For seventy years, a series of unsolved mathematical problems posed by the late academic Paul Erdos have challenged the brightest minds. This month, the AI startup Harmonic announced its system, Aristotle, in collaboration with OpenAI’s GPT-5.2 Pro, had cracked one such “Erdos problem.”
The breakthrough has ignited discussion among scientists: is this genuine AI innovation, or just advanced pattern recognition? While some hail it as proof that artificial intelligence can now conduct original academic research, others remain skeptical.
Terence Tao, a highly-respected mathematician at UCLA, put it bluntly: “It feels like a student who has memorized everything for the test but doesn’t have a deep understanding of the concept.” The AI’s solution, he argues, relies on vast existing knowledge to simulate true insight.
Why This Matters
The debate isn’t merely about one solved equation. It strikes at the core question of whether AI can generate truly new ideas. Currently, most AI excels at identifying and remixing existing information. But if it can’t leap beyond that, its value to science may be limited to accelerating work humans still conceive.
This distinction is critical because:
- Scientific progress relies on fundamentally new concepts, not just faster calculations.
- The hype around AI’s creative potential must be grounded in reality. Overstating its abilities risks misallocation of resources and unrealistic expectations.
- The nature of intelligence itself is up for debate. If AI only mimics understanding, it raises questions about what intelligence actually means.
AI as a Tool, Not a Replacement
Despite the skepticism, Harmonic’s work demonstrates that AI is already a powerful tool. When combined with human expertise, it can accelerate research and uncover solutions faster than ever before. This suggests that the future of science may not be AI replacing researchers, but rather AI augmenting their abilities.
Whether AI is generating new ideas or not, it is becoming an indispensable tool in the hands of skilled scientists.
The question of true AI creativity remains open. But one thing is clear: the rapid pace of development forces us to re-examine what it means for a machine to “think.”
