AI Use in My Research
I use AI tools as a routine part of my research, and I believe in being open about it. This page describes what those tools do in my workflow and, just as importantly, how I stay fully responsible for everything that carries my name. It follows the spirit of the American Physical Society’s policy on the appropriate use of AI tools, updated in June 2026, which allows substantive AI assistance in preparing manuscripts as long as the use is disclosed and a human author remains accountable.
My one governing rule is simple: AI assists, I decide. An AI tool is never an author, because it cannot take responsibility for a result. Every physics claim, every number, and every figure that appears in my papers is verified by me against an independent reference (an analytic limit, a published benchmark, a separate code, or a hand calculation) before it is trusted. A good fit with a wrong physics picture is worthless to me; the physics argument comes first, and the tools only help me build and check it faster.
How I keep AI-assisted work accountable
Wherever I use AI tools, the output has to clear a fixed standard before it can enter my research. The standard is the point; the tool only helps me reach it faster, and it never gets the final word.
| Task area | My verification standard | Supporting role of AI |
|---|---|---|
| Literature | I read the primary sources myself, and every citation is checked against the original paper before it enters a manuscript. | Searching, summarizing, and cross-linking what I read into a personal knowledge base. |
| Code and numerics | No code is trusted until it reproduces a known reference (a legacy result, a published value, or an analytic limit) to stated precision; I review code before it runs on shared resources. | Drafting, refactoring, and debugging; explaining unfamiliar numerical methods. |
| Figures | The data and the analysis are mine; I check every figure for correctness and re-plot until it faithfully represents the result. | Generating figures from my own data and plotting specifications. |
| Writing and language | I write the physics and the central argument, and I read and own every sentence; the science is never edited by a tool, only the expression. | Tightening prose, improving clarity, and translation. |
| Cross-checking | I treat a single-model answer as unverified, synthesize across independent checks, and make the final scientific judgment myself. | Running more than one independent model on the same problem so they catch each other’s errors. |
Confidentiality
When I act as a referee, I do not share manuscripts, reviewer reports, or any other confidential peer-review material with AI tools that do not guarantee data protection. Reviewing is a human responsibility, and the confidentiality of the work I am entrusted with comes before any convenience a tool might offer.
How I disclose
I disclose my use of AI tools to the editors in the cover letter of every submission, stating which tasks the tools assisted, how I verified the output, and that I take full responsibility for the manuscript. When a paper warrants a note in the published text, I place it according to what the tools did: computational, data, or numerical assistance in the methods description, and language assistance in the acknowledgments. An AI tool is never listed as an author, because it cannot take responsibility for the work.
A representative disclosure reads:
During the preparation of this work, the author used AI-based tools (large language model assistants) to support literature search and synthesis, code development and debugging, figure generation from author-produced data, and language editing. All physics content, derivations, and numerical results were verified by the author against independent references, and the author takes full responsibility for the contents of the manuscript. No AI tool is listed as an author.