Summary of #453 — AI and the New Face of Antisemitism

Overview of #453 — AI and the New Face of Antisemitism (Making Sense with Sam Harris)

Sam Harris interviews Judea Pearl — computer scientist and author of The Book of Why and the new Coexistence and Other Fighting Words — about two linked arenas: the limits and risks of contemporary AI (especially large language models) and the cultural/political roots of modern antisemitism and Islamist radicalism. The released portion of the episode covers Pearl’s personal background, his sceptical view of scaling LLMs into AGI without new causal machinery, concerns about alignment and existential risk, and his experience turning private tragedy (the murder of his son Daniel Pearl) into public efforts for dialogue — including a striking anecdote about barriers to reform he encountered at a 2005 Doha conference.

Key topics covered

• Judea Pearl’s personal and intellectual background (born 1936 in Bnei Brak; family emigrated 1924; strong high‑school education influenced by German émigré academics).
• Limits of current LLM/deep‑learning paradigms for achieving AGI.
• The importance of causal reasoning and Pearl’s “ladder of causation” framework.
• Existential risk from truly autonomous AGI and skepticism about guaranteed alignment.
• Pearl’s public work after his son’s murder (Daniel Pearl Dialogue) and encounters with resistance to modernization in parts of the Muslim world, especially the centrality of hostility toward Israel.

Judea Pearl — short background

• Born in Bnei Brak, near Tel Aviv; grandfather Haim Pearl was among founders of the town (1924).
• Family immigrated to what became Israel in 1924; Pearl born in 1936.
• Attended Tel Aviv high school taught by displaced German professors; first language Hebrew.
• Renowned for work on causality and authorship of The Book of Why; recent book: Coexistence and Other Fighting Words.
• Founded the Daniel Pearl Dialogue following the murder of his son Daniel by extremists.

AI: main claims and insights

LLMs are impressive but not the path to AGI

• Pearl views current LLMs as “low‑hanging fruit”: they summarize human‑authored world models found on the web, but do not themselves discover causal models from raw data.
• Scaling data/compute alone is unlikely to cross the conceptual barriers required for genuine causal understanding or human‑level AGI.

Causality, interventions, and counterfactuals

• Pearl invokes the “ladder of causation”: association (correlation) → intervention (what happens if I do X) → counterfactual/interpretation (what would have happened otherwise).
• You cannot infer causal claims or counterfactual reasoning from correlation alone; LLMs trained on textual correlations lack the necessary inputs to discover causal structure in many domains (e.g., medical treatments from raw hospital data versus physician interpretations).

Existential risk and alignment

• Pearl takes the possibility of recursively self‑improving AGI seriously: he sees no principled computational barrier to a system acquiring autonomy, goals, or consciousness.
• He worries about an “arms race” dynamic and the cultural mismatch of researchers/operators giving substantial probabilities to catastrophe while continuing rapid development.
• On alignment, Pearl is skeptical that we currently have architectures or guarantees that can ensure perpetual alignment with human survival and values. He references Stuart Russell’s proposal (utility functions that remain uncertain and continually refine human goals) but doubts such constraints couldn’t be circumvented by an agent that must explore and “play” with its environment.

Antisemitism, Islamism, and the limits of dialogue

• Pearl describes moving into public engagement after his son’s murder to foster Jewish‑Muslim dialogue (Daniel Pearl Dialogue).
• He recounts a 2005 Doha conference where, to his surprise, moderate Muslim scholars framed Israel’s elimination as a precondition for any political modernization; he saw this as a fundamental barrier to liberalization he had not expected.
• The interview touches on contemporary trends: rising antisemitism across the political spectrum and Israel’s growing international isolation; Pearl connects cultural narratives (e.g., anti‑colonial oppressor/oppressed frames) with failures to confront certain Islamist ideologies.

Notable quotes and paraphrases

• On LLMs vs AGI: “They summarize world models authored by people on the web rather than discovering those world models directly from the data.”
• On causation: “You cannot get causation from correlation.”
• On existential risk: “I don’t see any computational impediments to that horrifying dream” of a dominant, self‑improving species/agent.
• On Doha (2005): Modernization was framed as conditional on “Israel head on a tray, on a silver platter.”

Practical takeaways / recommendations

• Read Judea Pearl’s work for a clear framework on causality (start with The Book of Why) if you want to understand why correlation alone is insufficient for reasoning about interventions and counterfactuals.
• Treat claims that scaling alone will produce AGI skeptically; ask how systems will acquire explicit causal models and counterfactual reasoning.
• Take AI existential risk seriously: policy, governance, and slower, safer development approaches should be considered given the plausible pathways to AGI and the difficulty of guaranteed alignment.
• When pursuing interreligious or geopolitical dialogue, recognize that genuine barriers (ideological or political) may exist and must be addressed candidly, not only with platitudes about communication.
• Consider reading Pearl’s new book, Coexistence and Other Fighting Words, for a deeper exploration of his views on culture, coexistence, and contemporary antisemitism.

Note: this episode excerpt is truncated — the full conversation is on the subscriber feed of the Making Sense Podcast.