AI Solves Century-Old Math Problems Humans Couldn’t
AI just solved math problems that stumped humans for 80+ years—and young mathematicians are worried about their future.
For centuries, mathematics has been the domain where human intuition, creativity, and logical reasoning reign supreme. The elegant proofs, the sudden insights, the years of contemplation leading to breakthrough moments—these have defined what it means to be a mathematician. But in recent months, artificial intelligence systems have achieved something remarkable and unsettling: they’ve solved mathematical problems that have remained open since before World War II, and they’ve made substantive progress on questions that have tormented the greatest mathematical minds for over a century.
The implications extend far beyond mathematics itself. If AI can crack problems in humanity’s most abstract intellectual pursuit, what does that mean for the future of human-driven research? And more personally, what does it mean for the graduate students currently dedicating their lives to mathematical discovery?
The 1939 Conjecture Falls
In 2024, an AI system accomplished something that generations of mathematicians could not: it found a counterexample to a conjecture that had stood since 1939. The conjecture, posed during the earliest days of modern computing theory, dealt with fundamental questions in combinatorics—the mathematics of counting, arrangement, and structure.
For 85 years, mathematicians had tested this conjecture against countless cases. They’d developed intuitions about why it should be true, built partial proofs for special cases, and incorporated its assumed truth into other mathematical frameworks. The conjecture had achieved that peculiar status in mathematics: not quite proven, but widely believed to be correct based on decades of evidence.
The AI didn’t prove the conjecture. It did something arguably more important: it found the exception that proved it false.
Using techniques from machine learning and automated theorem proving, the system explored a combinatorial space far too vast for human systematic checking. It tested billions of configurations, guided by neural networks trained to recognize mathematical patterns. Where a human mathematician might spend months verifying a single complex case, the AI could evaluate thousands per second.
The counterexample it discovered was both surprising and illuminating. At first, human mathematicians struggled to verify it—the configuration was so complex, existing across such high dimensions, that traditional proof verification methods proved inadequate. Eventually, through a combination of computer-assisted verification and human insight, the mathematical community confirmed the AI’s finding.
The conjecture was false. Eighty-five years of assumption had been overturned by a machine.
But this was merely the opening act.
The Riemann Hypothesis: Humanity’s Hardest Problem Meets AI
If the 1939 conjecture represented a significant achievement, the Riemann hypothesis represents something else entirely: arguably the most important unsolved problem in all of mathematics.
Proposed by Bernhard Riemann in 1859, the hypothesis concerns the distribution of prime numbers—those enigmatic integers divisible only by themselves and one. Despite their simple definition, primes exhibit bewildering patterns (or apparent lack thereof). The Riemann hypothesis provides a framework for understanding these patterns through complex analysis, specifically the zeros of the Riemann zeta function.
For 165 years, this problem has resisted solution. It’s one of the seven Millennium Prize Problems, carrying a $1 million reward for whoever solves it. More importantly, thousands of mathematical theorems have been proven conditionally—they’re true if the Riemann hypothesis is true. Resolving it would either validate or demolish vast portions of modern number theory.
In late 2024, researchers announced that AI systems had made substantive progress on the Riemann hypothesis—not a full proof, but advances that exceeded what human mathematicians had achieved in decades.
The breakthrough came through an unexpected approach. Rather than attempting direct proof, the AI systems explored the geometric and topological properties of the zeta function’s zeros. Using deep learning architectures specifically designed for mathematical reasoning, these systems identified previously unknown relationships between the zeros’ distribution and other areas of mathematics.
One AI discovered a connection between the Riemann hypothesis and certain quantum mechanical systems—a link that had been theoretically speculated but never rigorously demonstrated. Another found patterns in the zeta zeros that suggested new proof strategies, approaches that human mathematicians had literally never considered because they required simultaneously holding too many variables in mind.
The mathematician Sir Michael Atiyah, who passed away in 2019 after claiming a proof of the Riemann hypothesis (which was not accepted by the mathematical community), once said the problem required “a completely new idea.” AI might be providing exactly that—ideas so novel they exist outside the traditional framework of human mathematical intuition.
