2023-05-09 Conversations with Coleman Multiverse of Madness with David Deutsch
Duration: 01:35:57
Transcript
Coleman Hughes
Welcome to another episode of conversations with Coleman. My guest today is David Deutsch. David is a renowned physicist and philosopher, best known for his work on quantum computation and his contributions to the field of quantum mechanics. He’s a fellow of the Royal Society and a visiting professor of physics at the University of Oxford. David has written two books called The Fabric of Reality and The Beginning of Infinity. In this episode, we talk about the purpose of science. We discuss the concept of an explanation and its crucial role in the scientific process. We examine the famous double slit experiment. We discuss rival interpretations of quantum mechanics and what they imply about the nature of reality. We also talk about progress in physics and advances in artificial intelligence. Without further ado, David Deutsch. Okay, David Deutsch, thanks so much for coming on my show.
David Deutsch
Thank you for inviting me. It’s very pleasant to meet you.
Coleman Hughes
So before we get into the deep topics of science and philosophy and theoretical physics and the frontiers of human knowledge and scientific progress and all of the topics that you’ve dealt with professionally and have explained to the general public in your two great books, which are called The Fabric of Reality and The Beginning of Infinity, both of which I highly recommend. I just want to get a little bit about you. How did you come to be a theoretical physicist and what have you focused on for most of your professional life?
David Deutsch
I wanted to be a physicist really for as long as I can remember, even definitely before I knew the word physics or the word physicist. I wanted to and more specifically, I wanted to do research in theoretical physics. And so that’s what I ended up doing. And I have a tendency to gravitate towards the more fundamental side of whatever I work on. And so originally, I thought that was going to be quantum gravity. But I found after starting on a doctoral course, which was entitled quantum gravity, because I kind of thought, well, you know, I’ve got three years to do this and how hard can it be? But I never made any progress with quantum gravity. But I very quickly realized, and this thought had been growing on me, even as an undergraduate, that there was something deeply wrong with quantum theory as we were taught it and with quantum field theory as well. And so I turned in that direction. And fortunately, I had a very accommodating boss, Dennis Sciama, who basically wanted me to work on anything that I found interesting. And later, John Wheeler as well had the same attitude. And I was working on various aspects of quantum theory, trying to clean them up. I was led to the many universes interpretation, the Everettian quantum theory, as I would now like to call it. And then that led directly into quantum computers. And now more recently, I’ve kind of stepped aside from quantum computers, quantum information research into another direction that I’m hoping will be successful. It’s called constructor theory. And it’s basically a new way of formulating laws of physics, which I hope will make certain puzzles and problems easier to solve.
Coleman Hughes
Okay, so the audience of this podcast ranges from people that are currently getting PhDs in physics to people that took no more than high school physics. So no doubt many people will know exactly what you mean when you said you thought you could solve quantum gravity in three years. And we’ll sort of get the joke inherent in that. But for those who don’t, can you briefly explain what you mean when you said you wanted to solve quantum gravity in three years and failed to do so?
David Deutsch
Yes. Well, I said that I’m attracted by the fundamental side of anything. And in physics, the most fundamental theories are quantum theory, which has the reputation of being about very, very small things, and general relativity, Einstein’s theory, which has the reputation of being about very large things like galaxies and the whole universe as a whole. Now, actually, both of them are not limited in that way. Both of them, if they are true at all, if they were true at all, they would apply to the whole universe, to everything physical. And yet they are incompatible with each other mathematically and conceptually. So quantum gravity is the name of the field of trying to unify these two most fundamental theories. And the problem is still unsolved today. There are many ideas for solving it, which I now think with hindsight are very inadequate. And we’re going to need a lot of very good new ideas to address them.
Coleman Hughes
So I was going to ask this later, but I’ll ask it now. From my point of view, I’ve seen a decline in progress in physics over the past century. So for instance, if you were born in the year 1900, you would have been in your 30s when Einstein made his most famous discoveries about general relativity and special relativity and the curvature of space time and all those incredibly interesting concepts and improvements on Newtonian physics. And then you would have been in middle age for the discovery of quantum mechanics as we know it today. But if you were born around 2000 or in the mid 90s as I was, I notice a distinct lack of those paradigm shifting explanations in my lifetime and really in the 50 years preceding my lifetime. Would you agree that physics has slowed down or halted in terms of its progress in explaining the world? And if you agree, why is that?
David Deutsch
I think progress has definitely slowed down not just in physics, but I should say not just in fundamental physics, because there has been a lot of progress. Lasers were invented in that time and space travel and computers and so on. But fundamental physics, yes, has definitely slowed down during that period. And there are various theories about why this is, not just physics, I said, other fields, philosophy, slow down even more, you might say, apart from Karl Popper. And there are theories about why this is. One that I find particularly unconvincing is the low hanging fruit theory, which is that in 1900, there was lots of progress to be made that was relatively easy. So if you were interested in that sort of thing and you knew just enough mathematics, then you would solve things. And they solved things in 1900 and 1905 and 1913 and 50 and so on. I think that is completely absurd. First of all, fundamental physics is definitely more problematic today than it was in 1900. In 1900, many people thought that all that was left in physics was to dot the i’s and cross the t’s, just a few minor anomalies to clear up, even if they needed clearing up at all. And on the other hand, the mathematics that was needed in both relativity and quantum theory was not known to the physicists who solved it. They encountered the problems. They learned the mathematics in order to solve the problem. It’s not that they had gone through an undergraduate course where they were taught how to do dozens of problems that would come up, because nobody knew in 1900 that the mathematics of Riemannian manifolds would be of use to physics. And so it wasn’t in physics courses. What happened was that there were people who were interested in solving problems. They found the problems, even though there were far fewer of them than today. They worked on them with enthusiasm, and they did whatever was necessary to solve them. There were relatively few of these physicists. I have on my wall a picture of the Solvay conference of whenever it was 1913 or something. I forget. But anyway, near the beginning of the century, those were more or less all the physicists that were around at that time. If you look at the picture, they’ve got labels. Many of them won Nobel prizes. Many of the people in that picture won Nobel prizes. Many of them are, I won’t say household names, but they are names. Things were named after those people. So everybody who studies physics gets to know those names as names of things. Today, there are thousands of times more physicists than that. So what’s the difference? If it’s not the low-hanging fruit theory, what is it? Well, I think it is simply physics education. In 1900, physics was not a standard subject. There was a subject, science, that was maybe taught in schools. But I think even when Einstein first came to America, the newspapers referred to him as the famous mathematician Einstein.
David Deutsch
Physics wasn’t a thing. Only specialists had even heard of it, really. Whereas now, there is a physics career, which you start even in school when you’re designated as maybe you’d like to study, or maybe you’re clever enough or something to study physics at university. So you do the exams, and then you go to university, and you do the exams there. You’re taught a curriculum. You are taught the problems that you are supposed to be addressing. None of these problems are fundamental. The problems you’re supposed to be addressing are all incremental. Then you get a doctorate. By the way, many of these early people had worked on the problems that they were going to solve independently as undergraduates. There wasn’t a recognized path into physics which involved getting a doctorate. Anyway, once you have a doctorate, you have to get a postdoc position. For that, you need a grant. For a grant, you need a grant application. This grant application always includes a statement of what you are going to discover. Now, that is fatal to genuine research, because genuine research at the foundational level, and I would say largely at any level, but definitely at the foundational level, is incompatible with knowing what you’re going to discover. Because discovery is an open-ended problem-solving process, and the first problem you address isn’t going to be the one that is your discovery at the end of n years or whatever it is. It’s just the beginning. I always say of graduate students, beginning graduate students, that if they’re not totally confused by the end of their first year or second year, then they’re not doing it right. Yet, at the end of their first year, they must now, in most universities, produce a document of what they have discovered in the first year, and how they’re going to develop this, and how they’re going to have completed it in the third year. This whole attitude is inimical to fundamental research, and I don’t know whether it alone is responsible for the slowdown. But it’s certainly a heavy burden on a researcher, and it would have been a heavy burden on those people in 1900. I don’t know if any of them would have succeeded if they had gone through that kind of education.
Coleman Hughes
Do you think it’s possible that reconciling quantum theory and the theory of gravity is just inherently a harder problem than having figured out either one of them?
David Deutsch
I’m sure it isn’t. The reason I’m sure is that the enormous amounts of mathematics that have been devoted to attempting to solve this problem have not got anywhere. That is, not only have they not solved it, they haven’t made progress in understanding what the problem is. They haven’t made the problem harder. They haven’t connected it with other things. On the other hand, the problems that I see with quantum gravity are all physical, or you might say conceptual. They’re all about the worldview that quantum theory invites us to adopt in order to understand the world, and the worldview that relativity invites us to adopt. They are fundamentally at odds with each other. Now, you may ask me, what exactly is the incompatibility between them? That is very hard to put into words. In fact, that is the problem that this is very hard to put into words. We learn as professional physicists to change gear when we’re thinking in a quantum way, or whether we’re thinking in a relativity way. We change gear, but there is no smooth transition between the two. The best I can say, I think, if you were to ask me that, is that gravity is about space-time. Space-time is about where and when things happen. Quantum theory is about multiple realities. It’s about the way that many different things happen at the same time. If many different space-times happen at the same time, if gravity is different in different universes, then there can be no such thing as where a thing is. Then if you say, for example, the sun produces an effect on space-time, which affects a planet, if there’s more than one space-time, then it will affect the planet in more than one way. The planet, in turn, will affect the sun in more than one way. There is no way of reconciling, it’s called the back reaction problem, although that rather understates it. There is no way of reconciling the multiplicity of possible positions of the planet, its multiplicity of effects it has on the sun, and then the multiplied multiplicity of the effect that that has on the planet again. I’m trying to think about gravity in a quantum way by explaining this and trying to show you how it crashes, it breaks down. I didn’t have to say anything mathematical. When I’m trying to explain this, I’m not thinking that certain tensor isn’t equal to another tensor. That’s not the nature of the problem.
Coleman Hughes
Right. I want to give people the tools to understand exactly how interesting and strange this is. In order to do that, I want to hit the pause button on quantum mechanics and relativity for a moment and reverse back to the primacy of explanations in science, which is something you’ve written a lot about. Just to frame this, human beings have been around for well over 100,000 years, beginning on the African savanna and inhabiting the rest of the earth. We’ve had an inquisitive mind about why things are happening. We have a thirst to understand, if only to survive and to thrive. When we see a new kind of animal, we want to learn how it behaves. We want to come up with a kind of implicit theory that helps us predict the world around us so that we can survive. We want to predict the people around us. We want to predict the environment around us so that we can change it to our advantage, manipulate it to our advantage. It’s just a part of what it is to be human. A few hundred years ago, you could say there was a kind of break in history or there’s a fundamental change, which is often called the scientific revolution. Can you describe the significance of the scientific way of thinking, how it differed from the past 100,000 or more years of human curiosity and why that has led to so much progress and perhaps include in there your explanation of what an explanation is and why that is of prime importance in scientific thinking?
