was the early 1900s, when a variety of paradoxical results threatened to break the very foundations of the field. Mathematicians and philosophers needed to come collectively and rebuild on firmer logical floor. “That was traumatic,” Tao says candidly. This AI disaster is completely different – not certainly one of mathematical arguments, however reasonably values and practices. Till now, mathematicians have largely been capable of put philosophy again in its field after the disaster of the 1900s, however AI is opening wounds previous and new.

All of it goes again to Tao’s query of what precisely a mathematician does. Throughout their coaching, mathematicians change into extraordinarily proficient in technical expertise, he says, however how arithmetic works, and why it really works, is one thing that they solely study implicitly from speaking to their mentors and friends.

As they progress, younger mathematicians tacitly choose up on how finest to current their work, find out how to train their method to others, when to belief another person’s work and what it means to have good style and discernment when deciding which issues and analysis matters to work on. With a wry smile, Tao jokes that he needs AI fashions might spend a while in graduate faculty and study all of this earlier than being unleashed onto some open downside. “Lots of people which might be selecting up AI are treating maths as a sport the place you simply acquire tick marks,” he says.

This aggressive spirit could also be impressively productive on the floor, but it surely doesn’t essentially advance arithmetic. In a well-attended lecture on the convention, Tao laid out what he sees as the actual work of mathematicians. He targeted on downside fixing. In his framework, step one is posing an issue, adopted by resolving it with a proof. The answer then must be verified, but it surely doesn’t change into significant until it’s also effectively written, he argued. Such proofs should be simple to know (for an professional, no less than), participating to learn and clearly lay out the modern elements of the argument and its connections to current literature.

However this isn’t the final step, stated Tao. The method ends solely as soon as the well-written, verified answer has been digested and accepted by a group of specialists and turns into built-in into how the following technology of specialists is educated. “So as to really affect the longer term growth of arithmetic, it must be accepted and valued by [the] group. Different mathematicians want to truly wish to learn it and digest it and put it into their very own work,” he stated to a captivated viewers. Heads nodded throughout the room.

AI fashions are superb at producing proofs and proficient at verifying them, however AI arithmetic is basically irritating to learn, Tao continued. Generally, an AI mannequin will spend pages proving one thing that’s apparent to a human mathematician, however then compresses the toughest a part of the proof into only a few sentences, glossing over probably the most significant elements. Worse but, the best way AI fashions arrive at options is usually opaque, hidden by each the privately run nature of software program, like ChatGPT and Claude, and the truth that tech-transparency-is-key/”>we nonetheless don’t absolutely perceive their underlying know-how.

Lastly, group acceptance is just not one thing that may be optimised with AI instruments alone, he stated. If being a mathematician merely means producing proofs, then AI fashions stand an opportunity of being good mathematicians, however Tao’s evaluation, partially primarily based on his personal expertise as a remarkably versatile and prolific mathematician, clearly made the case {that a} mathematician is one thing far more.

One space that may want cautious consideration is the schooling of recent mathematicians, with Tao saying in his lecture that college students will must be restricted of their use of AI, permitting them to develop their very own instinct by old school elbow grease. Tao tells me that he has already heard tales of scholars whose in any other case commendable ability actually declined after they began overly counting on AI. “Sooner or later, it turns into irreversible,” he says. To keep away from this, he can already provide some suggestions. “You need to use AI in the event you can current your work to your advisors and reply questions intelligently with out your cellphone.”

I ask him whether or not he expects resistance from mathematicians with regards to implementing such adjustments, in addition to doing analytical and self-reflective work about their occupation that they aren’t used to. They may merely should, he tells me with out hesitation. Then, a small concession to simply how tough cultural change may be: “Denial is such a robust pressure, particularly if you must change the whole lot.” However the various can also be clear to him – not solely will arithmetic cease advancing, however mathematicians will lose the authority over what their occupation is, and the determine of the mathematician will as an alternative be outlined by know-how firms.

