AI
AI Disclosure Standards in Ethics Research Need Process Transparency, Not Just Reporting
A new framework proposes that ethics researchers using AI should transparently document their entire process—not just declare that AI was used—to preserve academic integrity and allow future evaluation.
Current AI disclosure policies in academic publishing require researchers to report when they've used AI tools, but leave a critical gap unfilled: they don't explain what that transparency is actually for. A new arXiv paper by Michele Loi argues this matters especially in ethics research, where the stakes for methodological clarity are highest.
The Problem with Output-Only Transparency#
Existing disclosure mandates work like checkboxes—you use AI, you report it, done. But this approach fails when applied to philosophical and ethical inquiry, Loi argues. Ethics research is "essentially contested" at two levels: people disagree both about what ethical questions are worth asking and about what standards the inquirer themselves should meet. This means you can't evaluate the quality of an ethics paper just by looking at what it concludes. You need to understand how the author thinks, what commitments guide their inquiry, and what role AI played in shaping that process.
Agent-Integrity as the Core Issue#
The paper grounds its argument in "agent-integrity"—the idea that readers need legibility into an author's identity-constituting commitments and intellectual voice. When AI participates in research, it can obscure whose thinking is actually guiding the work. Traditional reproducibility frameworks borrowed from empirical science don't address this; they focus on whether findings can be replicated, not whether readers can understand the author's philosophical stance.
A Documentation-Adequacy Framework#
Loi proposes a five-part transparency standard:
- Declaration: stating that AI was used
- Navigation: explaining where and how it was applied
- Documentation account: describing what was documented
- Process documentation: recording the actual workflow
- Development records: archiving the full record for later review
Rather than imposing a single evaluation standard, the framework aims at "tracking"—creating a comprehensive evidentiary record that lets different philosophical traditions assess the work according to their own criteria. This operationalizes Meaningful Human Control without assuming settled agreement on what good ethics research looks like.