David Deutsch
Yes. Well, I see the scientific revolution as part of a wider intellectual movement, often called the Enlightenment. Although there are two branches of the Enlightenment and only one of them is related to the scientific revolution. But if we go back to 200,000 years or quite possibly as long as two million years ago with species before Homo sapiens, something happened there, which is the beginning of everything else that we find good. You said that there are animals that are curious. Our ancestors, apes, ape-like creatures before this thing happened, you know, maybe two million years ago, maybe 200,000, but they were curious. Monkeys are famously curious. Puppies are curious. But what’s called curious in those animals is different from what’s called curious in humans. It just has the same name because people have a wrong idea about what humans are. An animal which is curious has a built-in impulse, genetically built-in impulse to investigate novelty in its environment, a thing that it hasn’t seen before. So if it has seen it before, it doesn’t go and investigate it. If it has, then it goes to investigate it. And different animals have different amounts of this. Some have basically none of this, but some are very curious, and this obviously has a danger, but it also has a benefit. So it’s a benefit. They can become familiar with something before it’s dangerous. But humans are completely different. Humans are attracted to going out to investigate familiar things. The first things that humans try to explain that we know of are the lights in the sky, the most familiar things in our environment, the most constant things in our environment, the sun, the moon, the stars. People were thinking about them. Why? Because although they were perfectly familiar, they weren’t well explained. They posed problems of understanding. For example, what makes them go up into the sky and come down again? Why don’t they fall down? What fuels them? Everything else that emits heat or light runs out of fuel. They don’t. Everything else is affected by things like the weather and so on. They are not. They’re completely immune. Now, how do we explain that? People try to explain this. Well, we only have records by definition going back to historic times. But I bet that in prehistoric times, people wondered about that as well. And I think when they invented things like campfires, they weren’t using that animal type curiosity. They were using the human type curiosity. When they invented the campfire or a better method of hunting, whenever they made an improvement, they were trying to improve familiar things or to change familiar things to solve a problem. Now, I said that the lights in the sky were unexplained. What I mean by that is not just that they couldn’t predict them. In fact, it’s pretty easy to predict that the sun is going to rise every day and that the seasons are going to come along one after the other. In order to go further than that, you need to have a working model of the world that you want to explain in your mind. And that has to not just predict, it has to correspond to the reality. So, when you’re interested in whether the seasons are coming because you planted a crop or whatever, it would be good to know when it’s coming. And once you do that, you have to guess that there’s going to be a solstice.
David Deutsch
You have to wonder about things about the sun, which are not obviously related to your crops. And that story that corresponds to reality, that’s explanation. And I, along with following Karl Popper, think that seeking explanations is the essence of science and that seeking agreement with experiment is merely one of the methods of science. Having a predictive theory about the sun rising every day is not going to take you to the theory of solstices. To take you to the theory of solstices, you have to observe not just the sun rising and setting every day, you have to observe some other mystery, the mystery about a familiar thing. Namely, for example, the sun doesn’t get as high in the sky at some times of the year. I say this in words, but to say at some times of the year is already a guess about reality. It didn’t arise at first as an observation, it arose at first as a guess about what reality is like, that it’s at different times of the year, rather than when the trees are doing so and so, or when the animals are doing so and so, or when it’s colder or hotter. No, to get to the idea of a solstice, you have to be curious about the sun, the actual sun, not just observations of it.
Coleman Hughes
So this is useful to understand why this is significant. It’s useful to contrast this with the view that scientific theories are just about predicting reality. So there is a view which was long popular among philosophers and still is subscribed to by some today, which says that basically what science is doing is we’re going out into the world collecting data, whatever that means. Maybe I’m looking through a telescope, maybe I’m picking up ants and putting them in jars. Once I collect enough data, that data becomes a theory about whether it’s about the motion of the sun and the planets or about electricity. And now that theory makes predictions. And I run experiments, I see if those predictions are correct. And the more correct predictions it makes, the more data validates the theory, the higher and higher and higher my confidence gets in that theory to the point where I say, I pretty much know that this is true. And your concept of explanations on this view, explanations for why the theory works, what is the fundamental reality, what really makes it tick. I can tell myself pretty much any story that I would like to tell myself, but the story is not really the point. The why is not the point. The point is I can predict that the sun is going to be there at this particular time and tell yourself any story you would like about why that is. As long as the predictions keep coming back right, my scientific theory has done its full job. What is wrong with that picture of a scientific theory?
David Deutsch
First of all, what you’ve just described is the common sense theory. It’s not just a common philosophical theory, common philosophical error. It’s the common sense theory of knowledge and how we interact with the world and how we obtain knowledge. Common sense says that we see things happening, then predict that they will happen again, and then optionally we may add an explanation of what makes them happen. Now, it’s hard to get one’s mind around the fact that this is nothing like what happens in reality. No one ever does that. No one ever goes out and looks at the sun rising in the morning again and again. For a start, it doesn’t. When it’s cloudy, you don’t see the sun rising. The regularity that even in this crudest, most frequently cited example of regularities, that regularity is only recognizable with hindsight once you have at least some kind of explanation. For example, that the sun is still there even when it’s behind a cloud, that the sun is still there even when I’m not looking at it. Most importantly, the sun is still there at night, because only then does it make sense to say that there has been a repetition of anything. The sun sets and then later the sun comes up again, so that is a second instance of having seen the sun. That’s another theory. That’s not a thing that you observe. It’s a guess that you make, and then after guessing you can test it. You can test it on the next day. Then the ancient myths about the celestial objects all assume that the objects are still the same when they reappear. Nobody has that theory about clouds, let’s say. They don’t say when a cloud goes away, if it rains again the next day and there’s a cloud, that’ll be the same cloud, which has changed its shape a little and now resembles a different animal. No one ever thought that, and they were right. No animal ever does this. No animal cares where the sun is or even that the sun is. As for the future, using it to predict the future, in fact the future is never the same as the past. That’s the same problem again. When we say that the future, that we think the sun is going to come up because it’s come up in the past and the future is going to be like the past, called the inductive hypothesis, it is not the case that the, how can I put this, it’s, most things we see around us are different from day to day, from minute to minute. I’m sitting in front of a computer, an hour ago I was not. If I had predicted that because I’m not sitting in front of a computer now, I won’t be in an hour’s time, I would have been wrong. What distinguishes those predictions of things being the same from the predictions of things being different? That cannot be induced from having seen it before because this particular thing has never been seen before, nor has any particular thing been seen before.
