Proposing a Classification Scheme for AI Usage in Scientific Articles
Year: 2026
Type: preprint
Abstract: Journals and preprint servers increasingly ask authors to disclose their use of artificial intelligence (AI), but such disclosures are almost always free text that varies from journal to journal and cannot be searched or compared. We propose AIL2026 (AI Labels), a compact, machine-readable classification of AI usage. Like the CRediT taxonomy for human contributors, it records who contributed what; like MSC2020, it uses short standardized codes, but with plain-word labels that humans can read at a glance. Each paper carries a tag: a headline level from AI-0 to AI-4 summarizing AI's role in the research itself, followed by labels stating where AI contributed and how much, each with a depth from 0 to 4. Examples are AI-0: Text1 Check2 and AI-3: Proof3 Text1. The same labels apply to every arXiv category, with field-specific interpretations. AI can contribute to research in very different ways, from light uses such as polishing text or double-checking signs and prefactors to substantial contributions such as suggesting research directions or proving theorems. We argue that light use and AI verification of any depth benefit everyone and should be encouraged: polishing makes papers easier to read, and checks with Python or Lean add a further layer of verification. Normalizing light use, which is already widespread but often undisclosed, also lowers the barrier to disclosing heavier use, since transparency, not the level of use, is what signals scientific integrity. Where AI contributes essential parts of the research, disclosure is required by good scientific practice: undisclosed use at that level confers an unfair advantage. We discuss design principles, edge cases, incentives, and a route to adoption beginning with a pilot in mathematics. 17 pages; AI usage (AIL2026): AI-3: Idea3 Result3 Proof3 Text3 Lit2 Figs3 Check2 [Claude Opus 5.5, 2026-10] (details in the section "AI usage of this paper"); MSC2020: 00A35 (primary), 68T01 (secondary)
Author: Prinz, David
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7222567945
Cites:
Cited by:
Related to: