Why mathematicians feel caught between AI and their work

Several high-profile mathematicians have recently claimed that large AI companies used their unpublished or niche research to help these companies solve famous, decades-old math problems. These mathematicians argue that the companies often fail to acknowledge the human work that made these breakthroughs possible. In response, AI companies have occasionally corrected their own public announcements to include missing references to earlier human papers. Despite their frustrations and concerns about being made obsolete, these same researchers are still using AI tools in their daily work because they are simply too effective to ignore.

To understand why this is causing such friction, you have to look at how these AI models are built. Training an AI involves feeding it massive quantities of text, data, and code—often everything from the public internet—so the system can learn patterns, logic, and relationships between concepts. This process effectively creates a compressed, mathematical map of human knowledge. When you ask the AI to solve a problem, it isn't browsing a library in real time; it is using that internal map to generate the most statistically likely path to a correct answer. The problem is that the path it takes often incorporates ideas or techniques found in the vast collection of human research it was fed during training. Because these systems are so complex, it is virtually impossible to look under the hood and trace exactly which human paper or specific idea influenced a particular answer. This makes it impossible for researchers to know if the AI is building on their work or just mimicking it without attribution.

The traditional scientific method relies on a clear trail of breadcrumbs: every new discovery cites previous work, ensuring everyone gets credit and the entire community understands how a solution was reached. AI threatens to erase that trail. If researchers can no longer distinguish between their own contributions and the work of an opaque, automated machine, the very foundation of how science advances could change. While many mathematicians are calling for new rules and better credit for human labor, the efficiency gains from using these tools are so high that avoiding them could put a researcher at a massive disadvantage. We are entering an era where the most effective way to do research might involve using tools that potentially undermine the very system of credit that makes research worth doing in the first place.