Human–ai Knowledge Provenance
Year: 2026
Type: preprint
Abstract: The increasing integration of generative artificial intelligence (AI) into scientific research, academic writing, and knowledge-intensive professional activities creates a fundamental problem of epistemic transparency: how can the origins of ideas, arguments, and conclusions be reconstructed when humans and AI systems participate in successive stages of knowledge production? Conventional authorship declarations and general statements of AI use provide limited information about the specific intellectual contributions, transformations, and verification procedures underlying individual scientific claims. Although established provenance standards and recent AI documentation frameworks offer important foundations, practical methods are needed to connect individual claims with their observable histories of generation, revision, evidential support, and human evaluation. This article proposes the Claim-Level Human–AI Knowledge Provenance (CL-HAKP) framework, a conceptual and methodological approach to documenting knowledge development in human–AI collaboration. The framework distinguishes the observable origin of a contribution, its subsequent transformation, its evidential justification, and the human responsibility associated with its acceptance into a final knowledge product. It introduces a structured claim-level provenance record, describes relationships between human contributions, AI-generated alternatives, external evidence, and final conclusions, and presents a proportionate documentation protocol compatible with established provenance and research transparency standards. A hypothetical scientific writing scenario demonstrates how the framework can distinguish an independently formulated human hypothesis from an AI-suggested modification and a subsequently verified scientific claim. The article also proposes an empirical evaluation strategy addressing traceability, attribution agreement, documentation burden, and error correction. The framework does not claim to establish the hidden internal origins of model-generated content or to determine legal authorship automatically. The central argument is that meaningful transparency in human–AI knowledge production requires more than identifying which tools were used. It requires preserving a sufficiently informative record of how particular claims emerged, changed, were evaluated, and became part of the final intellectual product.
Author: Shaban, Roman
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7232933547
