When artificial intelligence gives us a disturbing answer about Israel, Zionism or Jews, the natural reaction is to blame the machine.
But perhaps we are looking too far downstream. Before asking why AI learned something, we should ask how that something became accepted knowledge in the first place.
That is the unsettling implication of Ashley Rindsberg’s recent essay in SAPIR. His immediate subject is the manipulation of Wikipedia and other online platforms by activists determined to reshape the narrative about Israel. But the larger issue is more consequential: Who decides what counts as knowledge in the digital age?
Wikipedia once was simply an encyclopedia. You looked something up, read the entry and decided whether to trust it. Today, Wikipedia, Reddit and other enormous repositories occupy important places in a sprawling information ecosystem. Their material is quoted, linked, summarized, indexed, repeated and absorbed elsewhere. What happens there does not necessarily stay there.
Rindsberg describes how a relatively small group of highly active Wikipedia editors targeted entries concerning Zionism, Israel and the Palestinians. The lead definition of Zionism, for example, was changed to describe establishment of a Jewish homeland “through colonization” and to assert that Zionists sought as much land and as few Palestinian Arabs as possible. The revised language was then effectively locked in place for a year.
That is troubling. But Wikipedia is not really the story. The story is what happens next.
A disputed assertion gets published. It gets repeated. Repetition produces citations. Citations create apparent authority. Search engines reproduce it. Journalists, students and other writers encounter it. Other websites rely upon it. Somewhere along that journey, an argument can stop looking like an argument and begin looking like a fact.
Artificial intelligence raises the stakes because it sits at the end of that supply chain. It synthesizes information gathered from an enormous digital environment and delivers an answer, often without exposing the messy arguments, editorial decisions and ideological battles through which the underlying information acquired its authority.
That changes the nature of the problem. Traditional propaganda tried to persuade us. We knew someone was making an argument, and we could accept it, reject it or investigate it. The emerging danger is more subtle. Manipulation can occur upstream, long before most of us encounter the resulting claim.
By the time an AI system confidently tells a student what Zionism means or whether Israel is a colonial enterprise, the consequential battle may already have occurred somewhere the student will never see.
This does not mean Wikipedia is inherently unreliable or every criticism of Israel is the product of manipulation. Nor should anyone appoint an authority to determine what constitutes acceptable knowledge. The challenge is harder.
Jewish organizations have concentrated enormous energy on combating misinformation after it appears: correcting articles, protesting posts, answering accusations and demanding retractions. That work remains important. But we also need to understand the supply chain of knowledge — where claims originate, who establishes their credibility, how they migrate and when repetition begins masquerading as authority.
AI did not create that problem. It has simply made the consequences much greater.
Before worrying only about what the machine is saying, we had better start paying much closer attention to who taught it.
