Both causation and covariation manifest themselves as associations. Just because some features drive the black-box predictions doesn’t mean changing their values will actually change the outcome in real life. Kinda like that old joke — nobody ever promised to feed the masses in.

Thank you Elias for pushing our thinking this direction, and please pass my admiration to Drago for his productivity…. I’m wondering why did you use another LLM as the ground truth here? it just adds another abstraction… I mean, wouldn’t it be more convincing to estimate.

RT @yudapearl: I’m thrilled to share that the Second Edition of The Book of Why will be released at the end of this year. It will include b….

I’d go even further than Schopenhauer’s: “Man can do what he wills, but he cannot will what he wills.”. Ideas are determined; ideas don’t determine.

“Evidence-based enterprise” turns science into clickbait journalism; headlines get crowned as truth and then swung like a machete in every argument.

Intelligent agents that can’t plan?.And just leave planning to programmer? 😂.That doesn’t sound like free will… or agency.

That’s not correct! 👇🏻 The assumption is much milder. We just need to assume that the factors we condition on (or stratify on, in epi lingo) block all chains of dependency that could connect the exposure and the outcome. A minimally sufficient adjustment set may not include.

A quiet time in Whistler

RT @JacobJHutton: @AdanBecerraPhD @soboleffspaces @yudapearl Yes a counterfactual mediation analysis was used to estimate the Natural Direc….

Congratulations to @JacobJHutton on a successful defense!. What a feast for causal inference: causal diagrams, minimal adjustment sets, mediation analysis, direct and indirect effects, probability of benefit — all in the context of cutting-edge clinical practice!. @yudapearl

RT @yudapearl: Exciting breakthrough from @eliasbareinboim -- a counterfactual-calculus (cfl-calculus), akin to do-calculus, for handling i….

Language isn’t the center of everything, except linguistics. In the end, what matters in a counterfactual like “Y wouldn’t have happened if it weren’t for X” is whether it holds up in the real world — not that it’s a proper type 3 conditional.

How does this link to causal inference?. Chomsky’s critique of AI has value but risks missing the bigger picture. If explanation is judged by practice, we shouldn’t dismiss AI just because it doesn’t think in words or use innate linguistic structures. Science moves forward.

Third: Explanation grows through shared work, feedback, and testing. AI fits into this process too — even if it doesn’t “understand” like humans. Fourth: Practice matters more than linguistic form. Chomsky talks about innate grammar. I say — minds generate endless ideas. Most.

First: Science is a collective effort. It’s not about lone geniuses or inborn knowledge. It’s people testing ideas together through practice. Second: Not all understanding starts with language. Even animals understand things without using human language.

Chomsky says language is central to human understanding. He sees it as the key to explanation, creativity, and reasoning. He argues AI lacks this because it works through patterns, not innate structures. But I see it differently. Chomsky focuses on individual minds. I focus on.

This early post helps connect with @yudapearl’s question about “The False Promise of ChatGPT” in the NYT. My view reflects ideas from pragmatic philosophy — thinkers like Popper, Dewey, Peirce, and even Mao Zedong’s “On Practice.” I’m also using this to push back on Noam.

I briefly addressed this in my earlier posts on X:. Scientific discovery is a collective human enterprise. Many discoveries weren’t due to language; not all involved language. Minds constantly generate ideas. Most are wrong once tested against reality, but rare successes become.