How Physics Professors Use and Frame Generative AI Tools

Through interviews with 12 physics professors, this study identifies six epistemic frames through which faculty perceive Generative AI, ranging from a useful tool for coding and labor-saving to a significant threat to genuine learning, thereby revealing how these technologies are reshaping the identity and practices of modern physicists.

The Gist

Imagine the world of physics as a massive, high-stakes kitchen. For the last century, the chefs (physicists) have been using new tools to cook: first, they traded hand-written recipes for calculators, then for computers that could chop vegetables (solve equations) and whisk ingredients (run simulations) faster than any human could.

Now, a new, magical sous-chef has arrived: Generative AI. This new assistant doesn't just chop vegetables; it can write the recipe, taste the soup, and even explain why the soup tastes salty, all while you chat with it in plain English.

This paper by Skogvoll and Odden is like a series of interviews with 12 head chefs at a famous Scandinavian culinary school. The researchers wanted to know: How are these expert chefs reacting to this new magical assistant? Are they terrified it will ruin their cooking? Are they using it to make dinner faster? Or are they trying to figure out how to teach their apprentices (students) to use it without letting the robot take over the kitchen?

Here is the breakdown of their findings, translated into everyday language:

The Big Picture: A Double-Edged Sword

The chefs have a split personality regarding this new tool. On one hand, they see it as a super-powered helper that can do the boring, repetitive work. On the other hand, they are deeply worried that if they let the robot do too much, the students (and even the chefs themselves) will forget how to cook from scratch.

The researchers found that the chefs view this AI through six different "lenses" or frames. Think of these like different pairs of sunglasses the chefs put on to look at the same tool.

The Six Lenses (Frames)

1. The "Thief" Lens (Threat to Learning) 🔒

This is the most common lens. The chefs are worried that the AI is a cheat code.

• The Fear: If a student can ask the AI to solve a physics problem or write a lab report in seconds, how do we know they actually learned anything? It's like letting a student use a calculator for a test they haven't studied for; they get the right answer, but their brain didn't do the work.
• The Reaction: To fight this, some chefs are changing the rules of the game. Instead of giving take-home essays (which the AI can write), they are switching to oral exams (where the student has to explain their thinking out loud) or asking students to declare exactly how they used the AI. It's like saying, "You can use the robot, but you have to show us your notes on how you talked to it."

2. The "Encyclopedia" Lens (Source of Knowledge) 📚

Here, the chefs see the AI as a super-smart, slightly unreliable librarian.

• The Use: They use it to quickly get a summary of a complex topic or to check a basic fact.
• The Catch: The chefs know the librarian sometimes makes things up (hallucinates). So, they treat the AI's answers like a rumor: "It sounds plausible, but I need to double-check the source." They even teach students to play "detective" and find the AI's mistakes, turning the AI's unreliability into a learning exercise.

3. The "Sparring Partner" Lens (Discussion Partner) 🥊

In this view, the AI is a practice dummy or a debate buddy.

• The Use: Chefs use it to bounce ideas around. "What if I tried this experiment?" they ask the AI. The AI suggests three possibilities. It helps them brainstorm.
• The Reality Check: The chefs noticed that students often treat the AI like a vending machine (just asking for the answer) rather than a conversation partner. The chefs wish students would use it to argue and refine ideas, but often, students just want the homework done.

4. The "Code Monkey" Lens (Coding Tool) 💻

Physics involves a lot of computer coding. For decades, chefs had to learn to write code by hand.

• The Shift: Now, the AI can write the code for them.
• The Debate: Some chefs say, "Great! I don't need to spend hours debugging; I can just ask the AI to write the script so I can focus on the physics." Others worry, "If we outsource the coding, will the students ever learn how to build the engine themselves?" It's a debate between efficiency and mastery.

5. The "Editor" Lens (Text-Processing Tool) ✍️

This is the most accepted use. The AI is seen as a super-powered spellchecker and translator.

• The Use: Fixing grammar, making sentences flow better, or translating a paper from Norwegian to English.
• The Rule: The chefs are okay with the AI polishing the surface of the work, but they draw the line at letting it write the ideas. It's like hiring a professional editor to fix your typos, but not hiring them to write your novel for you.

