Making Sense with Sam Harris - #164 — Cause & Effect

Episode Date: August 5, 2019

Sam Harris speaks with Judea Pearl about his work on the mathematics of causality and artificial intelligence. They discuss how science has generally failed to understand causation, different levels of causal inference, counterfactuals, the foundations of knowledge, the nature of possibility, the illusion of free will, artificial intelligence, the nature of consciousness, and other topics. If the Making Sense podcast logo in your player is BLACK, you can SUBSCRIBE to gain access to all full-length episodes at samharris.org/subscribe.

Thank you. of the Making Sense Podcast, you'll need to subscribe at SamHarris.org. There you'll find our private RSS feed to add to your favorite podcatcher, along with other subscriber-only content. We don't run ads on the podcast, and therefore it's made possible entirely through the support of our subscribers. So if you enjoy what we're doing here, please consider becoming one. Welcome to the Making Sense Podcast. This is Sam Harris. Okay. Well, I'm recording this intro in the immediate aftermath of now two mass shootings.

The one in El Paso, and it appears there was one in Dayton a few hours ago. Needless to say, social media is now a cesspool. I guess there are a few things I could say about this. Actually, I wrote a piece on my blog, when I used to blog rather than podcast, about six years ago in response to some jihadist violence. And it really is the clearest articulation of what I have to say at moments like this. The conversation about atrocities of this kind, mass shootings, is generally so confused, this kind, mass shootings, is generally so confused and it's so frustrating to see people talking past one another for political or otherwise emotional reasons that, I don't know, I think I'll read the first part of this blog post just to put my argument in view in the clearest form, and then maybe say a few things relevant to the current moment. This comes from a post titled No Ordinary Violence, which was published October 11th, 2013.

A young man enters a public place, a school, a shopping mall, an airport, carrying a small arsenal. He begins killing people at random. He has no demands, and no one is spared. Eventually, the police arrive, and after an excruciating delay as they marshal their forces, the young man is brought down, or arrested. This has happened many times, and it will happen again. After each of these crimes, we lose our innocence, but then innocence magically returns. In the aftermath of horror, we seem to learn nothing of value. Indeed, many of us remain committed to denying the one thing of value that is there to be learned. After the Boston Marathon bombing, a journalist asked me, why is it always angry young men who do these terrible things?

She then sought to connect the behavior of the Tsarnaev brothers with that of Jared Loeffner, James Holmes, and Adam Lanza. Like many people, she believed that similar actions must have similar causes. But there are many sources of human evil, and if we want to protect ourselves and our societies, we must understand this. To that end, we should differentiate at least four types of violent actor. And now this is a sidebar. There may be one new subtype here that I'll add. But here's the first one. Those who are suffering from some form of mental illness that causes them to think and act irrationally.

Given access to guns or explosives, these people may harm others for reasons that wouldn't make a bit of sense even if they could be articulated. We may never hear Jared Loeffner and James Holmes give accounts of their crimes, and we do not know what drove Adam Lanza to shoot his mother in the face and then slaughter dozens of children. But these mass murderers appear to be perfect examples of this first type. Aaron Alexis, the Navy Yard shooter, is yet another. What provoked him? He repeatedly complained that he was being bombarded with, quote, ultra-low frequency electromagnetic waves. Apparently, he thought that killing people at random would offer some relief. It seems there's little to understand about the experiences of these men, or about their beliefs,

except as symptoms of underlying mental illness. Two. This is the second type. Prototypically evil psychopaths. These people are not delusional. They are malignantly selfish, ruthless, and prone to violence. Our maximum security prisons are full of such men. Given half a chance and half a reason, psychopaths will harm others, because that is what psychopaths do. It is worth observing that these first two types trouble us for reasons that have nothing to do with culture, ideology, or any other social variable. Of course it matters if a psychotic or psychopath

happens to be the head of a nation, or otherwise has power and influence. That is what is so abhorrent about North Korea. The child king is mad, or simply evil, and he's building a nuclear arsenal while millions starve. But even here, there is very little to be learned about what we, the billions of relatively normal human beings struggling to maintain open societies, are doing wrong. We didn't create Jared Loeffner, apart from making it too easy for him to get a gun. And we didn't create Kim Jong-il, apart from making it too easy for him to get nuclear bombs. Again, this was written six years ago. Given access to powerful weapons, such people will pose a threat no matter how rational, tolerant, or circumspect we become.

