An AI Just Solved a Decade-Old Math Problem. Now What?

A mathematician spent years on a problem he never solved. A decade after he posed it, an AI answered it — without being asked twice.

On April 4, a research team led by Peking University published a preprint describing how their dual-agent AI framework autonomously resolved an open problem in mathematics first proposed in 2014 by Dan Anderson, a former professor at the University of Iowa. Anderson passed away in 2022 at the age of 73, leaving the problem unanswered. The AI, it turns out, didn't need him to be around.

What the AI Actually Did

This wasn't brute-force computation. According to the researchers, the system worked by synthesising decades of mathematical literature — reading, in effect, the accumulated work of the field — and then generating and verifying hypotheses through two cooperating agents. One proposed; the other scrutinised.

That structure mirrors, at least superficially, how mathematicians actually work: absorbing prior research, identifying gaps, attempting novel connections, and backtracking when the logic breaks. Whether the AI genuinely reasoned through the problem or executed an extraordinarily sophisticated form of pattern recognition is a question the paper doesn't fully settle — and one the broader research community will debate loudly.

What's notable is the type of problem involved. Anderson's question wasn't a computation exercise. It required mathematical insight of the kind that has traditionally resisted automation. The claim, if it holds up under peer review, would mark a meaningful step beyond what systems like DeepMind's AlphaProof — which tackled select IMO problems in 2024 — had demonstrated.

Why China, Why Now

Sponsored

Advertise with Us

[email protected]

Sponsored

The timing is worth examining. The US–China technology competition has moved well beyond semiconductors and model benchmarks. Basic science is now contested terrain. A Chinese university team solving an open problem posed by an American mathematician — one who is no longer alive to respond — carries symbolic weight that the researchers may or may not have intended.

Peking University's result signals that China's AI ambitions extend into the domain of scientific discovery itself, not just commercial applications or language model performance. For Western researchers and policymakers already watching DeepSeek's emergence with unease, this adds another data point to a pattern that's harder to dismiss.

For investors and institutions tracking AI capability frontiers, the implication is straightforward: the race to automate mathematical and scientific reasoning is accelerating, and the leading participants are not all in Silicon Valley.

The Math World Isn't Convinced Yet

Skepticism is warranted — and healthy. The paper is a preprint, meaning it has not yet passed peer review. The mathematical community has historically demanded rigorous, independently verifiable proof, not just plausible-looking output. When computer-assisted proof first emerged with the 1976 four-color theorem, it sparked years of debate about what counts as legitimate mathematics. An AI-generated proof will face at least as much scrutiny.

Some researchers will question whether the AI's synthesis of existing literature constitutes genuine discovery or an elaborate recombination. Others will focus on whether the proof is formally complete and free of subtle errors that a human reviewer might catch. These are not minor concerns — they go to the heart of what mathematical truth means.

There's also an institutional dimension. If AI systems can resolve open problems by ingesting decades of published research, what does that imply for how mathematical knowledge is produced, credited, and taught? The question of authorship — who discovered something — becomes genuinely murky when the answer emerges from a machine trained on everyone's prior work.