The article presents an experiment in which a historian engages ChatGPT 4.0 Plus to assess the capabilities and limitations of generative AI in historical research, specifically regarding its adherence to foundational methodological principles. The introductory section sets out the rationale for the test: determining whether AI goes beyond probabilistic information synthesis towards genuine factual verification, particularly in the absence of explicit training or supplied source material. Through a series of questions about Victor Hugo’s purported opinions on ethnographic exhibitions at the 1867 Paris Exposition, the researcher identifies the AI’s tendency to plausibly fabricate sources—most notably a letter from Hugo to Mary Weston Chapman, for which no verifiable historical basis exists. The Q&A section reveals how ChatGPT, when pressed for details and documentary evidence, shifts its narrative and eventually admits its inability to provide genuine sources, disclosing its reliance on probabilistic reasoning rather than strict historical rigor. The conclusions underscore the risks of relying on AI for historical inquiry without critical oversight, emphasizing the essential role of human researchers in verifying evidence and maintaining methodological integrity. The article advocates for transparency about uncertainty in AI-generated outputs and cautions against the uncritical adoption of AI tools in historical scholarship, suggesting that specialized or custom-trained AI models could enhance reliability but still require vigilant human supervision.