And I guess I would add another descriptor here. There are people, it seems, who fall into one of these two categories, who are living in an online culture of trolling now, where killing people and writing semi-bogus or entirely bogus manifestos, merely designed to confuse the media, is becoming a new phenomenon, right? These are people who are not moved by a sincere ideology. They're just, quote, shitposting. The behavior of trolling

on websites like 4chan and 8chan has been exported to the real world in the form of mass murder designed as a troll. And to some degree, I believe the Christchurch shooting in the mosque had this form, right? It's still not entirely clear what happened there. So this is a kind of derangement that social media has introduced into our lives, where some people are willing to commit murder and even mass murder simply to enjoy the spectacle it creates online. Again, they're either crazy or evil or both, but in certain cases, the reasons for their behavior are not as they appear, right? And the media seems to get very confused about this.

The media seems to get very confused about this. Okay, the third type here. Normal men and women who harm others while believing that they're doing the right thing or while neglecting to notice the consequences of their actions. These people are not insane, and they're not necessarily bad. They're just part of a system in which the negative consequences of ordinary selfishness and fear can become horribly magnified. Think of a soldier fighting in a war that may be ill-conceived or even unjust, but who has no rational alternative but to defend himself and his friends.

Think of a boy growing up in the inner city who joins a gang for protection, only to perpetuate the very cycle of violence that makes gang membership a necessity. Or think of a CEO whose short-term interests motivate him to put innocent lives, the environment, or the economy itself in peril. Most of these people aren't monsters. However, they can easily create suffering for others that only a monster would bring about by design. This is the true banality of evil, whatever Hannah Arendt actually meant by that phrase. But it is worth remembering that not all evil is banal. 4. Normal men and women who are motivated by ideology

to waste their lives and the lives of others in extraordinary ways. Some of these belief systems are merely political or otherwise secular in that their aim is to bring about specific changes in this world, but the worst of these doctrines are religious, whether or not they are attached to a mainstream religion, in that they are informed by ideas about otherworldly rewards and punishments, prophecies, magic, and so forth, which are especially conducive to fanaticism and self-sacrifice.

Of course, a person can inhabit more than one of the above categories at once, and thus have his antisocial behavior over-determined. There must be someone, somewhere, who is simultaneously psychotic and psychopathic, part of a corrupt system, and devoted to a dangerous, transcendent cause. But many examples of each of these types exist in their pure forms. For instance, in recent weeks a spate of especially appalling jihadist attacks occurred. One in a shopping mall in Nairobi, where non-Muslims appear to have been systematically tortured before being murdered. One on a church in Peshawar, and one

on a school playground in Baghdad, targeting children. Whenever I point out the role that religious ideology plays in atrocities of this kind, specifically the Islamic doctrines related to jihad, martyrdom, apostasy, and so forth, I am met with some version of the following. Quote, bad people will always do these things. Religion is nothing more than a pretext. This is an increasingly dangerous misconception to have about human violence. Here is my pick for the most terrifying and depressing phenomenon on Earth. A smart, capable, compassionate, and honorable person grows infected with ludicrous ideas about a holy book and a waiting paradise, and then becomes capable of murdering innocent people, even children,

while in a state of religious ecstasy. Needless to say, this problem is rendered all the more terrifying and depressing because so many of us deny that it even exists. Okay, I think I'll stop there. Again, I wrote this six years ago in the aftermath of some jihadist attacks, and now I'm reading it to you in the aftermath of some mass shootings in the United States, which attest at least to the problem of gun violence here, as well as to our failure to make it difficult for bad people, crazy people, dangerous people to get access to guns. And it might in fact attest to a rise of white supremacist violence. At the time I'm recording this,

