AI processing may obscure material context in Holocaust archives

Yael S. Hacohen, a researcher at Tel Aviv University, studied how artificial intelligence and computational tools change the interpretation of Holocaust artifacts.

Her research, published in AI & Society, examines the concept of material uncertainty. This occurs when the meaning of a historical text is tied to its damaged or improvised physical form, such as cloth or clay. Hacohen tracked the digital processing of three specific artifacts.

In the case of Sal Bramson, a prisoner in an Amsterdam jail, the software simplified the fabric's weave to make text clearer but removed the textile's role as a means of secret communication. For a poem by Eva Schulzová, a child in Terezín, the AI generated a description of an imagined meal as a still life, failing to capture the emotional contrast between the child's hunger and the depicted food.

For clay tablets by Eliezer Heiman, the technology struggled to distinguish between engraved Hebrew letters and smoke damage. Hacohen argues that these digital workflows can create a polished version of history that hides the physical struggle of the original creators. She suggests that archives should link transcriptions directly to original images and clearly mark where AI has inferred or corrected information to preserve the limits of what can be known.