Judea Pearl: Causal Reasoning, Counterfactuals, and the Path to AGI | Lex Fridman Podcast #56

Pearl stresses that correlation and conditional probability cannot, by themselves, yield cause-and-effect; you need explicit causal assumptions encoded in models (graphs) and a calculus that distinguishes observing from doing.

By conceptually cutting incoming arrows into a variable (e.g., blood pressure) and setting it by intervention, the do-operator allows us to mathematically reason about the effects of actions, even when real experiments are impossible.

Questions like "Was it the aspirin that cured my headache?" or "Would the prisoner have died if rifleman A hadn’t shot?" are counterfactuals; Pearl shows they can be rigorously defined via model "surgery" and are the highest rung of causal reasoning.

Deep learning systems are, in Pearl’s words, sophisticated conditional probability estimators; without causal models, they cannot answer intervention or counterfactual questions and will hit a ceiling on what they can do.

Children learn causality by acting on the world, receiving guidance, and mapping new situations onto familiar metaphors; Pearl believes AI must similarly combine simple causal models from many domains instead of just fitting patterns.