AIP report: Recent physics grads are no stranger to AI

A recent AIP report marks the first study to quantify the degree to which recent physics graduates are using AI in their work — and contributing to AI tool development.

“We all are, to a certain extent, learning and being exposed to AI tools,” says Patrick Mulvey, who manages the higher education studies carried out by the statistical research team at the American Institute of Physics, and who co-authored the report. And the intersection of AI with physics is interesting, he says, because some physicists use AI daily, while others build it.

The report found that 1 in 12 physics bachelor’s degree-earners and almost 1 in 5 physics doctorate-earners who entered the workforce in 2024 are working in AI development. Moreover, nearly 1 in 4 of physics bachelor’s degree-earners and 2 in 5 of physics doctorate degree-earners routinely use AI on the job.

“Currently working on computer vision models for various applications in the semiconductor industry,” wrote one private-sector respondent. One university researcher noted that they “develop and train convolutional neural networks for computer vision identification of objects.”

AI’s intersection with physics is decades old. First emerging as an area of interest in the 1950s, artificial intelligence — especially its use in consumer products — gained steam in the mid-2010s and exploded in 2022 with the launch of the large language model ChatGPT. In 2024, Geoffrey Hinton and APS Fellow John J. Hopfield were jointly awarded the Nobel Prize in Physics for foundational work that enabled the machine learning and artificial neural networks that underpin modern AI.

The AIP report’s findings were captured as a part of AIP’s 2024 Degree Recipient Follow-Up Survey, an annual census-type survey that captures initial post-degree outcomes of recent physics graduates. For over 40 years, the survey has provided insights to young physicists, including job options and salaries.

From the 2024 survey results, the most common job-related uses of AI tools reported by the recent physics bachelor’s degree recipients are for writing, debugging, or optimizing code; automating repetitive tasks or workflows; and data analysis or modeling. Many recent physics degree-earners also reported using AI to help them learn new or complex concepts.

Some physics educators have leaned into these tools, as well. Jennifer Klay, professor and chair of physics at California Polytechnic State University, has cautiously introduced AI into her department’s physics curriculum. In a study published earlier this year, Klay and colleagues reported on the use of conversational AI tools like ChatGPT to support introductory physics lab students with the analysis of periodic phenomena — in this case, tremors arising from diseases like Parkinson’s, which can be assessed using smartphone accelerometers.

“We didn’t just give them the prompt and ask for the output,” she says. The instructor team first trained students to manually extract the most important data. That background helped the students understand the utility of a particular algorithm, called a fast Fourier transform, and AI tools allowed them to implement quickly.

As a result, students in the lab now experience AI’s potential firsthand, says Klay, and the controlled approach gives the instructional team a chance “to show [the students] how to use these tools responsibly and ethically to enhance productivity or aid in learning without veering into academic dishonesty.”

Klay says other instructors in her department are using AI tools to support both student learning and assessments of learning, and publishers are introducing AI assistants trained on their texts as learning tools.

This exposure to AI in the physics curriculum could be helpful when students enter the workforce. “Employers are actually looking for these skills,” says Mulvey.

AIP’s report also found that, of the almost 40% of recent physics doctorate-earners who landed in potentially permanent positions (mostly in the private sector, says Mulvey) and routinely use AI tools, nearly half are using those tools to develop machine learning models, like ChatGPT.

One Ph.D. respondent, now a private-sector engineer, reported developing “camera-based detection models for autonomous vehicles.” Another respondent, a postdoc, said they “develop large AI models to estimate distances to galaxies.”

Some physics educators are concerned that students may be entering the AI-centric job market without sufficient training in ethics, important for developing new technology. In a 2023 study, Alice Olmstead and colleagues wrote that undergraduate physics programs “rarely prepare students to grapple with the complex, ethical decision-making that they will encounter,” even though STEM professionals often “make decisions that impact society in a wide variety of ways.” For example, when the AI system in a self-driving car decides whether to swerve around an obstacle, it leans on its decision-making architecture, which is imprinted with the ethical reasoning of its developers.

Olmstead’s team argues that ethical decision-making skills are as important for preparing physics students for the workforce as technical skills, like programming for AI tool development.

Klay echoes this sentiment. “Training students how to effectively and ethically use the tools,” she says, “is the most important area where [educators] can help prepare them for careers.”