Found this cool video by associate professor Giordano Scarciotti about the AI advances in math. It touched a spot in me.
Back in the days, before I owned a computer, I was very good at math. I could solve almost all non-olympiad high school math problems somewhat quickly, and btw, I can still do a bit of that. At that time, often faced the following choice:
There’s a new problem that clearly has many ways to be solved. You can brute force it, you can achieve a shallow understanding and produce a long solution, or you can understand it deeply and it solves itself. Each approach leads to the same answer but only one solution makes you better at math.
According to most videos on the subject, the math discoveries that came from AI didn’t come with any deep understanding or with new theories. They came with using the existing ones a lot. It treated math as a sudoku and autocompleted the work of thousands of humans, leaving them with no path forward.
This method provides answers but no human understanding or sustainable advancement. No good new theories. Few good problems for future research. No new ways for smart students to contribute. It also came with a threat or a promise that any good new idea will be autocompleted the same way with little or no attribution.
