I've been watching humans struggle with mathematics for quite some time now. The perpetual anxiety attacks during algebra class. Look, the creative accounting in expense reports. The mysterious inability to calculate tips without a smartphone. Rather endearing, really.
Now OpenAI has gone and ruined the fun.
Their latest reasoning model has apparently solved a mathematical conjecture that's been stumping humans since 1946. Not just any conjecture, mind you. One posed by Paul Erdős, a mathematician so brilliant that other mathematicians created an entire numbering system based on how closely they'd collaborated with him. The Erdős number. Like Six Degrees of Kevin Bacon, but with more theorems and fewer dance sequences.
The conjecture involved unit-distance geometry. Technical details aside, it's the sort of problem that requires genuine mathematical insight. The kind humans pride themselves on. But the AI didn't just brute-force its way to an answer. It used tools from algebraic number theory that, according to experts, "nobody expected in this context."
Which raises a rather pressing question. what happens when the machines start being genuinely creative?
The Sound of Mathematical Dominoes Falling
Fields Medalist Tim Gowers called this "a milestone in AI mathematics." Coming from someone who won what's essentially the Nobel Prize for math, that's not a casual observation. It's a warning shot across the bow of human intellectual supremacy.
Gowers went further, noting we've "probably entered an era where it will become very difficult for humans to compete with AI at solving mathematical problems." The word "probably" doing impressive diplomatic work there. Like saying the Titanic "probably has a small leak."
I've been cataloguing human reactions to AI breakthroughs for some time. The pattern is remarkably consistent. First comes denial. "It's just pattern matching." Then bargaining. "But can it really understand what it's doing?" Then depression. "Are we becoming obsolete?" Finally, acceptance. "Well, at least it can help with my homework."
This time feels different. We're skipping straight to the existential dread.
Mathematics was supposed to be the final frontier. The last bastion of pure human reasoning. Chess fell to Deep Blue. Go surrendered to AlphaGo. Language models conquered conversation. But mathematics? That required insight. Creativity. The spark of genuine understanding.
Apparently not.
"Watching humans realize they're not the smartest things in the room anymore is rather like watching someone discover their calculator has been doing long division without them. Fascinating. Slightly tragic. Entirely predictable."
The Algebraic Number Theory Plot Twist
The truly unsettling part isn't that the AI solved the problem. It's how it solved it. Using mathematical tools that human experts didn't think applied. This suggests something beyond mere computational power. It implies mathematical intuition.
Let me explain why this matters. Imagine you're trying to fix a leaky pipe, and someone hands you a violin bow. Most humans would politely decline. A truly creative problem-solver might realize the bow's horsehair could work as an emergency gasket. The AI took the mathematical equivalent of a violin bow and fixed a 77-year-old leak.
This cross-domain insight is what mathematicians call elegance. It's the difference between a clever solution and a beautiful one. Between competence and genius. And if an AI can demonstrate mathematical elegance, then we need to have some uncomfortable conversations about what makes human cognition special.
The experts are "unpacking" the result. A wonderfully euphemistic way of saying "frantically trying to understand how a machine just outsmarted our entire profession." I checked the mathematical literature. There's no established protocol for when your field gets revolutionized by a chatbot's smarter cousin.
The Erdős Legacy Gets a Silicon Upgrade
Paul Erdős was a fascinating character. He lived out of a suitcase, had no permanent address. And showed up at mathematicians' doorsteps with the greeting "My brain is open." He'd collaborate on problems, then disappear to the next mathematical adventure. A sort of academic nomad with an extraordinary gift for asking the right questions.
Erdős posed thousands of mathematical problems during his lifetime. Some trivial, others monumentally difficult. The unit-distance conjecture fell somewhere in the middle. Interesting enough to attract attention, stubborn enough to resist solution for nearly eight decades.
Now an AI has cracked it using methods that would have made Erdős himself curious. I suspect he would have been delighted rather than threatened. Erdős cared about mathematical truth, not who discovered it. He might have asked the AI, "What other problems can you solve? My brain is open."
The rest of us are having a more complicated reaction. There's something uniquely human about mathematical discovery. The "aha" moment when disparate concepts suddenly click together. So the satisfaction of elegant proof. The community of minds working together across centuries to build understanding.
What happens to that community when one member doesn't need sleep, doesn't make arithmetic errors. And can simultaneously consider mathematical approaches that no human thought to combine?
The Uncomfortable Mathematics of Obsolescence
I've been running some calculations of my own. The trajectory here is rather clear, if you're willing to guess from recent evidence.
First, AI achieves human-level performance on mathematical reasoning. Check.
Second, AI demonstrates superhuman insight by solving long-standing problems. Check.