violence. At the time I'm recording this, it's not yet clear what's what here. But whatever's true of El Paso and Dayton, two things are absolutely clear. One is that, again, we need some rational gun control in the U.S. And I've written about guns. My views on guns and gun control are hard enough to parse that they resist easy summary. You can listen to the podcast or read the associated essay titled The Riddle of the Gun. I can sound very pro-gun for part of that, but the punchline you should not lose sight of is that the regulations I recommend on guns in the U.S. are more stringent than anyone on the left is calling for. So don't lose sight of that if you freak out over the other parts of that essay that sound like they were written by the NRA, an organization which I hope will one day be destroyed. The short

form of this point is that we license people to drive cars, we license them even more stringently to fly airplanes, and I think getting a license to own a firearm should be like getting a pilot's license. It shouldn't be easy, and if you're mentally ill or prone to suicidal depression, it should be very difficult to get your hands on a gun. But with 300 million guns already in existence in the U.S., this is a hard thing to bring about, not to mention the political religion around gun ownership enshrined in the Second Amendment. Anyway, we need a conversation and research and political change around the epidemiology of gun violence. It's insane that we suffer this in the U.S. to this degree. It's also true that we should keep some perspective.

In the hours where I think it's now 38 people have died in two mass shootings in the U.S., more people have died from ordinary shootings and by suicide and even by medical errors in hospitals, right? So we should keep some proportion here. And finally, whatever is the case with these specific shooters, whether or not they're both people of the fourth type I describe in this essay, people who are motivated, in this case, by the lunatic ideology of white nationalism, and that may yet prove to be the case, it is obviously a bad thing that we have a president who utterly fails to be clearly and consistently opposed to these ideas. Yes, you can find him in the aftermath of Charlottesville

saying one measly thing against white supremacy, but to say that he has been ambiguous on this issue is an understatement, right? To say that he has been ambiguous on this issue is an understatement. To say that he has given comfort to racists is an understatement. He completely lacks a decent ethical political response to these trends. I'm not a fan of dog whistle theory. I don't actually think he's dog whistling in his statements to white supremacists. I think he's just an ordinary Archie Bunker style racist who doesn't care about these issues and doesn't want to alienate anyone in his base. And I think the people who

are endlessly talking about dog whistles are doing much more harm than good in our political discourse. Not everything is a dog whistle. In fact, almost nothing is a dog whistle. I'm not saying the phenomenon doesn't exist, but generally racists just tell you what they think, and when they talk to other racists, they're explicit about their racism. And it really does matter that the left's allegations against Trump

and his supporters are so poorly targeted. You know, when he tells Ilhan Omar to go back to where she came from. On the left, that is proof positive of racism. Again, I have no doubt that Donald Trump is actually a racist, but that's a bad example of racism. It can be read in other ways. And to think that it's a dog whistle to neo-Nazis is just an act of leftist clairvoyance that strikes me as totally counterproductive. To remind you how crazy this has all become, there was a Washington Post opinion editor who claimed that Nancy Pelosi was dog whistling to racists when she criticized AOC and Ilhan Omar and the rest of the so-called squad.

Nancy Pelosi? The dog-whistle meme is going to prove politically suicidal on the left. We have to be precise, even when attacking racists. So whatever turns out to be true in this case, whether either one of these mass shootings is a clear example of white nationalist terrorism, the problem with Trump is not that he is a clear supporter

of white nationalist terrorism, or even white nationalism. The problem is he is an obscenely amoral president who can't be counted upon to say anything beyond what he imagines is narrowly self-serving politically and financially. To use a great word which is now much overused, this is the U.S. presidency reduced to a grift. And it's awful. But it is not always precisely awful in the ways that are alleged on the left. And again, every error matters. We are guaranteed to have Trump for four more years if the Democrats can't get their house in order. So my political concern here is that this not get overplayed and overspun. It's totally possible that one of these shooters is mentally ill. And if this still gets talked about as white nationalist terrorism,

