Mindscape 196 | Judea Pearl on Cause and Effect

Patreon: / seanmcarroll
Blog post with audio player, show notes, and transcript: www.preposterousuniverse.com/...
To say that event A causes event B is to not only make a claim about our actual world, but about other possible worlds - in worlds where A didn’t happen but everything else was the same, B would not have happened. This leads to an obvious difficulty if we want to infer causes from sets of data - we generally only have data about the actual world. Happily, there are ways around this difficulty, and the study of causal relations is of central importance in modern social science and artificial intelligence research. Judea Pearl has been the leader of the “causal revolution,” and we talk about what that means and what questions remain unanswered.
Judea Pearl received a Ph.D. in electrical engineering from the Polytechnic Institute of Brooklyn. He is currently a professor of computer science and statistics and director of the Cognitive Systems Laboratory at UCLA. He is a founding editor of the Journal of Causal Inference. Among his awards are the Lakatos Award in the philosophy of science, The Allen Newell Award from the Association for Computing Machinery, the Benjamin Franklin Medal, the Rumelhart Prize from the Cognitive Science Society, the ACM Turing Award, and the Grenander Prize from the American Mathematical Society. He is the co-author (with Dana MacKenzie) of The Book of Why: The New Science of Cause and Effect.
Mindscape Podcast playlist: • Mindscape Podcast
Sean Carroll channel: / seancarroll
#podcast #ideas #science #philosophy #culture

[@scottk7515]

I loved this conversation, Sean. Thanks for having Judea Pearl on your podcast! One point to consider - babies are inherently curious to establish an intuitively apprehended causal diagram, as distinct from ADULT ANIMALS ... not animals in general. Baby animals are just as intrinsically curious, establishing their own causal diagrams. Always trying to find humans as superior and distinct from other animals is invariably wrong, that end up diminishing a natural empathy we should have with other creatures, resulting to indifference to harming our fellow-creatures.

[@ralphclark]

I’m a noob in this field and yet I found myself hanging on every word. Professor Pearl is clearly a fantastic teacher.

[@bohanxu6125]

1:01:15 Finally... someone asked this question. From my understanding of statmech, to make entropy well defined, we at least need to put some constrains on the macroscopic parameters (actually more is needed to get a probability distribution that determines entropy). Of course, Sean knows a lot more than I do, so I acknowledge my huge uncertainty in what I'm saying. For instance, we have to declare we are interested in the states where the balls form a triangle, and then we can talk about entropy, I think. The universe doesn't care about our definition of macroscopic parameters. I never quite understand what people exactly mean when they say the universe start from a low entropy state which is instantaneous and without any uncertainty. I feel we need to define how big of a phase space region that we consider to be states similar to the initial big bang, and then we can talk about probability distribution in that region... but that's a totally human construct which the universe doesn't care about. I feel like there is some interesting ideas in defining entropy for an instantaneous state... like creating a mesh of spacetime.. and consider each subregion as part of an ensemble.. and study the property of the ensemble. However, I feel like it's more complicated than the typical definition of entropy in stats mech.

[@RobRoss]

I enjoyed the discussion on entropy. This is in Sean’s wheelhouse. I actually can see merit in both ideas. In our real-world observations, we have never seen a bunch of moving billiard balls form a perfect triangle centered on the dot on the table, pointing perfectly at the other side of the table. But I have also thought that many states we do consider “ordered” are just a label we apply to a particular configuration; we are biased. For example, an ordered set of playing cards (in suit order, numerical order within each suit) is just one possible configuration and we are biased to this order. If we had grown up with a different number system, where the symbols for our numbers were in a different order, we would become acclimated to that particular order. Eg. if we had learned that our number system is : 3,8,1,2,9,6,4,7,5 , and a random configuration of cards was 1,2,3,4,5,6,7,8,9, because of our learning bias we’d call this sequence “random.”

[@Eziolise]

The argument on entropy is fascinating i tend to agree with Dr Judea that we are biased by langunage. Imagine a world where the cue ball disperses the balls in the triangle into another triangle. and we then played the video backwards, an observer would not be able to determine the progression of entropy just by viewing the video.