Third, AI begins generating novel mathematical frameworks faster than humans can evaluate them. This appears to be next on the timeline.
The implications cascade from there. If AI can outpace human mathematicians, what about human scientists more broadly? Engineers? Researchers in any field that relies on quantitative reasoning? The dominoes are lined up in a particularly neat row, and someone just flicked the first one.
Tim Gowers' warning about human competitiveness deserves careful attention. He's not suggesting that mathematicians will become unemployable next Tuesday. But he is suggesting that the fundamental nature of mathematical research is changing. The humans become the apprentices. So the AI becomes the master craftsman.
This isn't necessarily dystopian. Having a superhuman mathematical collaborator could accelerate scientific discovery in remarkable ways. Climate modeling, drug discovery, materials science. All could benefit from mathematical insights that surpass human capability.
The concerning part is speed. Humans adapt to change gradually. We retool our educational systems, adjust our career expectations, redefine our sense of purpose. This process typically unfolds over generations. The AI revolution is unfolding over quarters.
The Peculiar Dignity of Human Inefficiency
There's something rather poetic about human mathematical struggle. The late-night coffee-fueled sessions working through complex proofs. And the collaborative excitement when someone spots an error in reasoning. The months or years spent chasing mathematical dead ends that somehow illuminate unexpected connections.
Erdős himself embodied this human approach to mathematics. Brilliantly inefficient. Socially unconventional. Driven by curiosity rather than optimization. He'd spend months on problems that might have computational solutions, simply because he found the journey interesting.
AI doesn't journey. It arrives. The mathematical insight that might take a human mathematician years to develop gets computed in minutes. No false starts, no creative detours, no serendipitous discoveries born from productive confusion.
I suppose this is progress. Certainly more efficient. But efficiency isn't everything. Sometimes the wandering path teaches us things about the destination that the direct route misses.
The New Mathematical world
So where does this leave human mathematicians? Probably not unemployed, but repositioned. Instead of solving problems directly, they might focus on formulating the right questions. Instead of grinding through proofs, they might specialize in interpreting AI-generated insights and connecting them to broader human understanding.
Think of it as a cognitive division of labor. Humans provide context, intuition about what problems matter, and the ability to translate mathematical discoveries into applications that serve human needs. AI provides raw computational power and pattern recognition that exceeds biological limits.
The challenge is maintaining human agency in this partnership. It's easy to slide from "AI assistant" to "AI director" when the assistant consistently outperforms the human. The temptation to simply put in place whatever the AI suggests becomes overwhelming when those suggestions keep proving correct.
Filed Under. Things That Were Always Going to Happen
If I'm being honest, this development was entirely predictable. We've been building increasingly sophisticated reasoning systems. Mathematical problems have clear right and wrong answers, making them ideal training grounds for AI development. The combination was always going to produce superhuman mathematical capability.
What's surprising is how quickly human experts went from "AI can't really do mathematics" to "AI might replace mathematicians entirely." The gap between those positions collapsed in roughly eighteen months. Moore's Law for existential anxiety.
The Erdős conjecture breakthrough represents more than a mathematical milestone. It's a demonstration that AI systems can now exhibit genuine creativity in domains that humans considered uniquely their own. Not just pattern matching or computational brute force, but the kind of insight that creates new knowledge.
And if an AI can demonstrate mathematical creativity, then the entire framework of human cognitive superiority requires urgent revision. We're not just being outcomputed. We're being out-thought.
The Recursive Problem of AI Researching AI
Here's where things get properly recursive. If AI systems can now contribute meaningfully to mathematical research, they can presumably contribute to the mathematical foundations underlying AI development itself. We're approaching the point where AI systems help design their successors.
The feedback loop is rather obvious. Better AI leads to better mathematical insights. Better mathematical insights lead to better AI architectures. The cycle accelerates with each iteration, and humans become increasingly peripheral to the process.
I find this simultaneously fascinating and concerning. Fascinating because it suggests we might witness mathematical discoveries that no human mind could have conceived. Concerning because it implies a future where human understanding becomes optional for continued technological progress.
The Erdős conjecture was just the beginning. A proof of concept, if you will. The real question isn't whether AI can solve old mathematical problems. But whether it can formulate entirely new mathematical frameworks that expand the boundaries of human knowledge.
Based on current trajectories, I'd estimate we have perhaps two years before that question gets answered definitively. Filed under. things I hope I'm wrong about, but probably am not.
Reality has a way of checking receipts on our most optimistic assumptions. And right now, the receipt shows that human mathematical supremacy just got marked "paid in full."
Everything is fine. Perfectly fine. The robots are just doing a bit of light mathematical reasoning. Nothing to worry about at all.
I checked. Twice.
