Documenting antisemitism should be a remarkably simple undertaking. Find a public post, take a screenshot, preserve what was said and classify it according to an established definition. Nobody is hacking private accounts, peering through bedroom windows or stealing secret diaries. The material has already been published voluntarily, often proudly, before an audience of strangers.
The trouble begins when you try to preserve it.
The original Jew-hating post may remain online long enough to collect likes, laughing emojis and approving comments. The account that published it may continue operating normally. Yet the moment somebody captures that post and attempts to present it as evidence of antisemitism, automated systems suddenly develop a keen interest in public safety.
Apparently the hatred is acceptable as content, but documenting the hatred is suspicious.
While creating material for ArchivingHate.ca, I asked ChatGPT to produce a graphic explaining that Jew-hatred requires lies. The message condemned antisemitism. Its purpose was unmistakably educational, and the requested wording did not threaten, encourage or celebrate hatred. Nevertheless, the image request was rejected by the safety system.
The machine recognized certain words but failed to understand why they were being used.
Reminds me of a poem I once read:
“First they came for the porn addicts who pasted their neighbours’ faces onto naked bodies, and I said nothing—because those creepy fuckers had it coming.
Then they came for the propagandists manufacturing hateful images, and I said nothing—because those lying fuckers had it coming too.
Then they came for the people documenting Jew-hatred, because the machine could recognize the words but couldn’t understand why they were being shown.”
—Martin Screenshöller
Apparently artificial intelligence can identify suspicious material. It just cannot tell the arsonist from the fellow pointing at the fire.
The absurdly comical part is that thelsighting it is pure comedy. The serious part is that these failures always impose their burden on somebody. When a system cannot distinguish antisemitic propaganda from documentation opposing antisemitism, the result favours the person publishing the hatred. The propagandist gets an audience while the archiv archivist gets a rejection notice.
That is canny be described as antisemitism by declared intent. Nobody needs to program a machine to dislike Jews or announce an antisemitic policy. It is antisemitism by neglect: systems are built, deployed and defended without sufficient concern for whether they obstruct people documenting Jew-hatred.
Institutional indifference provides everyone with a clean pair of hands. The hater says he was only criticizing Israel. The platform says it merely hosted user content. The filter says nothing because it is a machine. The company says the system worked according to policy.
Meanwhile, the evidence is blocked and everyone claims innocence.
Antisemites already depend upon confusion,lk, denial and disappearing evidence. They post something hateful, receive the reaction they wanted and then delete it when scrutiny arrives. If confronted, they explain that they were only criticizing Israel, only opposing Zionism, only referring to certain Jews or merely repeating something they heard elsewhere. The original statement vanishes and the alibi takes its place.
That is precisely why screenshots matter.
A screenshot preserves the moment multiline before the writes retreat. It records what was actually written, how it was presented and, where possible, who published it. It allows the statement to be considered against the IHRA Working Definition of Antisemitism instead of against the author’s revised explanation after being caught.
Unfortunately, the same systems that freely carry the original hatred can obstruct the evidence documenting it. The result does not become harmless merely because it arose from automation, careless policy or corporate indifference. If a system repeatedly protects the circulation of hatred while hindering those who document it, then its effect deserves to be named, examined and corrected.
THE HATERS POSTED IT.
THE PLATFORM HOSTED IT.
THE FILTER BLOCKED THE EVIDENCE.
EVERYBODY CLAIMED INNOCENCE.
The excuse may change. The screenshot won’t.
The post may disappear. The evidence won’t.
ArchivingHate.ca
- AI BATTLE.png (2.88 MiB) Viewed 24 times