Sessions

AI, especially large language models like ChatGPT and Claude, is rapidly transforming science. Machine learning has already proven a powerful tool for finding patterns in large datasets and speeding up parts of the research process, such as writing code and performing literature searches. Now LLMs are emerging as full-fledged participants, developing hypotheses, designing experiments, and interpreting data. They’ve already had some striking successes, including solving a longstanding problem in theoretical particle physics and coming up with potential new drugs for a liver disease. There was even a scientific conference at which all 48 studies were partly done by AI. Already enthusiasts refer to LLMs as collaborators that in some ways outdo humans: tireless, all knowing, creative, available night and day. 

With the models becoming more powerful by the day, a dramatic acceleration of science may be upon us, first in mathematical specialties such as theoretical physics and molecular modeling but spreading fast to laboratory science. The structure and culture of science are sure to change as well, as senior scientists can do more with less—fewer graduate students, less interaction with collaborators—raising questions about the role of mentoring and the value of collaboration in an AI-transformed research world. Meanwhile, AI is also changing scientific publishing, presenting both the threat of low-quality or fraudulent research papers enabled by AI and more rapid, efficient peer review and quality control.