AI Disclosure Requirements
PHYS 4321/4322: Advanced Laboratory I/II
1 Overview
Based on American Physical Society (APS) Physical Review journals (American Physical Society 2026b, 2026a), we follow a similar policy on AI tools. The idea is simple: if you use AI tools, you must say how, and you are responsible for everything in your report.
This page covers generative AI tools such as ChatGPT, Claude, Gemini, Copilot, and similar systems, including AI features built into editors and coding environments.
AI use is entirely optional. You are never required to use generative AI in this course.
IBM training slide, 1979 (IBM 1979)
Relying on AI is not free. There is growing evidence that it can erode the skills it replaces:
“The ADR of standard colonoscopy decreased significantly from 28.4% (226 of 795) before to 22.4% (145 of 648) after exposure to AI, corresponding with an absolute difference of -6.0% (95% CI -10.5 to -1.6; p=0.0089). […] Continuous exposure to AI might reduce the ADR of standard non-AI assisted colonoscopy, suggesting a negative effect on endoscopist behaviour.” (Budzyń et al. 2025)
Writing a report, fitting your own data, and propagating your own uncertainties are the skills this course exists to build. Improper use of AI leaves you worse off, rather than better.
2 The Principles
- You are accountable. An AI program cannot be held accountable (IBM 1979), so it cannot be an author. You are responsible for the accuracy of every sentence, number, equation, figure, and reference in your report, whether or not AI helped produce it.
- Check everything. Review and correct all AI output before it goes into your report. AI tools confidently produce wrong derivations, wrong code, and references that do not exist.
- Disclose substantive use. If AI shaped the science or the content of your report, say so in the report.
- Keep a record. Note your AI use as you go, the same way you record your procedure in your lab notebook. Save chat links or transcripts until your report is graded.
3 When Must I Disclose?
We draw a line between light editing and substantive use.
No disclosure needed: using AI to polish, condense, or lightly edit text you wrote. This means spelling, grammar, punctuation, word choice, and trimming a sentence that runs long. The meaning, the claims, and the structure stay yours. If the AI rewrote a paragraph into something you would not have written, that is no longer light editing.
Disclosure required: any use that affects the content or the results. For a lab report, that includes:
- Writing or debugging code that produces your fits, numbers, or plots
- Performing or checking derivations, calculations, or uncertainty propagation
- Generating or restyling figures, schematics, or diagrams
- Finding, summarizing, or synthesizing literature for your Introduction or Theory
- Interpreting your results or suggesting explanations for a discrepancy
- Drafting or restructuring text (see What Is Not Permitted for the limits)
- Translating text from another language
A disclosure costs you one sentence.
4 Where Does the Disclosure Go?
The disclosure goes inside your report, in the section closest to where the AI was used. It is not a separate document.
| What AI did | Where you disclose it |
|---|---|
| Generated or restyled a figure | That figure’s caption |
| Wrote or debugged analysis code; did or checked calculations, fits, or uncertainty propagation | Experimental methods or Data and analysis |
| Literature search, summaries, drafting help, feedback on organization, translation | Acknowledgments |
If needed, add a short Acknowledgments section after the Conclusion and where you can collect everything before the Appendices and References.
5 What Goes in the Disclosure?
Every disclosure answers four questions:
| Required item | What to include |
|---|---|
| 1. Tool and version | The tool’s name and version or model, e.g. “ChatGPT (GPT-5)”, “Claude Sonnet 5”, “GitHub Copilot in VS Code”. |
| 2. How it assisted | What the AI did: wrote a fitting function, checked a derivation, suggested a figure layout. |
| 3. How you directed it | What you asked it to do, and what you gave it (your code, your equations, your draft). A one-line summary of your prompt is enough. |
| 4. How you verified the output | How you know the result is right: compared to a hand calculation, tested the code on simulated data with known parameters, checked each reference in the library. |
Item 4 is the one that matters most. “I read it over” is not verification.
6 Examples
Figure caption:
Fig. 3. Normalized counting rate versus absorber thickness for aluminum and lead, with weighted exponential fits. The plotting script was drafted with Claude Sonnet 5 from a description of the desired layout; I checked the plotted points against the data table and the fit curves against the parameters in Table 2.
Data and analysis:
The least-squares fit to Eq. (4) used
scipy.optimize.curve_fit; see Appendix A for code. GitHub Copilot (VS Code, September 2026) suggested the fitting function and the conversion of the covariance matrix to parameter uncertainties. I verified the code by comparison with the least-squares formulas in Ref. (Weisstein, n.d.).
Acknowledgments:
I thank my lab partner, J. Doe, for help collecting the data. AI disclosure: I used ChatGPT (GPT-5) to find review articles on the Zeeman effect; I read the two sources [3, 4] that I cited. I also asked it for feedback on the organization of my Discussion, which led me to extend my discussion of systematic error. All text is my own.
7 What Is Not Permitted
While APS now allows authors to use AI substantively, with disclosure, this course must assess your understanding. The following are not permitted, disclosed or not:
- Generating, altering, or “filling in” data. Every data point must come from measurements you and your partner took in the lab this semester.
- Having AI write your report or its sections. Your Abstract, Discussion, and Conclusion must be written by you. AI may give feedback on your draft; it may not produce the draft.
- Citing sources you have not read. AI tools invent references. Every citation must point to a real source that you have looked at and that says what you claim it says.
- Presenting AI-generated images as real. A schematic of your apparatus may be AI-assisted and disclosed; it may never be passed off as a photograph.
- Using AI substantively without disclosing it.
Undisclosed substantive AI use, or any of the uses above, may be treated as a violation of Georgia Tech’s academic integrity policies.
8 Why This Matters
This policy is not meant to discourage legitimate use of useful tools. Working physicists use them, and APS (and other journals collections) acknowledges that. This is meant to keep your reports honest about how they were made, and to make sure the physics in them is yours.
If you are not sure whether a use is allowed or needs disclosing, ask before you submit.
9 Acknowledgements
This webpage was first drafted by Claude Opus 5.5 based on a prior AI policy I had written for a different course. I then edited and reworked the generated draft to my liking except for these Acknowledgements — which I wrote by hand — em dashes and all.