AI evidence extraction

Generative AI as a historical source: source criticism, citation integrity, and the jagged frontier of digital history

Iryna A. Selyshcheva · CTE Workshop Proceedings · 2026

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

9/10

Relevance

0/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.55056/cte.1438

Methodology & findings

Study design

Narrative review synthesizing evidence across multiple domains including peer-reviewed evaluations, professional-society guidance, primary legal filings, documented failure cases, document analysis, public history, and pedagogy.

Main result

The article develops three connected claims: First, "generative models are best understood as an algorithmic cartography of the digitised record whose jagged frontier of competence maps which pasts have been absorbed into training data and which remain silent." Second, "the same architecture that enables transcription of damaged manuscripts and large-scale corpus analysis also produces hallucinations and fabricated citations at rates incompatible with the evidentiary standards of the discipline." Third, "the responsible integration of these tools depends on extending traditional source criticism to the model, on non-negotiable verification of every reference, on cryptographic provenance and Indigenous data-governance frameworks, and on assessment redesign rather than prohibition."

Research paradigm

Hermeneutic/Critical interpretive

Author conclusions

The article concludes that "the responsible integration of these tools depends on extending traditional source criticism to the model, on non-negotiable verification of every reference, on cryptographic provenance and Indigenous data-governance frameworks, and on assessment redesign rather than prohibition." The authors propose a framework for "a critically literate, symbiotic historical scholarship" that acknowledges both capabilities and limitations of generative AI in historical practice.

Risk of bias

Selection bias in which failure cases and audits are cited (may overrepresent high-profile cases); Potential confirmation bias in reviewing peer-reviewed evaluations critical of AI; Limited discussion of positive applications or successful integrations with appropriate safeguards

Open questions raised

The paper identifies the need for: (1) frameworks extending traditional source criticism to LLMs themselves; (2) systematic verification protocols for AI-generated references; (3) integration of cryptographic provenance tracking; (4) application of Indigenous data-governance frameworks to AI tool development; (5) redesign of historical assessment methods rather than outright prohibition of these tools.
The paper identifies the need for extending source criticism methodologies to large language models, developing verification protocols for AI-generated references, implementing cryptographic provenance systems, incorporating Indigenous data-governance frameworks, and redesigning assessment methods for historical scholarship in the generative AI era.

Data: not_statedCode: not_statedExtracted from: pdfAgreement 71%

Explore related topics

Citation Integrity Research Integrity Literature Review