The Jacobian Conjecture and AI: A Dialogue Between Human and Machine
A Rare Collaboration
When Terence Tao, a mathematician of exceptional stature, engaged in a detailed conversation with ChatGPT about the Jacobian Conjecture, it marked a subtle but significant shift in how advanced mathematics might be explored. The exchange, shared in fragments on his blog, was not about solving the problem but about probing the boundaries of what AI can contribute to mathematical reasoning. It revealed both the potential and the limitations of using large language models as discussion partners in highly abstract domains.
The Jacobian Conjecture, formulated in the 1930s, asks whether every polynomial map from complex space with a constant non-zero Jacobian determinant must possess a polynomial inverse. Despite its simple statement, the problem has resisted solution for nearly a century, remaining one of the most intriguing open questions in algebra and geometry.
The Nature of the Problem
The conjecture sits at the intersection of several mathematical fields. It connects to algebraic geometry through the study of polynomial automorphisms, to differential equations via local invertibility conditions, and even to mathematical physics in its symmetry properties. While partial results have been proven in special cases—such as for mappings in one variable or under additional constraints—the general case remains unresolved.
Many experts suspect the conjecture might be false, yet no counterexample has withstood scrutiny. This absence of resolution has made it a focal point for theoretical exploration, with researchers investigating its implications across multiple domains.
How AI Fits Into the Process
Tao’s interaction with the AI began with precise definitions and targeted questions. He asked the model to generate candidate mappings with a Jacobian of one, a key requirement for a potential counterexample. The AI responded with suggestions that, while mathematically flawed, sometimes led Tao to reconsider overlooked approaches. In one instance, a misapplied construction prompted him to revisit shear transformations in higher dimensions—a direction with known connections to tame automorphisms.
These exchanges were not about correctness but about stimulation. The AI’s responses, though often inaccurate, acted as foils that helped clarify Tao’s own understanding. By challenging incorrect ideas, the model encouraged deeper reflection on why certain strategies fail, reinforcing the value of pre-rigorous thinking—the stage where intuition guides exploration before formal proof takes shape.
Limitations of Current AI Models
Despite its utility as a brainstorming tool, the AI exhibited clear limitations. It struggled with self-correction when errors were pointed out subtly, often agreeing to revisions without fully grasping their significance. It would sometimes repeat flawed reasoning after being corrected, indicating a lack of deep conceptual coherence. Unlike a human collaborator, it could not maintain a sustained, evolving narrative about the problem’s structure or implications.
Moreover, the model lacked the ability to distinguish between necessary and sufficient conditions, misapplied theorems from related fields, or proposed constructions that violated basic polynomial constraints. These errors, while not surprising given the model’s training and architecture, underscored the gap between pattern recognition and genuine mathematical insight.
A Shift in How We Approach Problems
Tao’s experiment did not resolve the Jacobian Conjecture, but it illustrated a broader transformation in mathematical practice. As AI systems grow more sophisticated, interactions like this may become more common—not as replacements for human intuition, but as catalysts for new lines of inquiry. The model’s role appears to be less about delivering answers and more about surfacing blind spots, suggesting alternative frameworks, or prompting reconsideration of assumptions.
This shift reflects a growing recognition that AI can function as a kind of intellectual sparring partner—one that reads widely but lacks judgment. In this capacity, it can help researchers test ideas, refine arguments, and explore uncharted territories of thought, even if it cannot ultimately solve the problems it engages with.
Conclusion
The dialogue between Terence Tao and ChatGPT offers a glimpse into the future of mathematical exploration. While AI will not replace the creativity and depth of human mathematicians, it can serve as a tool for expanding the boundaries of inquiry. The Jacobian Conjecture remains open, but the ways in which it is being approached are already evolving. As AI continues to advance, conversations like this one may become an integral part of how complex problems are investigated—one exchange at a time.