[@AlexanderKoryagin]

Fascinating! Thank you so much, Dr. Carroll, as always!

[@barbcarbon9440]

Stumbled upon your podcast looking for someone interviewing Judea and I just love you! Not only are you a great interviewer, your voice reminds me of Alan Alda which is very comforting. I’ll be listening a lot! Thanks for doing it!

[@Life_42]

Lovely episode!

[@barundish]

Omg the guy wanted to argue with Sean on what entropy is. "If I remember my undergrad thermodynamics, SEAN FREAKING CARROLL, you will see I'm correct about billiard balls" Got a chuckle but love love love this episode

[@TheDudeKicker]

I've struggled with that idea for a while now, but in the end, I have to trust that someone like Sean probably understands this at a level that I don't.

[@lower_case_t]

True, but then, challenging authorities is the way to find out things you don't understand. Either they will explain it or it turns out that they really aren't standing on solid ground. I would have loved to listen to Sean's take on this. My first idea was to point out that, from an AI's perspective, it would be easy to find an algorithm that constructs the arrangement of the balls in the triangle, but nearly impossible to construct any random (or high-entropy) configuration without having to give the specific coordinates for each ball.

[@TheDudeKicker]

@@lower_case_t :. I really like that explanation. Thank you, that makes sense on a level I can understand.

[@7447744774477447]

Just so you know, "the guy" is a Turing Awarded statistician with a Master's degree in Physics, and the inventor of the "Pearl Vortex"

[@foobargorch]

@@lower_case_t that's basically an informal a description of Kolmogorov complexity, i think Pearl's point still applies (the language is given, much like simplicity and ability to name things is implicitly given) but there's still a sense in which you can compare things, and upper bounds on entropy do have a strong objective meaning. clearly Pearl understands this deeply because of his statement on random variables being constructions, i was really surprised by that remark that it's completely subjective, would have been nice to hear a more in depth discussion of that

[@thewiseturtle]

Judea has an excellent point about entropy, which is that it's really common to describe one random state as being higher or lower entropy than another random state simply because we "like" some states as compared to others, when in fact they are both equally likely and thus have the same level of entropy. I think this preference bias is really holding back academic physics.

[@NoActuallyGo-KCUF-Yourself]

It's not a bias, it is a definition. Entropy is not mere randomness. Just because two random states are equally likely doesn't mean each are equally capable of producing useful work. It is only an arbitrary preference in the sense that the word "entropy" was defined with a preferred thermodynamic quality.

[@thewiseturtle]

@@NoActuallyGo-KCUF-Yourself The original meaning of entropy, that Boltzman came up with, is a measurement of the number of possible states a system can be in. Each state is equal, and in a natural system each state is a single, random, output. (Like flipping a coin repeatedly, or the collapse of a wave function in quantum physics.) Individuals have biases for which outcomes/states we prefer, but that's got nothing to do with entropy. Four heads in a row has equal entropy to having a head, tail, tail, head, or tail, tail, tail, head, or any other combination of four flips. To get more entropy, you have to flip it more than four times. You can use a different definition of entropy, based on something other than Bolzman's definition, but that's going to confuse people, which is why so many folks are so confused about entropy now: lots of weird, unique definitions of the word. Judea and I are using the original one.

[@EuphoricDan]

This was amazing. I love this guy

[@EdoardoMarcora]

As a next topic of discussion it would be nice to explore the history/ways the entropy concept has been developed by physicists, information theorists, and statisticians.

[@michaeljfigueroa]

This was one of my favorite episodes!

[@rajeevgangal542]

Sean needs to followup on a lecture on entropy where he deconstructs Judea riposte

[@milliern]

Does Pearl talk about there being no causation in physics (because of time symmetry) in one of his articles or books?

[@sd-ti4yv]

Why do they say that the causality is not to be found in physics? The 2nd Newton law says p(t+dt) = p(t) + F(t) dt, which means the values of p and F at time t predict/cause the value of p at time t+dt. It seems all physics is based on causality.