David Deutsch
We have to have a guess that it’s going to be the same because of the nature of what it is. I have a theory of why this computer is here and why I’m going to be sitting in front of it. That wasn’t obtained by induction either, wasn’t being obtained by thinking of this event having happened before. You might say, yes, but animals do this. If you feed the dog at a certain time of day, then it salivates. When you ring a bell or something, then it salivates. That is because genetic expectations have already been tuned to a particular regularity, such as when to find food, where to find food. Like I said, humans can look at a familiar thing and because of their explanation of it, they can predict that that familiar thing is going to be different from now on. Animals can’t do that. By the way, computers can’t do that yet either. One day they will be able to, but that’s the difference between having an explanatory theory and having a purely predictive theory. When we have a predictive theory, it’s always because we are gaslighting ourselves into forgetting that we or someone else in the past had to guess that regularity before they had the data.
Coleman Hughes
One problem with the purely predictive picture of science is that often the same data can be compatible with two different models of how the system is working. And you give the example in one of your books, which is a very interesting historic example of Galileo’s discovery and advocacy of the heliocentric theory of our solar system, namely that the Earth is not the center of our solar system, that the Sun is, and the Earth is simply a planet revolving around it. As you say in your book, the Catholic Church, the Inquisition had a competing theory, which was the official line of the day and which was something you were not supposed to contradict, which was that the Earth was actually the center of the solar system and everything revolved around the Earth in just such a way so that it looks as if the Sun is at the center of the solar system. That was their theory of why it looked so much like the Sun was at the center. Basically, the Earth was at the center, but just in such a way that everything revolved around it so as to create this superficial illusion that the Sun was at the center. So what you notice about the Inquisition’s theory and Galileo’s theory is that they by and large predict the exact same sensory data. They predict that Mars will be exactly where, in the same place when you look at it through a telescope at any given time. So how do you choose between two different explanations that predict the same data? There must be some criterion. What is that criterion?
David Deutsch
Again, if you think of the purpose of science as explanation rather than predicting data, then the first thing you have to do before you even look at the data is examine the theory to see whether it’s an explanation, whether it’s a good explanation. So for example, part of the Inquisition’s position was that the, as you said, the motions of the planets and the Sun and the Moon are exactly such as they would be if the Earth was going around the Sun. But in fact, it’s not. Now, the thing about that kind of theory is that it can be used to explain anything. The flat Earth people explain all observations about the Earth, including satellites and space travel and so on, with exactly that philosophical move. They say the Earth is in fact flat, but light moves in curves and the orbits of the Moon and other things move in exactly such curves as to make them appear as if the Earth were moving about its axis and around the Sun. So I have called that move, that rhetorical move, bad explanation because it can be used to explain anything. I think something that can be used to explain anything actually explains nothing. So when we try to explain, try and form a scientific theory about something, we have to first get a good explanation and only then can we test it. But in fact, we can’t test it even then because in reality, as people have pointed out, Duhem and Quine pointed out that you can always say, you can always make a slightly different move and say in a scientific theory, that it’s an experimental error. That it only looks as though it’s conforming to the theory, but in fact, there’s a different theory and that this actually holds true. This is true. You can always do this. Moreover, you should do this. If you only have one explanation and the observation violates it, it’s simply not the case that scientists or detectives or anyone who’s trying to find the truth changes their theory just because they find a counter example. It’s overwhelmingly likely, more likely, that the counter example is wrong than that the theory is wrong, if it’s a good theory, if it’s a good explanation. So the example I give in my book is a few years ago, an experiment with particle accelerator seemed to show that neutrinos were traveling faster than light and they announced it to the press and there was a sensation about this and people were wondering, have they proved Einstein wrong? Now, they should have known, everyone should have known, that this is absurd. You can’t conclude that something travels faster than light unless you have an explanation of how. You have to have a rival explanation to general relativity to explain how the neutrino can travel faster than light when nothing else can. The inability to travel faster than light, according to relativity theory, is not an intrinsic property of light or neutrinos, it’s a property of space-time, as is gravity altogether.
David Deutsch
You can’t say that light travels around curved space because it’s traveling in a straight line from its own point of view, but the neutrinos don’t. That doesn’t make sense. The curvature of space-time either is or isn’t the explanation of gravity. So the first, the reasonable conclusion from the result of the neutrino traveling faster than light is experimental error. In real life, again, experiments in physics are hard. Errors happen all the time and we generally detect errors precisely because something happens which the theory says can’t happen. When people discover a new phenomenon, it’s when they’ve ruled out as much as they can that the theory is true and they have invented a candidate for a better explanation. If you have two explanations, you can use experiments to perform a crucial test. That’s what the philosophers call it, a crucial test between the theories where one of them predicts one thing, another one predicts a different thing. Then you can test the two explanations. If you have one explanation and just a rival prediction, like neutrinos travel faster than light, you can’t test that. The theory that it was an experimental error will always be better. What would happen if neutrinos really could travel faster than light is that this would worry people. The results of the experiment would worry people and they would start making up alternative explanations of not only how neutrinos work but how everything else works, light, gravity and so on. Somebody would come up with a rival explanation. It too would first be tested theoretically to see whether it was a good explanation and then they could modify the apparatus to distinguish between those two explanations. In the event, it never got that far because they found that the reason was a loose optical cable.
