[Digital Today reporter Yoonseo Lee] A student without formal mathematics training is drawing attention after solving a math problem that had remained unsolved for 60 years with help from ChatGPT.

On April 27, local time, online outlet Gigazine reported that 23-year-old math enthusiast Liam Price (리엄 프라이스) found a solution to an unsolved problem posed by mathematician Paul Erdos (폴 에르되시) using a different approach from existing experts.

The key lay in how he used artificial intelligence. Rather than starting with applying formulas or building a proof, Price first let ChatGPT generate ideas freely. He then selected and refined the results himself.

ChatGPT did not directly provide the correct answer. It did suggest approaches that were not easy for a person to think of, and Price chose viable clues and connected them to an actual solution. The final submission took a direction that even math experts had not considered.

The case is drawing attention because AI did not replace a mathematician’s role, but operated as a tool that expanded the range of human thinking. Mathematician Terence Tao (테런스 타오) pointed out, "People usually went in a slightly wrong direction from the very first step when tackling this problem." He thought the reason it remained unsolved for so long may have involved not only the problem’s difficulty itself, but also human habits of thought.

In the mathematics community, there are expectations that this kind of approach, less bound by existing intuition or convention, could be applied to other unsolved problems. In research settings, there are also reported cases of using ChatGPT to revisit past literature or find partial solutions that had been missed. Studies are also emerging in which new proof ideas were uncovered with AI support.

AI’s role also differed somewhat from typical generative AI use cases. In this collaboration, ChatGPT was not a calculator or proof tool that produces a finished answer, but closer to a brainstorming tool that broadly presents possible approaches. The key was the process in which a person distinguishes errors from possibilities and leaves only directions that are mathematically meaningful.

For this reason, the case also shows that human verification ability matters more than AI-use skills themselves. Because not all ideas produced by AI are correct, it required the ability to judge what to keep and what to discard rather than accept them as they are. Price’s achievement was not so much the result of following ChatGPT’s responses 그대로, but closer to a result in which a person interpreted the AI’s clues to the end and organized them into a form that could be proved.

Still, limits were clear. AI still faces constraints in autonomously constructing rigorous mathematical proofs. Here too, the proof produced by ChatGPT was incomplete, and a process was needed in which a person interpreted its meaning and verified it. It also confirmed that AI is not yet at a stage where it can immediately replace mathematicians.

Even so, the significance of this case is not small. A person without specialized training collaborated with AI to approach a solution that even experts had not reached. In mathematical research, what matters may be not only the ability to generate answers automatically, but also how to find paths that break out of existing frameworks of thought. Against this backdrop, AI is increasingly likely to first establish itself as an exploratory tool that broadens possibilities humans missed, rather than a tool that writes proofs in their place.