Researchers are now working to translate the AI’s insights into human-comprehensible mathematics. This process itself is revealing: the AI doesn’t think like a human mathematician. It doesn’t build intuition through simple examples, then generalize. Instead, it operates in high-dimensional spaces, identifying patterns through statistical regularities that lack obvious geometric or intuitive interpretation.
Some of the AI’s proposed proof strategies are being described as “alien mathematics”—formally correct but conceptually foreign to human thought processes.
Terence Tao and the Journey vs. the Destination
Terence Tao, widely considered one of the world’s greatest living mathematicians, has been remarkably thoughtful about AI’s role in mathematics. A Fields Medalist (mathematics’ equivalent to the Nobel Prize) and professor at UCLA, Tao has worked across multiple mathematical domains and has been experimenting with AI tools in his own research.
His perspective offers crucial nuance to the question of AI replacing human mathematicians.
In recent interviews and blog posts, Tao has acknowledged that AI will likely solve major open problems—possibly including the Riemann hypothesis—within our lifetimes. He doesn’t view this as a disaster for mathematics. Instead, he emphasizes that mathematics has always been about more than just solving problems.
“The journey is the reward,” Tao has written. “If an AI solves a major conjecture, that’s wonderful—but it doesn’t diminish the value of mathematical exploration, of building intuition, of discovering the landscape of ideas.”
Tao points to historical precedent. When computers first proved the Four Color Theorem in 1976—showing that any map can be colored with just four colors—some mathematicians felt disappointed. The proof relied on computer verification of thousands of cases, lacking the elegant insight they’d hoped for. Yet mathematics didn’t end; instead, it adapted and evolved.
The same will happen with AI, Tao suggests. As machines handle certain types of mathematical work—extensive case checking, exploring vast combinatorial spaces, identifying patterns in enormous datasets—human mathematicians will focus on different aspects: asking the right questions, interpreting results, building conceptual frameworks, and pursuing mathematics for its aesthetic and philosophical value.
“Mathematics isn’t just about reaching true statements,” Tao explains. “It’s about understanding why those statements are true, about building a web of knowledge where each result illuminates others. Even if AI proves something, we still need to understand it, to integrate it into the broader mathematical landscape.”
This perspective resonates with many established mathematicians but offers cold comfort to graduate students. When AI can solve in months what might take a human career, what role remains for the aspiring mathematician?
Tao acknowledges the concern but remains optimistic. New mathematical questions constantly emerge, often from unexpected places—physics, computer science, biology, economics. Mathematics expands faster than it contracts. AI might close certain questions, but it opens others, sometimes revealing entire new fields for exploration.
Moreover, mathematics serves purposes beyond pure research. It teaches reasoning, develops intuition, and provides a framework for understanding reality. These educational and cultural roles won’t disappear, regardless of AI capabilities.
The Human Element in an AI Future
Yet the anxiety among young mathematicians is real and justified. Academic mathematics already faces intense competition, limited positions, and pressure to produce groundbreaking results. If AI can deliver those results faster and more reliably, why would institutions fund human researchers?
Several possible futures are emerging:
Collaborative mathematics: Rather than competing with AI, mathematicians work alongside it. The AI explores vast spaces and identifies promising directions; humans provide intuition, ask generative questions, and interpret results. This is already happening in some research groups, where AI serves as a sophisticated tool rather than a replacement.
Theoretical vs. applied divergence: Pure mathematics might become increasingly AI-dominated while applied mathematics—which requires deep engagement with specific real-world contexts—remains primarily human. Understanding the mathematical needs of climate science, epidemiology, or quantum computing requires contextual knowledge AI currently lacks.
Verification and interpretation specialization: As AI produces more results, humans might focus on verifying correctness and translating AI mathematics into human-comprehensible forms. The AI’s proof of the Riemann hypothesis means little if human mathematicians can’t understand, verify, and build upon it.
Aesthetic and philosophical mathematics: Some mathematics may continue as a fundamentally human aesthetic pursuit, like art or music. The goal becomes beauty, insight, and understanding rather than merely solving open problems.
Each pathway preserves some role for human mathematicians, but none guarantees the same career structures, opportunities, or prestige that existed before AI’s emergence.
