In May 2026, an AI model solved a famous math problem first posed in 1946, and within weeks, more than a dozen other historic problems have fallen to artificial intelligence. These events are not just reshaping mathematics—they also offer a fascinating window into how human problem-solving compares to machine learning.
The Research
According to a Quanta Magazine report (August 2026), OpenAI's internal models have solved multiple legendary problems from Paul Erdős, the itinerant Hungarian mathematician who posed over a thousand questions during his lifetime. On May 20, 2026, an AI found a counterexample to the "unit distance problem," first conjectured in 1946. The result was not fully rigorous, but it was innovative, drawing from a distant branch of mathematics. Soon after, techniques from that solution were used to solve other open problems. By August 1, OpenAI's unreleased model Astra achieved 10 more advances, including solutions to three additional Erdős problems.
Why Erdős problems? Many are simple to state but require deep insight and creativity—exactly the kind of challenge that might be a good benchmark for AI. Thomas Bloom, a mathematician at the University of Manchester, created erdosproblems.com in 2023 to track these problems. He used ChatGPT to help code the site, an early sign of how AI is becoming part of a mathematician's toolkit.
Why It Matters for Your Brain
These AI advances show that machines can now handle tasks that were once thought to require human intuition and leaps of logic. But as a cognitive researcher, I see this as a challenge to us: how do we nurture the human abilities that AI still misses—like making novel connections across domains, or questioning the problem itself? AI excels at exploring vast solution spaces, but humans still frame the questions and decide what is worth solving. The fact that AI needed to borrow ideas from a far-off field reminds us that cross-disciplinary thinking is a signature of human creativity, and we can train it.
What You Can Do
To strengthen your own problem-solving, follow these evidence-based strategies:
- Practice cross-domain thinking: deliberately apply ideas from one field to another (e.g., use a musical pattern to solve a math puzzle).
- Engage with open-ended puzzles (like those on our brain training platform) that have no single right answer.
- When stuck, take a break and return later—incubation helps your unconscious mind work.
- Explain problems to someone else (or a rubber duck) to force yourself to clarify your thinking.
These techniques can boost flexibility and reasoning. And remember: even the most powerful AI still needs a human to pose the problem—that part is all you.
Source: Quanta Magazine
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.