6. The "Time-Saver" Lens (Labor-Saving Device) ⏱️

Finally, the chefs see the AI as a robot vacuum cleaner for their busy lives.

• The Use: Writing boring emails, formatting documents, or summarizing long lists of research papers.
• The Goal: The hope is that by letting the AI do the "grunt work," the chefs can spend more time doing the "fun work": talking to students, designing cool experiments, and thinking deeply about physics.

The Conclusion: A Kitchen in Transition

The paper concludes that physics is in a transition period, much like when the world moved from hand-cranking cars to driving automatic ones.

• Right now: The chefs are mostly using the AI to make small tweaks to their routine. They are using it to save time on boring tasks, but they are still holding the steering wheel.
• The Future: The researchers wonder if, in a few years, the AI will change the job description of a physicist entirely. Will the chef become more of a "manager" of the AI, or will the AI become the main cook?

The Takeaway:
The professors aren't Luddites (people who hate technology). They are excited about the potential to do more meaningful work. But they are also fiercely protective of the learning process. They believe that if you let the robot do all the thinking, the human brain atrophies. Their goal is to find a balance where the AI is a tool in the toolbox, not the hand that holds the hammer.

In short: Use the robot to chop the onions, but make sure the student still learns how to taste the soup.

Technical Summary

1. Problem Statement

The rapid emergence of Generative AI (GenAI), particularly Large Language Models (LLMs), is reshaping the landscape of scientific research and education. While GenAI offers capabilities in coding, text generation, and data analysis, it simultaneously poses significant challenges to academic integrity, authorship, and cognitive development.

• The Gap: Despite the transformative potential of GenAI, there is a scarcity of empirical research regarding how physics faculty specifically perceive, adopt, and integrate these tools into their professional workflows.
• The Context: Physics is an empirical and computational discipline. Unlike fields relying solely on argumentation, physics relies on mathematical derivations, numerical models, and experimental results. The authors posit that physicists, being generally tech-savvy, may adopt GenAI differently than other disciplines, yet the specific nature of this adoption remains under-explored.
• Research Questions:
  1. How are physics professors using GenAI tools in their teaching and research?
  2. How do physics professors frame (conceptually understand) these tools?

2. Methodology

The study employs a qualitative research design grounded in Epistemic Framing theory.

• Participants: 12 physics professors from a major Scandinavian research university (University of Oslo).
• Data Collection: Semi-structured interviews conducted in late spring and early fall of 2024. Questions focused on current GenAI usage in teaching and research, student feedback, personal experiences, and future outlooks.
• Data Analysis:
  • Inductive Thematic Analysis: The first author coded transcripts to identify initial themes regarding GenAI perception and usage.
  • Epistemic Framing Framework: The authors applied the theoretical lens of epistemic framing (how individuals perceive the nature of knowledge and learning situations) to categorize the data.
  • Iterative Coding: The team grouped initial themes into six overlapping "epistemic frames" and identified 19 specific "practices." A second round of deductive coding was performed to ensure consistency and accuracy.
  • Output: The analysis resulted in a taxonomy of practices organized by the epistemic frames through which professors interpret GenAI.

3. Key Contributions

The paper makes three primary contributions to the field of Physics Education Research (PER) and the broader discourse on AI in academia:

1. Taxonomy of Practices: It identifies and categorizes 19 distinct professional and pedagogical practices where physics professors utilize GenAI, ranging from coding assistance to assessment redesign.
2. Epistemic Framework: It establishes six overlapping epistemic frames that explain how professors cognitively process GenAI. This moves beyond simple "adoption vs. resistance" binaries to a nuanced understanding of how the tool is contextualized.
3. The "Threat" Overlay: It identifies a unique meta-frame where the perception of GenAI as a threat to genuine learning permeates and colors all other positive frames, suggesting a pervasive anxiety about academic integrity that influences tool usage.

