A student raises her hand and asks a question into the microphone.
asserted
student → raise → microphone
The woman on the screen in front of them pauses, blinks, and answers.
asserted
woman → pause → them
The classroom falls quiet as Holocaust survivor Sonia Warshawski describes what she lived through in her own words.
asserted
she → fall → words
For a moment, it feels as if she is sitting with the students.
asserted
she → feel → students
I have watched these exchanges through The Blue Card’s interactive hologram program.
asserted
I → watch → program
What students experience is built around Warshawski’s recorded testimony, with technology that retrieves her answers instead of creating new ones.
asserted
that → build → ones
If Warshawski answered a question during the recording process, the system plays her response.
asserted
system → answer → response
The artificial intelligence uses keywords from the question asked to find the closest answer.
asserted
intelligence → use → answer
The AI does not create new answers by splicing or picking parts out of other recordings.
asserted
AI → create → recordings
It can only play a single prerecorded video response that most closely connects to the keywords in the question.
asserted
that → play → question
If she did not, it pauses, protecting the difference between documented testimony and a machine-generated guess.
asserted
it → do → testimony
AI will help shape how the Holocaust is remembered long after the last survivors are gone.
asserted
survivors → help → ?
We should embrace its ability to extend the reach of authenticated testimony, while drawing a firm boundary around the truth: No system should invent a survivor’s words or present generated history as fact.
asserted
system → embrace → fact
Technology companies and the institutions responsible for teaching and preserving Holocaust history need to build that boundary into these tools now, before chatbots become one of the main ways a generation encounters the past.
asserted
generation → teach → past
That possibility is approaching quickly.
asserted
possibility → approach → ?
More than half of American teenagers have used chatbots to search for information or get help with schoolwork.
asserted
half → use → schoolwork
Meanwhile, the Claims Conference estimates that only 31,000 Jewish Holocaust survivors remain in America, with a median age of 87.
uncertain
survivors → estimate → 87
Young people are turning toward AI at the same time that the generation able to answer their questions from lived experience is disappearing.
asserted
generation → turn → experience
Generative AI can make archives searchable, translate testimony, and bring history to students who may never enter a museum.
uncertain
who → make → museum
It can also produce language that sounds authoritative even when it is wrong, stripped of context, or entirely invented.
asserted
it → produce → context
Holocaust denial has always depended on repeating falsehoods until they begin to sound credible.
asserted
they → depend → falsehoods
AI can give that old strategy enormous reach and a polished appearance.
asserted
AI → give → reach
In January, the Anti-Defamation League tested six leading AI models through more than 25,000 conversations and found gaps in every model’s ability to detect and counter antisemitic or extremist narratives.
asserted
League → test → narratives
A second ADL study released in July found that four major chatbots were consistently less effective at recognizing and rejecting antisemitism in Persian than in English.
asserted
chatbots → release → English
These failures affect what people learn, which sources they trust, and whether conspiracy theories receive a new appearance of authority.
asserted
theories → affect → authority
AI has a place in Holocaust education when it expands access without altering the historical record.
asserted
it → have → record
Answers should be grounded in authenticated testimony and primary sources, with clear links showing users where the information came from.
asserted
information → ground → where
When the record does not support an answer, the system should say so.
asserted
system → support → answer
Models should also be tested in multiple languages against Holocaust denial and false quotations, including antisemitic conspiracy claims, before they reach classrooms.
uncertain
they → test → classrooms
I work with Holocaust survivors every day through The Blue Card.
asserted
I → work → Card
I know how carefully they choose their words and what it can cost them to return to these memories.
asserted
it → know → memories
They are bearing witness to what they lived, not offering material for someone else to complete or reinterpret.
asserted
someone → bear → material
The Innovative Holocaust Education Program respects that choice by allowing her testimony to reach new audiences without her having to relive these moments over again, while keeping her own words at the center.
asserted
her → respect → center
Technology companies should work with historians and organizations that educate about and serve Holocaust survivors before these systems reach the public.
asserted
systems → work → public
Jewish institutions should invest in verified archives and the expertise needed to evaluate new tools responsibly.
asserted
institutions → invest → tools
Boards and funders should treat AI as a question of memory and education, not simply an information technology purchase.
asserted
Boards → treat → memory
Accuracy has to be part of the design from the beginning.
asserted
Accuracy → have → beginning
A teenager can now ask a chatbot about Auschwitz before opening a textbook, speaking with a teacher, or visiting a museum.
asserted
teenager → ask → museum
The answer may shape what that student believes is true.
uncertain
student → shape → ?
The evidence exists, but the system will influence how much of it the student sees and whether fact is clearly separated from invention.
asserted
fact → exist → invention
…and 2 more, not listed.
The Aporia