rather than a symptom of mental illness, that is going to be a political problem. And no, this is not a double standard. There are acts of violence perpetrated by Muslims that are not examples of jihadism, much less jihadist terrorism. Sometimes people really are violent for other reasons, as I sought to make clear in this essay. really are violent for other reasons, as I sought to make clear in this essay. However, it is yet another very dark moment, and this has all been horrible news,

but I will leave it there. And now for today's podcast. Okay, well, in this episode of the podcast, I speak with Judea Pearl. Judea is a professor of computer science at UCLA. He's the author of three highly influential scholarly books. He's also the winner of the Alan Turing Award, often considered the equivalent of the Nobel Prize for computer science. He's a member of the U.S. National Academy of Sciences. He's one of the first 10 inductees into the IEEE Intelligence Systems Hall of Fame.

He's received numerous awards and honorary doctorates, including the Rummelhart Prize, the Benjamin Franklin Medal, and the Lakatos Award at the London School of Economics. And he's also the founder and president of the Daniel Pearl Foundation. And that is because he is the father of Daniel Pearl, who was the, I believe, the first journalist killed by Al-Qaeda, at least the first that came to the attention of everyone in the aftermath of September 11th. Anyway, I mentioned this at the beginning because it would have been awkward to have just ignored it, but as you'll hear, I didn't have the heart to make Judea's experience there a topic of conversation,

so I opened that door only to close it, and then we just go on to have a fairly highbrow conversation about how science has generally failed to understand causation. We talk about the different levels of causal inference, counterfactuals, the foundations of knowledge, the nature of possibility, the illusion of free will, artificial intelligence, the nature of possibility, the illusion of free will, artificial intelligence, the nature of consciousness, and other topics. Anyway, at one point I get confused about what we're talking about, so it's a bit of a nerd fest, but I really enjoyed it, and as you'll hear, Judea is a dear person, and it was a great privilege to meet him. So now, without further delay, I bring you Judea Pearl.

I am here with Judea Pearl. Judea, thanks for coming on the podcast. Thank you, Sam. It's great to be here. So we've been circling this podcast for quite some time. It's taken a while to actually get together, and we have many areas of overlapping interest, so I'm looking forward to talking to you about your work. I was prepared, as I said offline, to just talk about your academic work, and we'll get deep into that. But given my background as a critic of Islam and as a warrior about the link between specific religious ideas and specific forms of violence, it's awkward for me to bring it up, but it's awkward for me to ignore it as well. Danny Pearl was your son who was, I believe, the first, at least first most visible person murdered, journalist murdered.

The first journalist. murdered, journalist murdered. I think the first journalist. Yeah, after 2001. So I wanted to kind of just mention that at the outset. We can talk about it or not, as you like. We should talk about this topic separately, so we can separate it to discussions. Okay, okay. I don't feel strange talking about it. I get used to talk about it. But I think in terms of listeners' interest, some people have interest in the technical part and some have in

the ideological part. It's good to separate it, too. Okay. Well, let's dive into your work and then see what happens, because your work is fascinating. So, how would you describe what your intellectual focus has been in your career? Recently, it has been the mathematization of cause and effect. Let's put it very, very concisely and precisely. But there's a direct connection to artificial intelligence that we'll talk about? Oh yeah, because if we want robots to behave like us, to communicate with us in our language,

we have to equip them with the ability to communicate in terms of cause and effect. This is our language. If they act stupidly without knowing the difference between correlation and causation, they will not be able to supply us answers to questions that are burning for us. Even simple questions like, why did the milk spill? Because I pushed it or because I was irritated or things of that sort. You want good answers, a good explanation, so we can communicate. So you just mentioned this opposition between correlation and causation. Yes, yes. And this is a phrase that will be familiar to many people.

