In July 2026, the U.S. Department of Energy launched the Genesis Mission, funding 278 projects that aim to integrate artificial intelligence into scientific research. One of these projects involves using AI to assist physicists in detecting an extremely rare event: a muon transforming into an electron. This experiment, called Mu2e, is conducted at Fermi National Accelerator Laboratory near Chicago. Scientists like Sarah Demers, a physicist at Yale University and chair of the physics department, are exploring how AI can help optimize the experiment’s setup. “We have many variables to adjust,” Demers said, explaining how AI could quickly fine-tune factors like magnetic field strength, runtime, and system configurations. She described the process as sometimes feeling more like an art than a science, and said AI could be a valuable tool in that space. Despite her enthusiasm for AI in experimental physics, Demers is cautious about its use in other areas of her work and life. One of her concerns is that large language models (LLMs), which power many AI tools, are often trained on vast datasets that include copyrighted material, including the work of scientists like herself. These models frequently fail to credit the original sources, which she sees as a serious ethical issue. As AI becomes more integrated into physics research, scientists are also rethinking what it means to do physics. Demers is helping lead an effort through the American Physical Society to develop a policy statement on AI in the field. She emphasized the need to define what aspects of physics remain unchanged by AI, such as the importance of attribution and the scientific process itself. The rapid evolution of AI is already reshaping the skills needed in physics. Demers noted that in the past, physicists were required to read German to access key research papers, but such skills may not be as crucial now. As AI tools handle more complex calculations and data analysis, the focus is shifting toward understanding how to effectively use and validate these tools. While AI can speed up research and allow scientists to ask new questions, it also raises concerns about how to maintain scientific rigor and ensure proper credit is given to original work. Some physicists worry about the potential for misinformation or the erosion of traditional research practices. Demers also discussed the impact of AI on education and training in physics. She suggested that future students may be able to engage with data and answer complex questions more quickly than previous generations. However, she also acknowledged the value of the human experience of learning through trial and error, and the importance of meaningful conversations with peers. While some researchers are finding AI tools helpful as “thought partners,” Demers personally avoids using them for tasks like writing emails or planning personal activities. She believes that relying on AI in these areas could prevent her from addressing the root challenges she faces. Additionally, she worries that using AI to answer questions might reduce the opportunities for meaningful dialogue with others in her field.