When to jump? The boiled frog problem of using AI in physics education

From the diary: Physics teacher

A visual representation of the boiled frog problem produced by Sora that illustrates the main points of this paper in several ways. That is, the author's ability to provide a photo does not prove that he was able to create the photo. Credit: Sora (OpenAI)

WASHINGTON, January 6, 2026 — Generative AI is becoming ubiquitous in daily life. Large language models like ChatGPT can help you answer questions, write emails, and solve problems with seemingly lightning speed by engaging users in conversations from vast datasets. Generative AI tools are increasingly being used in the classroom, both to supplement learning and in some cases to cut corners.

Because of this, many teachers have expressed concerns about the impact this powerful tool will have on student learning and development. In The Physics Teacher, a joint publication of AIP Publishing and the American Association of Physics Teachers, the ETH Zurich physics professor turned AI researcher explores the use of generative AI to teach physical science, both its usefulness and its obstacles to learning.

In his paper, author Gerd Kortemeyer likens the ever-improving physics capabilities of generative AI to the fable of the boiling frog. In this fable, it is predicted that the frog will fail to recognize the danger of the slowly heating pot and will be too late to jump out.

“Generative AI could definitely be a helper, for example as a tool to quickly look up definitions, explain terms, write analysis programs, give instant feedback to students on explanations, or translate physical concepts into another language,” Kortemeyer said. “But they support human sense-making and collaboration, not the primary act.”

Research shows that students are using AI tools more frequently. Kortemeyer shows situations in which the use of generative AI may be justified, and situations in which it may not be useful for education. Therefore, “jumping out of the pot” is justified.

Courtmeyer explains that while AI tools can correctly solve homework problems from photos, unsupervised online assignments no longer hold up as examples of mastery because AI detection tools are questionable at best. But removing AI tools from education in general risks alienating students.

“Change is hard, and we've spent years perfecting our lecture notes and PowerPoint slides, and building beautiful problem banks for the slightly trickier scenarios,” Kortemeyer said. “As physics educators, I think we really need to completely recalibrate what we want to teach.”

Kortemeyer focuses on how to integrate AI into the learning process, allowing students to cite and critique the use of AI. These research-based teaching methods require stamina, but by adapting rather than ignoring them, educators can ensure that the water in the pot doesn't get too hot for the proverbial frog.

“We introduced generative AI ourselves. Physicists played a big role in developing the underlying principles. Now we have to live with the results, and that's not necessarily a bad thing,” Kortemeyer said. “If we can use this as an opportunity to focus on reasoning, collaboration, and true understanding rather than speed and mundane problem solving, our students may be doing the right thing in the end.”

###