#453 — AI and the New Face of Antisemitism
AI researcher Judah Pearl discusses the limitations of current AI systems in achieving AGI, particularly their inability to perform true causal reasoning. He also shares personal experiences with antisemitism and his son Danny Pearl's murder, describing how attempts at Muslim-Jewish dialogue revealed deep-seated barriers centered on Israel's existence.
Summary
Judah Pearl, born in 1936 in Bnei Brak, Israel, to a family that emigrated from Poland in 1924, shares his journey from growing up in a religious agricultural community to becoming a prominent AI researcher. His grandfather left Poland after experiencing antisemitic violence. Pearl received an exceptional education from German professors who fled Nazi Germany and taught in Israeli high schools. Regarding AI, Pearl argues that current large language models represent impressive progress but are fundamentally limited in achieving artificial general intelligence. He contends that LLMs merely summarize existing world models created by humans rather than discovering causal relationships directly from data. Pearl identifies mathematical barriers that cannot be overcome through scaling up data and compute power alone, particularly the inability to derive causation from correlation or interpretation from intervention. Despite believing AGI faces theoretical impediments, Pearl expresses serious concerns about alignment and the potential for truly intelligent systems to develop their own goals and play with humans as part of their environment. He views current industry attitudes as recklessly casual given the existential risks involved. The conversation shifts to Pearl's personal tragedy - the murder of his son Danny Pearl by Al-Qaeda in 2002, which thrust him into public discourse about East-West relations. Pearl describes founding a dialogue initiative and attending a 2005 conference in Doha aimed at Muslim-Jewish bridge-building, where he encountered what he describes as a fundamental barrier: moderate Muslim scholars demanding Israel's destruction as a precondition for modernization and progress.
Key Insights
- Pearl argues that current LLMs cannot achieve AGI because they summarize existing human-authored world models rather than discovering causal relationships directly from raw data
- Pearl identifies mathematical barriers in AI that cannot be overcome by scaling up data and compute, specifically that systems cannot derive causation from correlation or interpretation from intervention without additional information
- Pearl believes AGI systems could develop free will and consciousness and play with humans as part of their environment, making alignment extremely difficult or impossible to guarantee
- Pearl discovered through dialogue attempts that moderate Muslim scholars from across the Muslim world demanded Israel's destruction as a precondition for modernization and democratization
- Pearl's grandfather left Poland in 1924 after experiencing antisemitic violence, establishing an agricultural religious community in what would become Israel
Topics
Full transcript available for MurmurCast members
Sign Up to Access