4. Results

The analysis revealed that professors hold simultaneous positive and negative framings of GenAI. The findings are structured around six epistemic frames and their associated practices:

A. The Six Epistemic Frames

1. GenAI as a Threat to Genuine Learning and Assessment:

   • Core View: GenAI bypasses learning, replaces writing, and offloads thinking, making it difficult to assess student competence.
   • Dominance: This was the most prevalent frame, appearing twice as often as warrants for other frames. It acts as an "overarching concern" that colors all other interactions.
   • Associated Practices: Shifting assessment formats (e.g., oral exams), requiring AI-use declarations, discussing AI use with students, assigning AI-use reflections, and vetting AI reliability.

2. GenAI as a Source of Knowledge:

   • Core View: GenAI is an information repository capable of explaining concepts and solving problems, but requires critical verification.
   • Nuance: Professors treat it like a textbook or colleague but emphasize the need for "critical evaluation" due to hallucinations.
   • Associated Practices: Vetting subject reliability, critically evaluating AI output, getting answers/hints, and using AI for topic exploration.

3. GenAI as a Discussion Partner:

   • Core View: GenAI acts as a "sparring partner" or "buddy" for brainstorming and refining ideas (Socratic dialogue).
   • Reality Check: While professors hope for deep dialogue, they report students mostly use it for low-level technical tasks.
   • Associated Practices: Brainstorming, getting formative feedback on writing, refining language, and assisting in literature reviews.

4. GenAI as a Coding Tool:

   • Core View: GenAI is a pragmatic utility for writing, debugging, and visualizing code.
   • Specialization: Physicists who view coding as a "means to an end" outsource it to AI. Those who view coding as part of their identity worry about students losing foundational skills.
   • Associated Practices: Outsourcing simple code generation, creating visuals (plots/diagrams), and refining/debugging code.

5. GenAI as a Text-Processing Tool:

   • Core View: GenAI is a superior tool for grammar, style, summarization, and translation compared to traditional tools (e.g., spell checkers).
   • Hierarchy of Acceptability: Refining existing text is acceptable; drafting initial ideas is viewed with skepticism.
   • Associated Practices: Drafting content, refining language, getting formative feedback, assisting literature reviews, and translating content.

6. GenAI as a Labor-Saving Device:

   • Core View: GenAI increases efficiency by automating "grunt work" (administrative tasks, routine grading, literature sorting).
   • Associated Practices: Automating administrative work, exploring automated grading, and assisting literature reviews.

B. Key Observations

• Specialization: The study highlights a divergence in adoption based on identity. Physicists who do not identify as coders outsource coding to AI, while those who do engage with AI as a partner.
• The "Threat" Permeation: Even when using GenAI as a tool, the fear of it undermining learning remains a background constant. For example, using AI for coding is accepted only if the student still understands the underlying physics.
• Assessment Shift: To counter the "Threat" frame, professors are actively shifting from unsupervised written exams to oral defenses and in-person assessments.

5. Significance and Implications

• Redefining the Physicist: The paper argues that GenAI is beginning to redefine what it means to "do" physics. Just as computers shifted physics from analytical integration to numerical modeling, GenAI may shift the role of the physicist from a generator of code/text to an engineer of learning environments and a critic of AI output.
• Pedagogical Shifts: The findings suggest that the role of the professor is evolving from a knowledge provider to a facilitator of critical thinking. The practice of assigning students to find errors in AI-generated solutions is a prime example of this shift.
• Future Directions:
  • AI Literacy: There is an urgent need for AI-focused training for both professors and students to foster "AI literacy" specific to physics.
  • Longitudinal Studies: As the technology evolves, future research must track how these frames change over time and across different institutional contexts.
  • Social Dynamics: The authors raise concerns about the potential reduction of human-to-human collaboration if GenAI becomes a constant, non-human discussion partner.

Conclusion: The study concludes that while GenAI is currently being integrated as a set of efficiency tools (small perturbations), its long-term impact may fundamentally alter the epistemic practices of physics. The "Threat" frame serves as a critical guardrail, ensuring that the integration of these tools does not compromise the core cognitive development required to be a physicist.