What is the Jacobian Conjecture, and why a Tao-ChatGPT chat about it went viral

A link circulating widely among programmers and researchers this week was not a product launch or a research paper, but a shared ChatGPT conversation: a transcript showing the prominent mathematician Terence Tao working through ideas related to the Jacobian Conjecture, one of the more stubborn open problems in algebraic geometry, with an AI chatbot as his sounding board. The exchange offers a rare, unfiltered glimpse into how a leading mathematician actually uses AI tools while thinking through a hard problem, rather than a polished summary of a finished result.

The Jacobian Conjecture, first proposed by the mathematician Ott-Heinrich Keller in 1939, concerns a specific class of polynomial functions that map multidimensional space onto itself. Roughly stated, it asks whether a polynomial map with a particular algebraic property, a constant, nonzero Jacobian determinant, a quantity from calculus that measures how a transformation stretches or compresses space, must always be invertible, meaning it must always be possible to undo the transformation with another polynomial map.

The conjecture sounds narrow, but it sits at the intersection of algebraic geometry, complex analysis, and dynamical systems, and has resisted proof for over eight decades despite sustained attention from serious mathematicians. It has also become somewhat notorious within the field for attracting a steady stream of proposed proofs over the decades that were later found to contain subtle errors, a track record that has made mathematicians particularly cautious about claiming a resolution, even a partial one, without extremely careful verification.

What makes the shared conversation notable is not that it resolves the conjecture, and nothing in the exchange amounts to a verified proof or disproof; formal mathematical results require rigorous peer review that a chat transcript cannot substitute for. What it shows instead is a working mathematician using an AI system as a collaborative thinking tool, testing potential approaches, checking specific algebraic manipulations, and exploring whether a particular line of attack toward a counterexample might hold up, in much the same way a mathematician might think out loud with a colleague or a whiteboard.

This kind of use, sometimes described as AI-assisted mathematical exploration, differs meaningfully from asking an AI system to simply produce a proof outright. Complex, novel mathematical reasoning remains an area where AI systems can make convincing-looking but ultimately incorrect claims, particularly on problems, like the Jacobian Conjecture, where subtle errors have tripped up expert humans for decades. Mathematicians who use AI tools in this exploratory mode generally treat the system's suggestions as hypotheses to be independently verified, rather than as trusted conclusions.

Tao has been a prominent, relatively public voice on how mathematicians might productively use AI tools, including experimenting with AI systems as aids for exploring proof strategies, checking calculations, and formalizing arguments in verification systems. His willingness to share a raw, unedited conversation, including whatever dead ends or corrections it may contain, offers a different kind of insight than a curated announcement, showing the messier, iterative process of mathematical thinking rather than a finished result.

The episode reflects a broader shift underway in how some researchers across mathematics and the sciences are incorporating AI chatbots into their daily working process, not as a replacement for formal proof or rigorous derivation, but as a fast, interactive way to test ideas, catch arithmetic slips, or explore a problem space before committing to a more careful, formal line of investigation. That shift is still relatively new and unevenly adopted across the mathematical community, with plenty of working mathematicians skeptical of relying on AI systems for anything beyond routine computation.

For those outside mathematics, the appeal of the shared conversation lies partly in its accessibility: watching a leading expert reason through a genuinely hard, unresolved problem in real time, mistakes and uncertainty included, offers a more honest picture of how mathematical research actually proceeds than the clean, retrospective narrative typically presented in a published paper.

Whether the specific line of reasoning in Tao's conversation ultimately leads anywhere toward the Jacobian Conjecture remains an open question, and one that will be settled, if it is settled at all, through the traditional process of formal proof and peer review rather than through a chat transcript. What the episode has already demonstrated, regardless of where that particular mathematical thread leads, is a glimpse of how AI tools are beginning to fit into the working habits of even the most accomplished mathematicians.