Coleman Hughes
There’s another key thing for people to understand here, which is usually called levels of explanation in philosophy of science. Sometimes you’ll get people saying things like, am I a human being or am I just a collection of atoms? As if those are two alternatives, mutually exclusive alternatives where I can only be one or I can only be the other. Another way of conceptualizing this is if I ask you to explain the Great Depression only by referencing the motion of atoms, it would strike you as a nonsense question, although in some way it can make sense because the Great Recession was a collection of atoms moving, those atoms being in my body, in dollar bills that were coming out of banks and so forth. Whatever explanation you gave about the Great Depression or whatever I said, Great Recession, whatever explanation you gave would have to be compatible with the laws of physics. If you were to explain to me what a run on the bank is, nothing in your theory of what a bank run is can violate quantum mechanics and nothing will, but it’s also an explanation occurring at a totally different level. The units you’re talking about are human beings and motives and self-interest and money, whereas the units of physicists is talking about are atoms or subatomic particles and yet at some level you’re describing the same thing. You’re not describing two different universes, you’re using two different languages to describe the same thing. So can you talk about how do you think of levels of explanation
David Deutsch
In our universe? It’s not just two different languages. There are two different problems. Suppose you did somehow magically have access to a giant computer provided by aliens that could simulate the motion of all the atoms on Earth and you fed it with the data of before the Great Depression and suppose it predicted that there would be a Great Depression after this. And that would not explain the Great Depression. That would not even begin to explain it. It would merely predict it. And although there is a description of you as a collection of atoms and a description of you as a human being, it’s perfectly possible that there is no explanation of you as a bunch of atoms, only an explanation of you as a human being, because it may be that even if you followed through the motion of all those atoms and you ended up with a prediction of exactly what you will do, you still wouldn’t have explained anything. You wouldn’t have learned anything about what caused your behavior or the advent of the Great Depression or whatever. This idea that a real explanation or a real theory has to be about microscopic things and that everything about higher level things like humans and economies is just a crude way of referring to atoms. It’s just a prejudice. It’s a philosophical prejudice that comes from maybe it’s physics envy, as some people say, or maybe it’s misplaced empiricism, or maybe it’s just a historical tradition that for some time really excellent explanations emerge from studies of microscopic things. But whatever it is, it’s wrong. There is no fundamental reason why an explanation has to exist at a particular level of size of atoms or people or whatever. On the contrary, there is every reason to believe even within physics that explanations and higher levels and lower levels have an equal status. Of course, they’re not allowed to conflict with each other at any level. They’re not allowed to conflict with each other at the high level or the low level. But there may not be an explanation at a low level, just like there may not be an explanation at a high level either. It’s just a matter of what the best-
Coleman Hughes
So why is it that lower level theories, and by lower we don’t mean lower in status, we mean dealing generally we mean dealing with smaller units. Why does a lower level theory like physics seem to have so much more predictive power than a higher level theory like economics to choose something? It seems like physicists can predict exactly where a particle is going to be to the smallest unit of measurement of the universe, but economists can’t predict anything. I would say they can’t predict anything, but their predictions are far less reliable. Then you have things in between like biology, which seems like biologists can sometimes predict with more precision than economists, but certainly not with the precision of physics. So why isn’t it the case that the more predictive power and precision you have, the better your level of explanation?
David Deutsch
The main reason is that one cannot predict the growth of knowledge. When you are trying to predict the economy or the behaviour of a human, you are trying to predict what knowledge that system is going to create in the future. And predicting the growth of knowledge, as Karl Popper taught us, is fundamentally impossible. So the more the phenomenon that you’re trying to predict depends on the future of growth of knowledge, the less predictable it is. So that’s one difference, that’s one huge difference. Certain things are fundamentally unpredictable, because if we knew what the price of gold is going to be in a year’s time, it wouldn’t be that. Because if people could know that, they would make investment decisions that would make that prediction not happen. The other thing is that it’s a bit of a mirage to say that physics can predict everything very accurately. We know as a matter of theory, theory, explanatory theory only, that every air molecule in this room exactly obeys an equation that we can write down. But we can’t do an experiment on most of the implications of that equation. To do an experiment with a precise outcome that we can measure and make a prediction that we can then test, we have to go to enormous lengths to make the experimental conditions single out the prediction that we want to make. That’s why I said earlier, doing physics experiments is hard. So a physics experiment in the laboratory has an optical bench which is insulated against possible vibrations from lorries outside and from earthquakes if possible and so on. The lighting is made so that it can’t interfere with the sensitive photo detectors, so the main lighting is off and so on. The laser shines a light which is stabilized exactly to a particular frequency, being controlled by a computer that keeps it at that frequency. If any of those things go wrong, the outcome will be wrong. People won’t say, oh well, their quantum mechanics is false. No, they’ll say, oh, one of our precautions hasn’t worked properly. Let’s think very carefully what might have gone wrong. I remember a few years ago, I went to see a beautiful experiment being done by people doing an experiment on ion traps, which are traps which trap a single ion, a single charged atom inside the apparatus being held in place by magnetic and electric fields. They could manipulate that atom to increase its electron energy up to a certain level, then down and measure that, and so on. And suddenly it stopped working. I was amazed that, you know, because when I went to university, I was told that we’ll never be able to do experiments on single atoms, but now it’s commonplace. I said, you know, it suddenly stopped and I said, what’s happened? And they said, oh well, probably an air molecule hit it. Well, this whole thing was being done inside a high vacuum, but what we call a high vacuum still has millions of air molecules in it, and sooner or later, one of them hits the ion and that’s the end of the experiment. And you’ve got to work hard to reduce that, and if you can’t reduce it any further, how to work with it so that the thing that you measure is not affected by the perturbations caused by air molecules? So it’s not really true that physics can predict things with very high precision.