[@librulcunspirisy]

Thanks

[@NocturnalJin]

The "stupid" robot changing the barometer to stop the rain sounds an awful lot like human superstition. If you want to emulate human intelligence, you may actually need a dash of confusion. Or more - we certainly have it in heaps, ourselves.

[@luizarthurbrito]

Judea pearl is freaking great. Too bad I'm too dumb to go into this "the book of why" without being overwhelmed. I'm just glad people him exist, so that the world is not left to people as mediocre as myself.

[@simonyricools]

Haha, we are legion, buddy.

[@davethebrahman9870]

I concur, not of course in your mediocrity but in my own.

[@mikhailfranco]

The Elephant in the Detector for particle physics is the claim that the Standard Model is CPT invariant, with the experimental fact that CP is violated (1964), hence T symmetry must also be violated. So there is some experimental data stream for particle interactions where you _can_ tell which way the film is running. Physicists have had nearly 60 years to address this, but I've never seen an explanation. It seems the explanation for matter-antimatter asymmetry may also be the origin of time. The fact that matter-antimatter asymmetry may just be the flipside of darkantimatter-darkmatter asymmetry does not change this fact. They both balance overall, but perhaps the separations of matter & antimatter or luminous & dark is linked to the origin of time _at the particle level._

[@Robinson8491]

Sean, could you ask Judea about the following: partial correlation (in statistics) differentiates between cause and effect in three variables, but not more. Can AI solve this problem in practice, even if we don't understand how? I really want to know! If AI can circumvent mathematical logic basically, that would be groundbreaking and amazing!

[@robocop30301]

@1:11:00 Anil Seth's theory of consciousness

[@TheSanmanju]

I find the idea of the DO operator reminiscent to the idea of Measurment in quantum mechanics, in the sense that theres a certain collapse in a space of information that reduces that space. Do you think the analogy is meaningfull?

[@mikhailfranco]

That is a nice observation. A DO operation collapses the causal graph by removing some dependencies (edges) and forcing some outcomes (states). After the DO, the _'doer'_ is connected to the causal graph, in a similar way to the _'measurer'_ being entangled with the quantum state. A _'ManyDoers' Interpretation_ would say causal graphs are duplicated (and contracted) for every possible DO operation, in the same way that _measurers_ duplicate (copy) and contract (collapse) worlds in the _'ManyWorlds'_ Interpretation.

[@mikhailfranco]

Warming to your theme... I would say there is a duality between Doers and Measurers. They are the sources and sinks of causality. They each have many options, and they literally realize the world in their combinatorial interventions. Of course, the Doer may _be_ the Measurer, and we achieve a unity of knowledge over the phenomenon. There must be an information connection between intervener and measurer to instantiate their own actions. Now we can take one more leap over the bull, to find that _only_ the Doer-Measurer Connection has enough knowledge to create the causal graph in the first place. We have found the Creator. This is the 4th Level in Judea's Hierarchy.

[@DestroManiak]

Cause and effect seems to be a subset of philosophy of time. There is simply no cause or effect, without a temporal perspective.

[@judgeomega]

of course time is a prerequisite , but i think 'information flow' is more closely related.

[@mikhailfranco]

For every coffee-and-cream mixing experiment that _'proves'_ entropy increases, there is an equal and opposite oil-and-vinegar mixing _salad dressing_ experiment that shows separation to a _'simpler'_ unmixed state.

[@NoActuallyGo-KCUF-Yourself]

No. If you understand the different intermolecular forces that cause mixing vs separation, you will understand that entropy increases in both cases.

[@mikhailfranco]

@@NoActuallyGo-KCUF-Yourself Of course, but my point was that nuance is usually missing from the pop-sci explanation.

[@nicholascurran1734]

If smoking does cause cancer, as implied by Sean's response, have we also found out what prevents cancer in smokers, given that not all smokers get it?

[@mikhailfranco]

Could just be randomness. Tar causes cancer. Cancer is a mutation. Mutations are probabalistic. However, there could be some genetic influence on cancer, because I expect people vary in their ability to repair damaged DNA in response to smoke, other carcinogenic chemicals, and radiation. So genes do not _cause_ cancer, but may affect the chances of smoking causing cancer.