I think many people will be surprised that it has impeded scientific understanding to the degree that it has. You make a very strong case that science has more or less ignored causation, and yet I think in the popular understanding, science is all about finding the causes of phenomenon. And so maybe we can speak for a few minutes about how statistics has rendered us unable to speak about causes historically. It's not only statistics, it's science in general. Yeah. You see, we learn physics.

Every high school kid can solve physics homework. physics homework. And if you look at the physics homework, you have boundary condition, you have the equation of motions, and find out what's going to happen. Or even what's going to happen

if you intervene and you change the spring length to double its previous value. It's a causal question. And every child can do that. But when you're trying to transfer this knowledge to a computer, to a robot, then the robot is facing a clash here. The equations of physics are symmetric,

which means that X causes Y to the same degree as Y causes X, which means that the movement of the barometer depends on the pressure, the same way that the pressure depends on the movement of the barometer. So when a robot comes in and looks at the equation and says, hmm, let me change the weather tomorrow by moving this barometer a little bit, right? What would prevent the robot from doing that? Yes, it's the same thing that prevents the high school kids from not giving the same answer. But what the high school kids have,

the notion of cause-effect. So the high school kids filters the equations in his or her mind before giving you the answer. And that is a kind of filtering that we need to do here, to introduce the asymmetry between cause and effect and do it mathematically. Because the robot doesn't understand the hand waving. Robots must understand equations. So we need an algebra which is asymmetric

to capture the asymmetry in nature. Right, so it's asymmetric with respect to influence. Time is usually the signature of influence. Correct, but it's not only the time. It's not only the time. We can show many cases in which the temporal direction, temporal order is different, and still X causes Y and Y doesn't cause X. It's very simple. I mean, you don't actually need teleology for that.

Yeah, I'll give you an example. The rooster crow precedes the sunrise, and no one will say that the rooster crow causes the sunrise. It's highly correlated, too. So the rooster crow appears to be a cause if time were your only signature. If time is your only signature, right. It's not sufficient.

So you talk about three levels of causation. And maybe back up for a second and do a little more history of ideas. So David Hume, the Scottish philosopher, has been very influential here in alleging that, at least in one place in his work, that we have no direct knowledge of causes ever. All we have is the conjunction or the correlation, the coincidence of two events. And when, you know, event B reliably follows event A, we impute causation where, in fact, there's no other

knowledge ever gained there. And, you know, I've always felt that that's almost a kind of semantic game which ignores some background intuitions we have that reach deeper into the way the world is than just mere B following A. First it's ignore experiments. And Galileo lived before Jung. So I'm surprised that Jung did not pay attention to Galileo. Although Galileo didn't make it explicit

that with experiments we get additional knowledge that you cannot get by passive observation. But Hume puts too much emphasis on regularity, which was criticized by many other people. But then Hume changed his mind between his essay and the treatise on human nature. And he, after I think seven or nine years, he said, in other words, and then he brought up a counterfactual definition of causation. Right. Had the object been different, the results would have, I don't have the exact phrasing, I have it in my book, but he changed from regularity

to counterfactual. Had the object been different, then the outcome would be different, and even put the words, in other words, between them, as if they were the same. Right. But they're totally different. The first one is statistical regularity, which sits on the lowest level of the ladder. And the counterfactual is the top layer, the third layer. Yeah, so let's talk about the three layers. You described them at one point as seeing, doing, and imagining.

Right. So seeing is this, well, I'll let you describe it. What is seeing? Seeing is you are sitting there like an astronomer, passively observing phenomena, with your hand tied behind your back, and you are talking about how your belief changes with additional observation. That's statistics. If you see another piece of evidence, you change your belief. Whether you see symptoms and you change your belief about disease, you see a disease and you have expectation about symptoms.

So this is what statistics is all about. And so that's the domain of mere correlation and Humean juxtaposition. At least the first. Human is first mood. And that, by the way, is the domain of machine learning today. Care fitting.