David Deutsch
It can predict things with very high precision that are very carefully prepared so as to be …
Coleman Hughes
Predictable. Okay, so now that we’ve laid the groundwork with the primacy of explanations in science, the fact that science is not just about predicting the world, it’s about explaining the world and predictions help us do that, help us choose between competing explanations. And we’ve understood that there are different levels of explanations which are compatible with each other, but none of them are second class in so far as they’re good explanations at their level. I want to talk a little bit about your interpretation of quantum mechanics, which is you’re an advocate of the many worlds or the Everett interpretation of quantum mechanics, which competes against mostly the Copenhagen interpretation of quantum mechanics. Now just to be clear, these are two different explanations for the same data, the same experimental result. And that experimental result is often called the double slit experiment. Many people will remember this from whatever physics course they’ve taken in life, but the double slit experiment is one of the strangest results that any scientific experiment has ever yielded in the history of human experience on planet Earth. It’s one of the most deeply counterintuitive, strange results that it essentially feels like you’re a child watching a magic trick and you’ve checked every possible way that the magician could be deceiving you, and he is not deceiving you in that way. So it seems actually like pure magic, like real magic. It’s difficult to describe just in an audio form, but I would encourage you to look up any five minute YouTube explanation of the double slit experiment if you haven’t seen it. But at the most basic level, particles can either behave like particles or like waves. And very simply, when something behaves like a particle, it’s like you’re throwing a baseball through a hole. It behaves exactly the way you think. It just goes through the hole and ends up on the other side. When it behaves like a wave, it’s like spraying water through a hole that is much too small. It diffuses on the other side like a wave. So if you imagine a baseball that certain times it’s thrown through a hole behaves like a baseball, other times it’s thrown through a hole behaves like water, and whether it behaves the first way or the other way seems to depend on whether you’re looking at it, whether you’re observing it. None of it makes any sense. So my question here is, do you have a simple way of explaining what the double slit experiment is? I know it’s a very tough and long thing to explain, but you’ve been doing this for years. So presumably, what is the simplest way for you to explain to a smart but uninformed person what the hell is the double slit experiment and what are the competing explanations for how the hell and why the hell it works that way?
David Deutsch
When people go and look this up on the internet, they should look for the single particle or single photon double slit experiment. Because if you imagine a torrent of photons going through the double slits, that doesn’t rule out some pedestrian ways in which the conjurer might be deceiving you. So just imagine one photon at a time, particle of light, but it can be done with other particles as well, electrons, neutrons, whatever. So one particle heads towards the two slits, and what do you see on the other side? You have a screen with an amplifier or something that tells you where the photon lands on the screen beyond the slits. And as you say, if the photon were not quantum mechanical, if it behaved like particles in our everyday experience, then you would see a slit-shaped shadow if each time it landed, it produced a mark, then you would see a shadow of the barrier. So that where the slits were, the photon sometimes lands, and where there’s a barrier and no slit, there’s just a shadow. So there would be a shadow with two holes. And indeed, if you have big holes and a lot of light, that’s exactly what you do see. Like if I hold up my fingers and have a lamp, then I will see a shadow of my hand with the same number of fingers as I have holding up part of the light, preventing part of the light from reaching the screen. But if we have a small screen with slits, and when I say small, I don’t mean atom size. You can do this experiment yourself if you have just a needle and a laser pointer, and you make needle-sized holes. Now needle-sized holes are millions of times as big as an atom or as a photon. So this is not microscopic. You can do this in your own home. You can do this experiment. It’s hard to do it with single photons because our eyes are not sensitive enough for single photons. But you just have to take my word for it that you get exactly the same results if you use single photons and photomultipliers to detect them. Now what happens when you do it with single photons is that once the slits get close enough together and are narrow enough, you no longer get a shadow of the slits. You no longer get what you would get if you just put your fingers in the way. Instead, you get a completely different pattern. Never mind what the pattern is. It’s called an interference pattern, but it doesn’t matter what the pattern is. They’re beautiful patterns. If you make two holes—by the way, it’s called double slit experiment. It’s easier to do it with holes, just a double hole experiment. You put a needle or a pin through a card and then shine your laser pointer at it. Switch the lights off a couple of meters away, put a screen, and you can do this experiment yourself. The strangest thing in the world. You see an amazing pattern. Make a third hole, you see a different pattern, a different pattern, and four holes, yet another pattern.
