Artificial intelligence has independently verified a proof for one of the most challenging problems in mathematics—the higher-dimensional sphere-packing problem—a feat that earned Ukrainian mathematician Maryna Viazovska a Fields Medal in 2022. This milestone marks a fundamental shift in how mathematical research is conducted, moving beyond AI as a mere computational tool to a collaborative reasoning partner.
The Evolving Role of AI in Mathematics
For centuries, mathematicians have relied on tools like abacuses, calculators, and computers to assist in calculations. However, these tools remained extensions of human intellect, never replacing the core reasoning process. The current emergence of AI in mathematics is fundamentally different: these systems now assist not just with calculation, but with reasoning itself, automating many underlying steps in mathematical arguments.
This change has been gradual. Modern mathematics already relies on complex frameworks and extensive catalogs of results that no single person can fully grasp. Computers have aided large proofs before, such as the four-color theorem and the Kepler conjecture, but today’s AI systems offer a new level of autonomy and reliability, especially when paired with formal proof assistants.
The Power of Formal Verification
Formal verification languages, such as Lean, express mathematical arguments in a way that computers can check step-by-step, ensuring logical soundness. Unlike traditional mathematical writing, Lean demands explicit definitions and inferences, meticulously checking each step. While unforgiving, this process eliminates hidden assumptions and leaps of faith. The result is mathematically certain, provided the proof passes Lean’s scrutiny.
Over the past few years, mathematicians have built extensive libraries within these languages, amassing definitions and verified theorems to tackle increasingly complex problems. The bottleneck was previously the time-consuming process of converting cutting-edge proofs into machine-checkable form—a task that could take months or years.
Breakthrough: Viazovska’s Proof Verified by AI
The recent verification of Viazovska’s higher-dimensional sphere-packing problem demonstrates the rapid progress in this field. The sphere-packing problem asks how tightly identical spheres can be packed together in spaces of any dimension. Viazovska solved the problem for eight and 24 dimensions, building on work that had only been completed for one, two, and three dimensions previously.
The AI startup Math, Inc., using its reasoning agent Gauss, played a pivotal role in translating Viazovska’s arguments into Lean code and verifying every step. The AI system was not working in isolation; mathematicians provided the initial blueprint and structure. However, once set up, Gauss completed the work in days—a task human researchers estimated would take months.
The Future of Mathematical Research
This is more than a technical achievement; it signals a fundamental shift in how mathematicians work. Fields Medalist Terence Tao suggests that AI’s immediate value lies in automating tedious but conceptually simple tasks, allowing mathematicians to focus on strategy rather than bookkeeping. This separation of creative idea generation from rigorous checking is the key.
Kevin Buzzard of Imperial College London warns against relying on unverified large language models, but argues that formal verification languages like Lean offer a solution. If a proof passes the program, it’s guaranteed to be valid. The main challenge now is translating more of modern mathematics into these formal libraries, giving AI systems the necessary concepts to work with.
Conclusion
AI isn’t replacing mathematicians, but redefining their role. The future of mathematics will likely involve building and tuning tools that extend human cognitive limits, pairing intuition with machine discipline. As verifiable mathematics expands, so will the demand for humans who can ask the right questions, create new definitions, and recognize genuine insights. The partnership between human intellect and artificial intelligence will drive the next era of mathematical discovery.
