AI reaches research-mathematics parity with human mathematicians ahead of 2030 horizon?

By late 2029, AI systems routinely solve open research problems that humans have worked on for years. Mathematics journals publish proofs first discovered by neural networks, refined by human mathematicians. Top universities hire AI-in-residence positions because the systems can sketch novel approaches to hard problems faster than a team of PhD students. The boundary between "tool" and "researcher" blurs—competitions like the IMO have added AI categories because human mathematicians can no longer compete on pure problem-solving speed.

Through 2029, AI stays excellent at routine mathematics but hits a wall on truly novel research. The o1 series and successors win benchmarks by pattern-matching training data and rehearsed competition tricks, not by inventing new mathematical insight. When researchers ask AI to tackle unsolved problems—the kind that require genuine originality—the systems either recycle existing techniques or produce impressive-looking but wrong answers. Human mathematicians remain the only source of truly new mathematical breakthroughs; AI is a powerful calculator, not a research peer.