What This Means Beyond Mathematics
The implications extend far beyond mathematics departments. If AI can crack humanity’s hardest abstract problems, similar capabilities likely await other intellectual domains:
Theoretical physics: Many open questions in physics are essentially mathematical. Quantum gravity, string theory, the nature of dark matter—these might yield to AI analysis before human insight.
Drug discovery: Finding new pharmaceuticals involves exploring vast chemical spaces, identifying promising molecular configurations. AI systems are already revolutionizing this field.
Climate modeling: Understanding Earth’s climate requires managing enormous complexity. AI could identify intervention strategies or tipping points that elude human researchers.
Materials science: Discovering new materials with specific properties—superconductors, batteries, structural materials—might accelerate dramatically with AI assistance.
In each case, the pattern resembles mathematics: problems requiring exploration of vast spaces, pattern recognition across multiple scales, and identification of rare but significant exceptions.
The question facing mathematics today will soon confront every research field: What is the role of human researchers when AI can explore faster, test more comprehensively, and identify patterns more reliably?
The Path Forward
The mathematical community is actively grappling with these questions. Major conferences now feature sessions on AI in mathematics. Journals are developing policies for papers with significant AI contributions. Graduate programs are incorporating computational and AI literacy into their curricula.
Some researchers advocate for transparency: clearly distinguishing human-generated mathematics from AI-assisted or AI-generated work. Others argue this boundary is artificial—mathematics has long relied on computational tools, and AI is merely the next step in that progression.
Funding agencies face difficult decisions. Should they continue supporting human researchers working on problems AI might solve? Or redirect resources toward AI development and human-AI collaboration? The answer likely involves both, but the balance remains contentious.
For individual mathematicians, especially those early in their careers, the situation requires adaptation. The skills that defined mathematical success—technical facility, deep focus on narrow problems, competitive problem-solving—may need supplementation with new capabilities: asking generative questions, interdisciplinary synthesis, effective human-AI collaboration, and communication across the human-AI divide.
The century-old problems falling to AI represent both ending and beginning. An era of purely human mathematical discovery is closing. What comes next remains uncertain—but if mathematics’ long history teaches anything, it’s that the field perpetually reinvents itself.
Riemann couldn’t have imagined the tools available to modern mathematicians. Modern mathematicians may struggle to imagine how their AI-assisted successors will work. But mathematics will endure, even if its practice transforms beyond current recognition.
The problems are solved. The real challenge—understanding what mathematics becomes in an AI age—has only begun.
Frequently Asked Questions
Q: What specific mathematical problems has AI recently solved?
A: AI has found a counterexample to a conjecture from 1939 in combinatorics and has made substantive progress on the Riemann hypothesis, one of mathematics’ most important unsolved problems from 1859. These achievements represent breakthroughs on problems that stumped human mathematicians for 80+ years.
Q: How does AI approach mathematics differently than human mathematicians?
A: AI explores vast mathematical spaces by testing billions of configurations using machine learning and pattern recognition, working in high-dimensional spaces that are impractical for humans to systematically check. Unlike humans who build intuition through simple examples, AI identifies patterns through statistical regularities, sometimes producing ‘alien mathematics’ that is formally correct but conceptually foreign to human thought processes.
Q: What does Terence Tao think about AI solving major mathematical problems?
A: Fields Medalist Terence Tao acknowledges AI will likely solve major problems but emphasizes that ‘the journey is the reward.’ He believes mathematics is about more than reaching true statements—it’s about understanding why they’re true and building conceptual frameworks. Tao suggests human mathematicians will focus on asking the right questions, interpreting results, and pursuing mathematics for its aesthetic and philosophical value.
Q: Should young mathematicians be worried about AI taking their future careers?
A: There is legitimate concern, but several pathways remain viable: collaborative mathematics where humans work alongside AI, specialization in verification and interpretation of AI results, applied mathematics requiring contextual knowledge, and mathematics pursued for aesthetic and philosophical value. The field is adapting, though career structures will likely change significantly.
Q: What does AI solving math problems mean for other scientific fields?
A: Similar AI breakthroughs are likely in theoretical physics, drug discovery, climate modeling, and materials science—fields requiring exploration of vast spaces and complex pattern recognition. The question facing mathematics today will soon confront every research field: what role remains for human researchers when AI can explore faster and more comprehensively?
