How I use artificial intelligence in my academic work

I use artificial intelligence tools in my research and teaching, and I think transparency about that use is now part of scholarly honesty. This page sets out how I use them, and where I do not.

Principles

The intellectual responsibility for my work is mine. AI tools assist; they do not author. Every legal proposition, every citation, and every argument that appears under my name has been verified by me against primary sources. A model's output is a starting point to be checked, never an authority to be trusted.

I treat these tools the way I treat a research assistant or a well-read colleague: useful for drafting, organising, and challenging ideas, but never a substitute for my own reading of the sources or my own judgment about the law.

Where I use AI

• Structuring and drafting. Organising notes, producing first drafts of expository passages, and reformulating my own text for clarity, which I then rewrite in my own voice.
• Language. Working across French, English, and German, I use these tools to check phrasing and to move between languages, always keeping control of legal terminology, which is where translation most often goes wrong.
• Exploration and synthesis. Getting a fast orientation in an unfamiliar area, summarising long documents, and surfacing arguments or counterarguments I should consider, all of which I then confirm against the actual materials.
• Administrative and planning work. Managing schedules, correspondence, and the documentation that surrounds research stays and institutional roles, so that more of my time goes to substantive work.

Where I do not use AI, or use it only with care

• I do not delegate legal reasoning. The analysis, the choice of method, and the conclusions are my own.
• I do not rely on AI for citations or authorities. Models fabricate references convincingly; I verify every source directly.
• I do not enter confidential, unpublished, or third-party material into tools without regard for its confidentiality, particularly work entrusted to me as an expert or as a supervisor.
• I do not present AI-generated text as finished scholarship. Anything published under my name has passed through my own writing and checking.

In teaching and supervision

I encourage my students and doctoral candidates to understand these tools rather than avoid them, and to use them within the same discipline I apply to myself: as an aid to thinking, held to the standard of verification against primary sources, and always transparent. Learning to use AI well, and to know its limits, is now part of legal training.

Environmental cost

These tools are not immaterial. Running them consumes energy and water, and I take the view that someone who uses them regularly should account for that rather than treat it as someone else's externality.

I therefore estimate the greenhouse gas emissions and the freshwater consumption attributable to my own use, and compensate both. The estimates are deliberately ranges rather than single figures: providers do not publish per-query emissions or water use, and my usage dashboard reports the share of my quota consumed rather than the number of queries, so a precise number is not available to me. I bound them from independent published research and review them weekly.

I compensate through monthly donations, adjusted twice a year: to the Fondation GoodPlanet (Action Carbone Solidaire) for carbon, and to Solidarités International, whose work is drinking water and sanitation, for water. These are donations to public-interest organisations rather than purchases of offset credits. That choice carries a cost I accept: a donation comes with no certified tonnage, precisely because French law does not permit a genuine gift to carry a quantified quid pro quo. I would rather fund the work and forgo the certificate than buy the certificate.

Two caveats I would rather state than be asked about. Compensation does not undo an emission: a tonne compensated is not a tonne avoided, and I treat it as a second-best to using these tools deliberately rather than reflexively. And the order of magnitude matters: my use of AI is small next to my academic travel. The same logic applies there: I take the train when I can, and would rather take part online, though not every event lends itself to it.

Why this matters

Private international law and the law of emerging domains, cyberspace and space among them, are themselves being reshaped by these technologies. I consider it consistent to be open about using, carefully and critically, the very tools whose legal governance is part of my research.