In the latest “huge if true” news to come out of AI’s ongoing conquest of mathematics, earlier today American mathematician and Anthropic researcher Levent Alpöge posted a claim on his personal website that is already reverberating around the globe.
“I believe I have, working with Claude, discovered a proof of the Riemann Hypothesis,” Alpöge wrote. He also included a Claude-generated proof written in Lean, a programming language used to confirm that the supporting math is logically consistent and correct; other mathematicians are racing to independently verify Alpöge’s work, with preliminary and informal responses already emerging that the proof is indeed correct.
But what would this actually mean for the rest of us? Would proof of the Riemann hypothesis make our computers faster, our online transactions more or less secure or our digital lives fundamentally different? The answers are murkier than you might expect.
On supporting science journalism
If you’re enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
Proposed some 167 years ago by mathematician Bernhard Riemann, the Riemann hypothesis has been widely considered one of the greatest unsolved mysteries of math. It concerns the distribution of prime numbers (numbers divisible only by themselves and 1) and an unproven—apparently until today—assumption about the almost mystical layers of numerical order that underpin them. The question of its veracity is also one of seven “Millennium Prize Problems” posed in 2000 by the Clay Mathematics Institute—resolving it would net a $1-million prize.
But the biggest implications for a solution won’t be monetary. Alpöge’s proof is technically for the “generalized Riemann hypothesis,” or GRH, which is a bigger, broader and more powerful application of Riemann’s initial insight to a much wider class of mathematical functions related to the primes. Proving it true, as my colleague Joseph Howlett explains, would cascade through all of math, in a flash revealing that many of the elaborate interlocking assumptions built on its presumed bedrock veracity are, in fact, true. Algorithms used in arcane realms like computational number theory and the study of algebraic number fields could see immediate improvements. These gains could even, eventually, lead to potential benefits for practical applications, but such advances would be invisible to most people.
Many have wondered how this could affect cryptography. Prime numbers are essential to RSA, a widely used public-key cryptographic scheme whose security depends on the difficulty of factoring enormous numbers into their original prime factors. Given this technology, you might think that proving GRH and being able to precisely map patterns of primes would threaten the security of online banking, encrypted communications and other digital infrastructure.
But it wouldn’t—not directly. Determining whether a number is prime—an obvious application of GRH—isn’t the same as finding which primes serve as factors for the gargantuanly unwieldy numbers underpinning RSA and other encryption systems. GRH would improve mathematicians’ understanding of the former but wouldn’t provide a known shortcut for the latter. So, for now, your encrypted financial transactions and online secrets are presumably safe from direct cracking via some GRH-derived method.
In fact, the most consequential impacts of Alpöge’s claim (if verified) wouldn’t really concern prime numbers—or math—at all. Instead they would signal AI’s emerging capacity for profoundly superhuman scientific discoveries.
For many experts, this milestone isn’t necessarily something to celebrate. As one prominent mathematician ominously told Howlett for a different story earlier this year, an AI that can prove the Riemann hypothesis would be something so powerful and potentially disruptive that he’d prefer to never meet it.
That’s because the problem, even in its nongeneralized form, is so ferociously complex that it’s resisted more than 150 years of effort to solve it by some of humanity’s best and brightest minds. “Proving” GRH doesn’t seem to be something that just takes more computation—novel techniques and ideas are probably required. So a large language model achieving it, even with considerable help and guidance from an expert such as Alpöge, would be the strongest evidence yet that the latest AI models can do much more than merely explain existing theorems or aid mathematicians in developing proofs. It would mean that they can, in fact, leverage the extant corpus of human thought to generate and rigorously establish genuinely new knowledge—in a manner that humans seemingly cannot.
The implications extend far beyond number theory. Mathematics underpins much of theoretical physics, engineering, computer science and other scientific disciplines, so an AI capable of proving GRH today could, tomorrow, help researchers tackle problems in those fields that might otherwise be beyond reach. This would be a new way of doing science, with AI systems moving from tools that assist in calculations and experiments to autonomous scientists capable of executing entire lines of theoretical investigation.
Whether humans could or would even understand the ultimate results of such potential AI-generated advances remains an open question.
For ordinary people, then, today’s news isn’t really about prime numbers at all—or any sudden, wrenching shift in daily life. Instead it’s about a future, perhaps imminent, if not immediate, in which machines supersede the capacities of human beings for making fundamental intellectual discoveries.
And that’s a sobering thought.