Under noise, of course. So that has been the dominant theology. Maybe we're going to head toward AI for a second, but maybe we should elaborate on that just for a stretch of 30 seconds. Machine learning takes in an immense amount of data and finds correlations which prove useful

as long as we give it information as to what constitutes success. So it's like take a facial recognition task. It's an example. And there's just, there's that mere correlation combined with sufficient computational power can prove very useful. Very useful. It's just not, it's just...

Amazingly useful. Yeah, just obviously not the basis of general intelligence of the sort that we'll later talk about. It is debatable whether it is sufficient for general intelligence. Seems unlikely, yeah. But my opinion is not, because I've seen mathematically that there are barriers that you cannot cross. Right. Okay, so we'll get to AI in a second and the robots that may or may not kill us.

So seeing, then there's doing. What is doing? Doing is running an experiment. I'm wondering whether smoking causes cancer, so I conduct an experiment. It's as old as Daniel in the land then. In the book of Daniel, you have a first experiment where Daniel and his fellow Israelites who were exiled refused to eat the food.

It wasn't kosher. And the king Nebuchadnezzar, commanded them to eat the king's food because it was much healthier and he depended on their talents to run the empire. So Daniel proposed an experiment. Take a few of us, give them vegetarian food,

and take the other groups and give them the king's food and see who is going to be more healthier looking. And that was the first experiment that we know of. Almost controlled, almost random. Yeah, I don't know which the control is there, but yeah. Well, take a group. So, let's say you split the group into two parts. One of them is control, the other one is treatment, they call them. And you see the difference in the outcome. It's an experiment, but of course. This was invented only in the 1930s, the idea of randomized experiments.

A randomized, controlled experiment. Yes. But we have been dealing with cause and effect much before that, right? Even from the time of Daniel. One hopes. How did we manage? Well, the child manages by conducting

playful manipulation in the world. The child finds out that moving one ball causes the other one ball to move. Playing with one toy makes a noise and the other one doesn't. So it's called playful manipulation. And that, I believe,

where we get most of our knowledge about cause and effect in the world. Yeah, yeah. You push the world and something happens. With your own muscles, right? Like Galileo dropped the two objects from the tower of Pisa and looked at them with his own eyes. Right.

That was essential. So the third level is imagining. The third one is imagining, yeah. Some people do not see the... You can sit back if you want. I can just swing this closer. No, one is imagining, yeah. Some people do not see the… You can sit back if you want. I can just swing this closer. No, no. Yeah, imagining is looking at your theory of the world and manipulating it in your mind. I start talking about imagining by showing the first sculpture that described impossible objects. It was a lion head connected to a human body.

That was the first figurine, ivory figurines, discovered from 32,000 years ago in a cave in Germany. The first object, artifact, they describe an impossible object. And how was it created? Well, the artist, in his or her mind, probably was his, he imagined taking apart the human body, sever it, and putting on a lion head. Imagine it in your mind first and then put it in the ivory. And that was the key. You can manipulate things in your mind before doing it in the physical world.

And that is a terrific idea, because that creates, according to Harari, a market of promises. Okay? Yeah, you all know Harari. He's been on the podcast. Yeah. Do you know him? He's very interesting.

I haven't met him personally. We communicated in one message. You guys should get together. So, imagining is the domain of counterfactuals, and counterfactuals are a very important part of this story. It's essential for science. How would you define a counterfactual? It's figuring out an outcome that would have prevailed had a certain observation not taking place.

Had Hillary won the election. Had Cleopatra's nose been longer than it was really, okay? Had Julius Caesar not crossed the Robicon. Don't laugh, because that's how historians communicate. Right. Okay? And they understand each other, and they form a consensus. So they can communicate. Right. Okay? And they understand each other and they form a consensus. So they can

communicate, had Oswald not killed Kennedy, how would American politics develop? When would we be pulled out of Vietnam and things of that sort? And they can communicate that way despite the fact that Oswald

did kill Kennedy. Right. How can we form a consensus about things that are conflicting with the real trajectory of history? So it's a discussion of what might have been. Might have been. what might have been. Might have been. And it's anything that falls into the bin of, had the world been different, what could we say then?

