AI Model Mentions in Biology Papers Nearly Catch Up to AlphaFold

Researchers at Epoch AI, a group that tracks AI progress, analyzed nearly every biology paper posted on bioRxiv, an online repository for preprints, meaning papers that have not yet gone through official publication, from January 2020 through September 2026. They used an AI algorithm to scan each paper and count how many times it mentioned AI. By September 2026, 17% of all papers referenced language models like ChatGPT, according to their data. That figure is close to the 19% that mentioned other kinds of scientific AI applications, a category led by AlphaFold, a protein-folding prediction tool built by Google DeepMind. At the start of 2023, mentions of language models were still rare. Only 4% of papers actually used a language model in the data-analysis process itself, as opposed to using it just for literature research, writing, or code.

The gap between 17% and 4% is telling. It shows that language models are taking on more and more of the "peripheral" parts of research, such as literature reviews, writing, and code, while the "core" tasks, meaning data analysis that needs human oversight, stay with people or with narrow, specialized AI tools like AlphaFold. Sometimes you do not need an all-powerful agent that does everything. For specific jobs, a specialized agent works better.