David Deutsch
The pattern that you see on the screen depends on the pattern of holes. Then you notice—the fact that it’s not a shadow of the holes that you’ve made. That’s one amazing thing already, because it shows that something is going through all the holes at the same time, even though you’re only sending one photon there. The more amazing thing—and this is the icing on the cake—is that there will be places on the final screen that receive photons when there is only one hole there and go dark when there are two holes. There are other places on the screen that receive photons are bright when there are two holes and go dark when there are three holes. Now that means that something is going through all the holes at the same time, because if it only went through one of the holes, it couldn’t detect whether the other holes were there or not. Something is traveling through each of the holes. No matter how many holes you put there, as long as you do this experiment carefully enough, the same phenomenon happens. You open up a new hole and some place on the screen goes dark. No shadow is like that. You can’t do that with shadows of your hand. With quantum mechanics, you can predict this. With any version of quantum mechanics, with any interpretation, the standard interpretation is what you call the Copenhagen interpretation. Though I think that’s probably not historically accurate, but it’s often called the Copenhagen interpretation. Anyway, with the standard interpretation, you can predict it, but if you ask how the photon got from the laser to the place where it landed, you will be told you’re not allowed to ask that question. This is a sure sign of the wool being pulled over your eyes. This cannot be the case that the explanation of something is that you’re not allowed to ask the question. So what is the answer? Well, sometimes people say, well, the particle is sometimes a wave and sometimes a particle. That won’t do, because whenever you put the photomultiplier anywhere at the screen, at the holes, after the holes, at the final screen, wherever you put it, when you put one photon through, you only ever detect one photon. So if there was some phenomenon of the photon becoming a wave, going through the holes, then going back to being a particle, if there was some phenomenon like that, then sometimes you would see a quarter of a photon when you had four holes. Sometimes you’d see half a photon, but you don’t. When you send in one photon at a time, the detector only ever detects one photon. It’s just a matter of where it detects it. The Everett interpretation is that whenever you do this experiment with a photon, there are a large number of copies of you also doing the same experiment, and they’re also sending photons through in other universes. And these photons, unlike you and the copies of you in other universes, the photons actually interact with each other.
David Deutsch
The ones from different universes interact with each other. And they go through all the holes at the same time. And when they get to the far end, to where the screen is, or the photomultiplier, they are interacting with each other. And so they can detect that they are affected by what the other instances of them were doing. So interference is explained as the effect of other instances of the photon in other universes on each other. This is the basis of quantum computation. In quantum computation, it’s not photons. It’s the states of atoms or whatever they’re doing, or ions or whatever they’re doing the experiment or building the quantum computer with. You can arrange things so that just as the photon could go through one of, let’s say, four holes, the computer can be performing one of four different computations, or one of two to the fourth different computations, because there are two to the fourth possible ways that the photon can either go through or not go through a hole. So if you had 100 atoms, and you set these 100 atoms to be doing all different computations, they would actually be doing two to the power of 100 different computations, one for each possible configuration of the atoms. All zeros then a one, all zeros then a one then a zero, and so on. So there are two to the power of 100 ways of setting 100 bits. And if you then run a computation, there are two to the power of 100 computations going on. And then you can arrange for those to interfere with each other and produce a final answer that depends on all of them. And that’s how quantum computers work. So I’ve told you about two different quantum phenomena, both of which are amazing.
Coleman Hughes
How is it that the other versions of you in parallel universes, what is the explanation of why the particles can interact between universes, but I have no interaction with my other selves?
David Deutsch
You do have interaction with the other selves. But unfortunately, you also have interaction with lots of other things like air molecules. And if the photons in the two-hole, two-slit experiment interact with something else on the way, then they’re no longer interacting with each other, or rather their interactions with each other are then different in such a way that you don’t get interference between them. So interference is prevented by these things in different universes interacting with something else. That’s called decoherence. So once decoherence has occurred, interference won’t occur. So when there’s a large object like a human, we’re decohering all the time. Billions and billions of decoherence processes are going on. And so the interference between ourselves and the other instances of us is suppressed. And that’s why we can’t see them. But if I proposed a thought experiment where in a quantum computer, if you could run an artificial intelligence, an AGI, artificial general intelligence, on that quantum computer, it could easily detect other instances of itself. And one day when we have AGI and also quantum computers, neither of which we have at the moment, then we will be able to do that experiment.
Coleman Hughes
So this pivots perfectly into my next topic, which is artificial intelligence. This is something you’ve written about in your book, The Beginning of Infinity, which you wrote, importantly, before the era of ChatGPT, Midjourney, DALL-E 2, MusicLM, and all of the other language model artificial intelligences that have sprung onto the scene in the past, let’s say, six months. So I know that you have thoughts about the significance of the Turing test, whether that is actually a good test for artificial intelligence properly understood. So let me just put that to you. Is ChatGPT an artificial intelligence? Or is it just some kind of clever program that has passed the original version of the Turing test? What do you make of our current progress on AI?
David Deutsch
I think to speak about this clearly, one has to distinguish between two completely different kinds of software, one of them called AI, called artificial intelligence, where the word intelligence is really slightly misused and artificial general intelligence, AGI, where the word intelligence would be properly used. AGI is general intelligence, that is, it is a form of computation with the same computational abilities as the human brain, or to be more precise, as the program in a human brain. The program in a human brain is a general intelligence, not artificial, and we could make such a program artificially, one day we will, and that would be an artificial general intelligence. At present, we have absolutely no idea how to make an AGI. We will one day, but I’m afraid that at present, everybody working on this thinks that an AGI is just a more powerful AI. In my view, for example, GPT and that sort of thing are all AIs, none of them is even remotely an AGI. What’s more, programming an AGI is considered as a programming task. It is qualitatively different from every other programming task. Every other programming task you can define in terms of a criterion, that is, it recognizes faces, or it produces sentences in correct English, or it tells you whether there’s a flaw in your program, or something like that, or it plays chess. So the criterion is doing that correctly. The art of writing a program to be an AI, all these AIs that I’ve described, is to prevent it from doing the wrong thing so that it homes in on doing the right thing, the correct thing. Now, an AGI, there is no specification for an AGI. If you say, for example, a person who chooses not to speak like a monk in a silent order is still a perfectly conscious being. If they decide not to speak for five years, they haven’t stopped having the property of general intelligence. The property of general intelligence resides in their thoughts, not in their output, not in the relationship between their inputs and their outputs. In fact, they might have no inputs and outputs. Supposing there was a religion that said, you’ve got to spend five years in a sensory deprivation tank, the person there would not be any less human provided they were doing it voluntarily. I suppose they might go mad if they weren’t. But a person can perfectly well think without having any inputs and outputs. So, a program to do that couldn’t be defined in terms of its inputs and outputs. I should say, in regards to the Turing tests, I don’t think Turing was aiming to set up a test of whether something’s an AGI or not. He wrote a paper in whatever it was, 1950, in which he was arguing that, as he put it, computers can think, though he really meant computer programs can think. He was arguing that that must be so because of what we now call Turing universality. That is, the behavior of any physical system can be represented with arbitrary accuracy by a program running on the Turing machine or on what we call a computer.