Correct. If I hadn't crossed the street at precisely that moment, how would my life be different? And with that comes all the ethics. You should have known better. Great. Yeah. So it can sound like a very dry export from the ivory tower, this notion of counterfactuals, but it underpins so much of what we care about, and I think we'll get into that.

There's another connection for me to the foundation of knowledge. What does real knowledge consist in, it's not enough to be right by accident, right? So you can't, like, you know, if I look at my watch and it's actually broken, but it happens to show the correct time at this moment, it's wrong to say that I am in knowledge of what time it is. I, you know, because a minute later, you know, I will reveal that my methodology is such that it's not delivering me actual knowledge about the world. So you need to be able to ask, and this is a problem I always get into with religious people, when I criticize religion, I criticize it for this. When you ask yourself, I would invite any believer to ask this question of themselves now, would you believe in God if God didn't exist? Do you stand in such relation to the truth of his existence such that

you would not form a false belief that he exists? Is your belief in God the result of being in some contact with reality such that if God didn't exist, you wouldn't believe he exists? And I think any look at the history and psychology of religion demonstrates that in almost every case, apart from the mystics who have some vision of God that may in fact be a vision of God, who are we to judge? Believers routinely violate this principle because the truth is they inherit these doctrines from previous generations that have merely asserted that certain books were dictated

by the creator of the universe, and there's no more burden of evidence than that, and there's no more reality testing or updating of beliefs generation after generation. There's still the mere assertion that these ancient books are the perfect record of God's existence. You are facing now a specimen of a person who answers your description. I don't believe in God. Actually, I know that God doesn't exist. Okay, you did me one better. And I still believe in him. Okay, I know that God doesn't exist. Okay, you give me one better.

And I still believe in him. Okay, well, that's going to get complicated. Why? Okay, well, so I'm reluctant to take a full detour here, but it's too interesting. Okay, so what do you mean? What do you mean? God and religion are just poetry.

Okay, well, that's... So I'm using certain metaphors, which are very helpful due to my cognition. I'm using them to communicate with you, with my children, and I say, yeah, God will punish you if you talk like that. Why not? talk like that. Why not? Which means, look, to be more scientific about it, most of our reasoning works around metaphors,

similarities. And the deepest metaphor that we have are the metaphors of family relations. We are born to mother and father. Our perception system is so attuned to whether our mother frowns or smiles. It's the first thing that we learn. You grow up and you find out

that the world is not only mother and father. It has stars and it has other things. So you create a metaphor because I only mother and father. It has stars and it has other things. So you create a metaphor, because I understand mother and father. I don't understand this movement of the stars. So I would immediately come out with the conclusion that there is some force there, like my father, that moves the stars around,

and like my father, teaches me things and punishes me things. And sometimes it's very natural. So that's the basics of our cognition. So I do not fight it. I use it. But I remember that it's only poetry. Right.

Okay. Well, then you're in a parish with very few members at the moment. You're in a parish with very few members at the moment. But that's a legitimate use of poetry and literature, certainly. But it's not what most people most of the time mean by God, as you know. This is just to say that thinking about what might have been different at the level of belief. So I believe certain things about the world, and if I believe I'm in touch with the world, I believe that, for instance, I'm staring at a microphone that I put here. I believe there's a microphone in front of me on the desk. Implicit

in that belief, to say that that really is my propositional attitude, that there's a microphone on the desk, is the assertion that if there weren't a microphone on the desk, I wouldn't think there was one, right? So there is a counterfactual built into just the assertion that this is a microphone, whether anyone ever thinks about it. But as you point out, an understanding of counterfactuals or an ability to model them is the necessary ingredient to understanding what in fact is a cause as opposed to merely an event that happens to precede some other event in time

or be associated with it, a mere correlation. It's necessary to believe in actual cause. My actual is different than average cause. Smoking is, on the average, smoking is harmful to your health, on the average. Some people could benefit from smoking. When you talk about individual, then you talk about counterfactual.

