Tel Aviv University Study Examines AI Limits in Decoding Holocaust Archives
A study by Tel Aviv University researcher Dr. Yael Hacohen explores how artificial intelligence struggles to preserve material context and meaning when digitizing fragile Holocaust testimonies.
A new study published in the journal AI & Society explores the complex challenges that arise when artificial intelligence attempts to process and digitize fragile historical testimonies from the Holocaust. Conducted by Dr. Yael Hacohen from the Department of English and American Studies at Tel Aviv University, the research investigates what is lost when computational tools successfully decode words but strip away the crucial material context of the original artifacts.
Three Unique Archival Case Studies
The research is based on three detailed case studies selected for their material complexity rather than as a representative statistical sample. The first case involves two clay tablets engraved in Hebrew by writer Eliezer Heyman while hiding near Kaunas, featuring family names and biblical language. Automated tools struggled to distinguish between engraved letters, peeling material, shadows, and soot marks—burn signs that physically tie the object to the destruction of the hiding place.
The second case examines an illustrated poem written by a young girl, Eva Schulzova, in the Theresienstadt ghetto shortly before her deportation to Auschwitz and murder. Faint pencil strokes proved exceptionally difficult for the software to identify, and the digital conversion severed the visual link between the poem and its accompanying drawings. The third case features a piece of fabric where a young Amsterdam resident, Sal Barmson, documented his prison cell through drawings and notes smuggled out in a laundry collar.
Preserving Context in the Age of AI
Dr. Hacohen concludes that responsible digital archiving must maintain the direct link between transcriptions and original physical objects, meticulously documenting alterations and uncertain readings. The research emphasizes that clearer text does not automatically equate to a more complete historical testimony.
"Artificial intelligence can open up access to Holocaust testimonies for new generations, but our responsibility does not end with decoding words," Dr. Hacohen stated. "When someone carves Hebrew into clay or a child draws an illustration alongside a ghetto poem, the material and physical damage are part of the message."
The study calls for the future development of AI-driven archival tools capable of factoring in material context, ensuring that authentic human voices continue to resonate clearly across generations despite technological mediation.
