AI evidence extraction

Governing Epistemic Risk in Large Language Model Use for Art History and Art Criticism Instruction: A Policy Analysis

Yuke Meng · Knowledge Commons (Lakehead University) · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.17613/1j19m-dgp27

Methodology & findings

Study design

Policy-centered analysis drawing on international governance instruments and instruments from UNESCO, OECD, NIST, and the European Union, combined with scholarship review on hallucination, bias, automation, academic integrity, and higher education governance..

Main result

The article argues that "the distinctive epistemic structure of art-historical and critical inquiry makes these subjects vulnerable to a specific cluster of harms: fabricated attributions and references, canon-reinforcing bias, decontextualized summaries, interpretive flattening, and the outsourcing of judgment to systems whose outputs carry an unwarranted aura of authority." The study identifies that existing higher-education policies on generative AI remain overly generic and fail to address discipline-specific epistemic conditions.

Reports effect sizes.

Research paradigm

Critical policy analysis / hermeneutic interpretation

Author conclusions

"The article concludes that policy for LLM use in art education should move beyond permissive tool adoption or blanket prohibition toward an institutional model of epistemic stewardship." The authors propose "a governance framework for art history and art criticism instruction organized around six principles: task classification by epistemic sensitivity; evidence-traceability requirements; assessment redesign to privilege documented reasoning; mandatory disclosure as pedagogical metadata rather than merely as compliance; procurement and tool approval standards for educational uses; and procedural safeguards to protect students from biased detection and opaque sanctions."

Risk of bias

Not applicable—this is a theoretical policy analysis, not an empirical study with participant samples or experimental conditions. No selection bias, attrition, or confounders are relevant.; Selection bias in policy documents analyzed; Geographic/institutional bias toward UNESCO, OECD, NIST, and EU guidance; Potential confirmation bias in identifying harms to humanities disciplines

Open questions raised

The article identifies that the integration of LLMs into higher education has proceeded faster than discipline-sensitive governance development. It specifically identifies a gap in existing policies that frame risk primarily in terms of plagiarism or disclosure while failing to address the epistemic conditions of disciplines organized around contested interpretation, visual evidence, and historically situated argument.
The paper identifies that "the integration of large language models (LLMs) into higher education has proceeded faster than the development of discipline-sensitive governance" and notes this gap "is especially consequential in art history and art criticism instruction, where learning depends not only on factual accuracy but also on source fidelity, context reconstruction, interpretive plurality, and accountable judgment."
The paper identifies a gap between rapid LLM integration into higher education and the development of discipline-sensitive governance. It notes that existing policies fail to address the epistemic conditions of disciplines organized around contested interpretation, visual evidence, and historically situated argument.

Data: not_statedCode: not_statedExtracted from: pdfAgreement 74%

Explore related topics

Institutional Policy AI Governance Policy Analysis