Sarah Demers has a complicated relationship with artificial intelligence. As chair of the physics department at Yale University and leader of the American Physical Society's effort to produce an AI policy statement for the field, she is at the forefront of physicists' reckoning with the technology. Yet she has never typed anything into ChatGPT or Claude.
I have enough objections that I've never typed anything into Claude or ChatGPT, Demers told Quanta Magazine. I'm in a very weird space, personally, with this.
Her reservations stem from a core concern: popular LLMs are trained on datasets that include physicists' intellectual property — likely including her own work — without proper attribution. It's outrageous what's happened, she said. This is a moral hazard. We can't let this stand.
But Demers is no Luddite. Her group works on the Mu2e experiment at Fermilab, searching for a vanishingly rare particle transformation — a muon becoming an electron — and has received US Department of Energy funding through the Genesis Mission specifically to integrate AI into the experiment. The AI handles real-time optimization of magnetic fields, target placement, and system configurations. Sometimes it can feel like it's more of an art than a science, to tune these machines, she said. Having AI capabilities help us populate that space is a godsend.
On the theoretical side, LLMs are already accelerating calculations and reshaping what skills physicists need. We're seeing the fastest changes in the theoretical physics realm, Demers noted, comparing the shift to the historical requirement for physicists to read German — a prerequisite that has long since become obsolete.
What concerns Demers most is the impact on attribution and intellectual honesty. If we lose this conversation among everyone who's contributing, we're attacking our foundation, she warned. We will lose track of what's valid and what isn't. She believes physicists must take responsibility for everything they publish, even if they don't fully understand how a model produced a particular result.
For junior researchers, she sees particular danger in using LLMs as thought partners: You have to understand enough to know when you're going off the rails. I think there are risks for more junior researchers, or people with less experience, to try to use LLMs as a thought partner. That sounds dangerous to me.
In her personal life, Demers maintains a firm boundary. An LLM is not writing my emails; it's not planning my vacations, she said. Why are some emails difficult to write? Because I'm navigating something that's tricky, that I need to figure out. So having a large language model write it is me ignoring the root of the challenge that I am trying to solve.
The APS AI policy statement is expected to be released around spring 2027.