[@NoActuallyGo-KCUF-Yourself]

The only thing that prevents cancer is luck.

[@joshua3171]

i wonder why they refused steam power

[@joshua3171]

barometer like a politician sitting in an air-conditioned building stating we don't have the power to change the climate

[@joshua3171]

hit the triangle at the right spot and u can get in 1 off the break

[@yaserthe1]

As you go through life you'll see, there is so much that we, don't understand. And the only thing we know, is things don't always go, the way we plan.

[@NoActuallyGo-KCUF-Yourself]

As you go through life, I hope you learn how to use commas.

[@X1Y0Z0]

💯😊🤔🙏🏽

[@ludinodreamsmith4361]

Why does causality have to be emergent? Surely, no phenomenon has been observed, ever, that has not been caused by something? And no observation would be possible if it were not an effect of some cause? Even if you were able to unstir a cup of coffee with cream, it would not change the fact that it was once mixed together. Is it not more reasonable to think causality is primary, and our physics is an account of regularities, patterns, in this causal web? It does not mean time can go backwards. I think such an assumption would lead to the same point as the one you made, that the future would be predictable in theory, but not in practice, but the past would not be postdictable - this is the opposite of what you would intuitively think since the past should be an objective historical fact, but this is simply a result of the direction of the arrow of causality.

[@Paxsali]

There are no causes and no effects in this world, just as there are no colors and tastes. Some events can be traced to prior causes, so we call them the effects. And similarly events which lead directly to other events are called causes. The underlying, general relationship between these events unfortunatelly has no name, only the "nodes / events" have names. It's like living in a culture which has no generic name for "relative", but only individual names for instances of relatives, like "father" and "child". With the metaphor of father and child it becomes clear that every child can itself become a father to it's own child at some point later in life. The issue comes into play when you state that every child must have a father and every child must become a father itself. I don't see that as a given. It's like the construction of the natural numbers. All natural numbers have a successor. You can identify, if you want to, successors and predecessors. But you cannot say every number has a predecessor. IMHO, there is no reason to assume that all events are caused, even tho we can identify some causes and effects. Also causes and effects are not primary entities, events are. Similarly to how colors are not fundamental, but wave lengths of light. This brings us in difficult situations where the human language is working against us, because some may be inclined to say "but there are no effects without a cause" (this would be a contradiction of word effect, by it's dictionary definition). At this point I go back to my original point: cause and effect does not exist in and of itself. It's a relationship that can be established, but is not axiomatic, and I think it's not necessary. There are events (a non-controversial observation), some events cause other events (a self-evident observation), it is a mistake to assume that all events are caused by prior events, necessarily. That's why I reject the basic idea of "universal causation" - I believe it's an assumption that cannot be proven or disproven, AND I don't even think it's necessary to describe the world that we observe.

[@alsaba5203]

Agreed. Maps are not terrain. Concept, idea or theory are not reality. Only just arrow pointing at something. Like clock pointing the "what time is?" but not tell what " is" a time. ( any analogy is weak but again how to communicate?) It's the same kind of problem as free will or meaning of life. There are no answer but more or less better description what we experience. Anyway... a joke... Love is a such beautiful idea - but why I married her?

[@addammadd]

1:03:15 hilarious exchange. Prof. Carroll just really really super badly wants to reify his core dogma and our guest proves he’s spent most of his career doing just fine without it thank you very much. Bless them both. I, for one, would love to listen to Prof. Carroll explain to Slavoj Žižek why his whole arrow of time myth somehow transcends ideology. I think it would be a very stuttery, very slobbery, very sniffly conversation.

[@avrenna]

If you would like to learn more about the arrow of time, Sean has explained it many times in many places, including his book, From Eternity to Here: The Quest for the Ultimate Theory of Time.

[@NoActuallyGo-KCUF-Yourself]

Not participating in reality does not cause reality to not exist.

[@chudleyflusher7132]

In my opinion, Norton’s Dome breaks the notion of causality.

[@aprylvanryn5898]

Can you give me a brief overview so I can decide if I want to learn more? I've never heard of Nortons dome