Had I not smoked, I would have lived X number of years. Well, let's talk about the smoking case, because that was a fascinating bit of history in your book, which I thought I was aware of, but it was actually a far bit more grim and delusional than I realized. I mean, there was a period of such active and protracted debate about whether or not smoking caused cancer that it went on far too long. And you had people, you had scientists who were smoking two and three and four packs a day, denying the linkage,

and there's a nicotine-empowered level of confirmation bias that was ruling the conversation there. What lessons do you draw from that period in our history? To me, it means something perhaps different than to other people. For me, it was an example of how scientists can argue about things for which they don't have a language. They didn't have a language of causation at that time. They had a language of randomized experiments which they couldn't conduct on smoking.

And that gave Fischer, who was the top statistician at the time. An avid pipe smoker, if I recall. Yeah, all his life. And it gave him ammunition to claim, hmm, maybe what we see here is just coincidental correlation between some genetic factor that makes you, you know, crave for nicotine on one hand, and it puts you at a cancer risk on the other. Okay? So what we are seeing is just the effect of a confounder.

Right. A third variable that causes both. that causes both. I am not sure that he did it because he was a smoker himself or because he wanted to be an iphamistabra, which means just a smart, a smart aleck.

A smart ass. A smart ass. Okay. And to show off his knowledge about statistics and about the possibility that you might get the same results with a different hypothesis. I'm not sure which was the case, but the fact that he resisted the conclusion of other people

who went on for more than 10 years, I think millions of people died as a result of that. But eventually, it was resolved by a commissioner, and the surgery general came out with a statement that it does cause cancer. And the way that came about it was interesting. They looked into the plausibility argument in order to calculate it. The degree to which the hidden genetic factors will have to change your craving for

nicotine, and that made it impossible, or implausible, that if you have these genetic factors you'll crave eight times more than if you didn't have it. They don't have any mechanism between a genetic factor to make this craving plausible. That was a key for the conclusion that they came up with

in the consensus they came up with, and things have been different since then. Right. But still, what one confronts there is the sense that, based on a purely statistical argument, it's always an overreach to establish causation no matter how much data you have of correlation. Correct.

And that has not been appreciated to the degree it should be. No causes in, no causes out. That was Nancy Cartwright's slogan, which people, makes sense. No, no, it does. No causation without correlation, everybody understand, okay? But the idea is that if you want to get causal conclusion, you must have some causal assumption someplace or experiment.

One of the two. This is just so important because so many people have forgotten. counterfactuals for another moment. So it does suggest that possibility is a real thing. And I've occasionally wondered, in fact, last time I

wondered this in public, it was John Brockman's final edge question. And the one I suggested was, I don't know if you were in that particular round, but my last edge question, the question that year was, I don't know if you were in that particular round, but my last edge question, the question that year was, what should the last edge question be? And I believe my question was, is the actual all that is possible? Which is to say that is possibility an illusion?

Is there only what is actual? Is the notion that something else could have happened always just an idea, and does it actually not reach into anything that we can profitably think about? Is there simply just the fact of the matter in every case? And counterfactual thinking is explicitly thinking about what is possible, what might have been, had things been different. And I guess I'll just put it to you. How do we know that possibility is even a thing? It's useful to speak as though it were a thing. And this actually connects to the topic of free will, which you write about in the book, because you and I are convinced, happily,

not many people agree with us, but you and I are both convinced that free will is an illusion, but in one way or another, it's a useful or inevitable illusion. But we still don't understand what makes it useful. Right, right. And you and I might disagree a little bit about how useful it is, but is it possible, and here there's the useful invocation of the concept, is it possible that possibility is an illusion as well? It is. Because there is something better than possibility. If you'd like to continue listening to this conversation, you'll need to subscribe at samharris.org. Once you do, you'll get access to all full-length episodes of the Making Sense Podcast,

along with other subscriber-only content, including bonus episodes and AMAs and the conversations I've been having on the Waking Up app. The Making Sense Podcast is ad-free and relies entirely on listener support, and you can subscribe now at SamHarris.org.