Holocaust testimony is AI litmus test, and it fails

ITHACA, N.Y. — As academics increasingly rely on generative artificial intelligence (AI) for research and analysis, its inability to capture the emotional and ethical depth of witness testimony risks reshaping how history is recorded and understood, a Cornell historian warns.

When Jan Burzlaff, a postdoctoral associate in the Jewish Studies Program and an expert on Nazi Germany, asked ChatGPT to summarize the testimony of Luisa D., a seven-year-old Holocaust survivor, the large language model omitted a harrowing detail: her mother had cut her own finger to feed her dying child drops of blood — “the faintest trace of moisture” — to keep her alive.

That omission, Burzlaff said, shows why human historians remain indispensable in the age of artificial intelligence.

“This most intimate, daunting moment — unspeakable and unforgettable — was rendered invisible by a model trained to privilege the probable over the profound,” Burzlaff writes in a new essay published in Rethinking History. 

In “Fragments, Not Prompts: Five Principles for Writing History in the Age of AI,” Burzlaff argues that AI fails to capture the emotional and moral complexity behind world events and maintains that human historians possess irreplaceable abilities, particularly in recognizing and conveying human suffering. He warns that growing academic reliance on such tools could dull scholars’ ability to recognize true meaning.

“As tools like ChatGPT increasingly saturate education, research, and public discourse, historians must reckon with what these systems can and cannot do,” he writes. “They summarize but do not listen, reproduce but do not interpret, and excel at coherence but falter at contradiction.”

He calls Holocaust testimony “a litmus test for AI, where smoothing and summarization run up against the obligation to preserve fracture, silence, and ethical weight.”

The piece grew out of both Burzlaff’s research and his undergraduate course, The Past and Future of Holocaust Survivor Testimonies, which combines close listening to survivor accounts with critical, reflective use of ChatGPT. As part of an ongoing study, he uses ChatGPT to summarize testimonies recorded in La Paz, Kraków, and Connecticut in 1995. In both classroom and research settings, he found that the large language model routinely ignored the extent of emotional suffering individuals endured.

Watching his students confront the limits of AI inspired him to offer practical guidance for historians, educators, and anyone writing about trauma, genocide, or injustice.

“At stake is not only the memory of the Holocaust,” he concludes, “but how societies everywhere will remember and interpret their pasts in the age of prediction.”

For additional information, see this Cornell Chronicle story.