In truth, Tao thinks that is already taking place. “Till not too long ago, solely mathematicians cared about arithmetic. We had full management,” he says. “Now, for the primary time, mathematicians have quickly misplaced the narrative.” In a single placing instance, Jacob Tsimerman on the College of Toronto in Canada, who was one other winner of the 2026 Fields medal, is going on hiatus from mathematics to join OpenAI as a researcher.

And whereas it’s simple that AI fashions are fixing mathematical issues, we’re removed from having a scientific understanding of what it takes to attain these outcomes, and even one thing so simple as what number of makes an attempt it takes a mannequin to spit out a proof – one thing that firms like OpenAI or Anthropic don’t essentially have an curiosity in clarifying. “We’ve kind of selectively disclosed outcomes, which look spectacular, however we don’t know precisely what the prompts have been. We don’t know the failure charge. And most of the firms which might be disclosing these have their very own incentives to possibly current the leads to as beneficial [a] gentle as doable,” Tao stated in his lecture.

He needs that mathematicians leveraged their authority extra and spoke up extra usually. The general public already doesn’t perceive what mathematicians do, and now there’s all this noise about it coming from a really completely different crop of researchers and technologists. In Tao’s view, tech firms could possibly shift the main target to easily producing as many proofs as doable, however mathematicians should get louder about all of the elements of the method of doing arithmetic the place these companies’ AI instruments can’t compete. “They tried to redefine what our occupation is,” he says. “[But] mathematicians have extra affect than they suppose.”

Moreover, Tao is actively concerned in efforts to guage the mathematical prowess of AI in a extra scientific manner. He’s a part of the First Proof project, which goals to check AI on unpublished issues in arithmetic, which means the fashions can’t depend on solutions discovered elsewhere. Earlier this yr, after posing 10 issues, the undertaking’s reviewers discovered seven options generated by AI fashions to be of a high-enough high quality to be revealed in arithmetic journals. “We have to establish the issues that basically are finest fitted to AI,” says Tao.

Even as soon as AI fashions’ competency has been completely evaluated by efforts like First Proof, questions will stay about one of the simplest ways to leverage them. It may be that, as an alternative of fixing a couple of very arduous issues in opaque methods, AI instruments might deal with 1000’s of barely much less arduous issues and advance arithmetic by a extra cumulative impact, Tao suggests, echoing a few of his previous group tasks that had the same goal. This might usher in an period of extra experimental arithmetic, possibly even mirroring how massive worldwide collaborations, as an example surrounding particle colliders, have superior physics. “If we get away from our obsession with fixing [famous] open issues and chasing status, these instruments may be helpful,” says Tao.

He’s chipping away on the tradition challenge, too. The form of a mathematical paper hasn’t actually modified for many years, however Tao has simply launched Mathematical Discourse, a video-based journal the place mathematicians’ contributions will probably be evaluated not simply on correctness of proofs, however how effectively they will clarify them to their friends.

But when one of many classes of the disaster within the foundations of arithmetic a century in the past was that mathematicians should come collectively, that can also be taking place now. Tao factors me to the Leiden Declaration on Artificial Intelligence and Mathematics, a group initiative that produced a collection of suggestions that have been then endorsed by the Worldwide Mathematical Union and, on the time of writing, has greater than 3000 signatories throughout all ranges of mathematical analysis.

“Arithmetic produces not solely a physique of outcomes, but in addition understanding, readability, and judgment among the many communities of mathematicians who’ve formed them,” the authors of the declaration state in its introduction. “This has been the results of months of group enter in regards to the elementary values and targets of the mathematical group. On reflection, these have been questions we must always have been systematically discussing years in the past,” Tao commented when including his signature.  

Arithmetic, then, stands at a massively vital second, one the place it should cope with a century’s value of historical past as it’s shortly barrelling into the longer term – a future that, with out motion, might be determined by know-how firms reasonably than mathematicians. For Tao, this wrestle for the guts of arithmetic is so simple as it’s private. “I like my occupation,” he says. “I believe that is value preventing for.”