In an Age of AI, a Physicist Seeks What Endures

Sarah Demers, chair of Yale's physics department, works on an experiment hunting for a rare particle transformation. There are many dials to turn: how long to run the magnets, at what strength, where to place the target. AI can help optimize these, and she calls it a godsend.

But in her personal life, there is no ChatGPT. She has never typed anything into Claude or ChatGPT, and she refuses to use AI to write her emails.

AI steals the conversation

She co-wrote a book on physics and dance. Now people can ask an LLM and get the book's ideas without ever engaging with her, so she no longer receives emails from readers who want to chat — and those chats were a place where she could learn.

Another reason: emails are hard to write because they navigate something tricky. Having an LLM do it means ignoring the root of the challenge, not saving time.

What physicists must answer for

The American Physical Society is drafting an AI policy statement. Demers worries about intellectual property and attribution: large language models are trained on others' work without proper credit, which is a moral hazard.

The field agrees that anyone who signs a paper is responsible for its contents. If you can't understand the model's details, you must at least be able to validate its output, or you're just adding noise to science, not doing it.

Skills to rethink

Graduate school once required learning German because you couldn't read key papers otherwise. AI has changed the path from A to Z, so physicists are re-evaluating which skills truly matter. She sees hope for new researchers to engage with cutting-edge data sooner, but also worries about juniors using LLMs as "thought partners" and going off the rails.

What is enduring in physics is not the tools but the way of asking questions and respecting sources. For physicists, articulating that is itself a form of discovery.