The pace of artificial intelligence's progress in mathematics is refreshing the outside world's perception. Recently, OpenAI publicly released as many as 722 proof manuscripts of difficult mathematical problems in one go. These results span 17 different mathematical fields, and the overall problem-solving cost is far lower than before, causing enormous tremors in the mathematics community.
From a technical and application standpoint, although this batch of AI-generated proofs can pass rigorous machine verification, they expose clear pain points in practical translation. The vast majority of the proofs are difficult to convert into knowledge that humans can easily understand, and in practice there have even been "failures" in which a single symbol error caused multiple related conclusions to collapse collectively.
This progress has in turn triggered deep concern in academia about the research ecosystem. Mathematicians point out that when AI mass-produces vast numbers of proofs at extremely high efficiency, the subsequent burdensome work of sorting, verifying, and interpreting and consolidating them all falls on the shoulders of human researchers. The more fundamental issue is that AI at its current stage mainly solves problems by calling on existing knowledge and combining it logically, and it is very difficult for it to truly give birth to the new ideas and new tools that are the core of mathematical research.
Industry experts worry that, over time, human motivation to carry out the related interpretive work may keep declining, ultimately creating an insurmountable barrier of understanding between humans and machines. This trend may not only lock down the deep development of human civilization, but to some extent also runs counter to science's fundamental purpose of getting to the bottom of things and pursuing essential understanding.