F(AI)2R: Verifiable AI Provenance as an Executable Skill
A recent publication on arXiv introduces F(AI)2R, a framework designed for the verifiable tracking of AI contributions in research. This system builds upon the initial F(AI)2R experiment, evolving into aiprov, an extension of PROV-O that documents AI inputs across any AI-in-the-loop artifact. It functions as an executable skill managed by the AI agent, which requires the human operator's ORCID ID to verify identity through the public registry. The framework supports continuous integration, ensuring compliance with graph standards for each update and publishing the latest version of the paper. The publication acts as its own case study, documenting all activities, claims, and sources in the provenance graph, adhering to two key principles: no artefacts without parents. This research tackles the issue of insufficient auditable records for AI contributions.
Key facts
- F(AI)2R stands for FAIR research with AI in the loop, twice.
- The framework includes an AI-assisted authoring pass and a machine-readable audit pass.
- aiprov is a PROV-O extension covering any AI-in-the-loop artefact.
- The method is packaged as an executable skill for AI agents.
- Setup requires the human operator's ORCID ID.
- Continuous integration gates every push on graph conformance.
- The paper is its own case study.
- Provenance graph has two invariants: no parentless artefacts.
Entities
Institutions
- arXiv