David Deutsch
Nowadays, to be perfectly precise, we’d have to say on a quantum computer. But I don’t think quantum has anything to do with the G of AGI, the general intelligence. So, the imitation game that Turing proposed in his paper, where a machine pretends to be a person and you converse with it via a teletype machine, and then if you can’t tell the difference, then what more reason have you got to believe that that thing isn’t human or isn’t thinking than you do with a human? So, that was the argument. It was a thought experiment that formed part of an argument. It was not a proposal for a thing which would, 50 years later, or whatever we are now, 73 years later, be used as a test of whether something’s really an AGI because, as a test, it’s completely flawed because, for example, a real AGI might decide not to participate and a real judge might decide, as is happening at the moment, a real judge in a sort of Turing test situation might be gullible and might decide that anything whatever is conscious or is thinking. And right back in the 1960s, when the first such program was written, ELIZA, people did indeed think it was conscious, even though it was a program of a hundred lines or so.
Coleman Hughes
So, with both AI and scientific progress in general, I would describe you as someone who is confident that we will make lots of progress and essentially solve most, if not all, of the problems that we spend time thinking about. One of those problems would be artificial general intelligence. Another one of those problems I would call the problem of consciousness, which is why is there anything it’s like to be this set of atoms? Why does it feel like something to be this set of atoms as opposed to the set of atoms in my microphone, which I presume has no subjective interiority, nothing it’s like, no qualia, to use a philosophy term. You’re confident that we are going to solve all of these problems, assuming we don’t destroy civilization or have some extinction event. I’m less confident because it seems to me that human beings don’t occupy the highest possible intelligence. We just occupy the highest intelligence that exists on earth. It seems to me it would be a remarkable coincidence if the smartest animal on earth were able to understand and solve every problem. It seems to me like every other animal has an upper bound in terms of the problems it can solve. Human beings logically should also have an upper bound unless we occupy the highest possible intelligence a being could have, which I think probably we are far from. I don’t see why that would be true. Can you defend your optimism regarding our ability to solve the most difficult problems: quantum gravity, AGI, the problem of consciousness, the other problems that have vexed science and philosophers since time immemorial?
David Deutsch
Yes. The argument can be labeled the argument from universality. The universality in question is at two levels appropriately enough. The lower level, as it were, is Turing Universality, which we’ve already mentioned. That is, there is such a thing as a universal computer. We are speaking to each other on very good approximations to universal computers. There is no computation that could be performed by anything in the universe that could not be performed by our computers that are on our desks at the moment, barring only speed and memory capacity. Apart from that, the computers that we use are universal. That wasn’t always true. I still remember when scientists went around with slide rules in their pockets, and the slide rule could perform certain computations, but not others. But Turing discovered that there is an escape velocity for computation, that there are such things as machines that can perform any computation that could be performed by any other machine, no matter how ingenious the machine is built, no matter how many extra types of chip are put in it. Apart from speed and memory capacity, which in this context is not philosophically important, a universal computer can compute anything. An argument that says that all computers that we’ve had before, all computers that exist in animal brains and whatever, all computers that exist in the DNA system inside our cells, they’ve all had a finite and tiny repertoire. What makes you think that our latest technology has an infinite repertoire, an unbounded repertoire? Well, it’s because of a feature of the laws of physics, which is Turing universality. Now, human brains are running a program which has Turing universality, but that’s not enough to be optimistic about the capacity of the human brain to solve problems, because quite likely there are animals whose brains are computers powerful enough to have Turing universality, if only we could input them in the right states. At the moment, we don’t have the technology and it would probably be considered unethical to try to use a dog’s brain as a computer. But anyway, it’s clear that humans can, because humans can in fact execute any program that a computer can execute in our minds or using a piece of paper. So that is not enough for solving problems. Now, when it comes to solving problems, we need an additional type of universality, which I have argued must exist, and that is explanatory universality. For explanatory universality, you need Turing universality, but you also need more than that. So, Turing universality is a matter of hardware. Some hardware has it, some doesn’t. And we have attained escape velocity in regard to hardware with our computers. Explanatory universality is a property that software may or may not have. And AI software doesn’t have it, but AGI software would have it.
David Deutsch
Now, how do I know that there aren’t levels of AGI software, so that there’s a certain level of generality you can reach, which can explain a certain range of things, but then there’s a better program which could reach? Now, this program must be runnable on a Turing machine, because everything is. So, we can’t guess that there are things we can’t explain because it’s at a level that our hardware can’t reach, because it’s only a matter of software. And since it’s a matter of software, let’s suppose, perhaps the best way of explaining this is, suppose that there are superior beings on another planet, and they come and visit us, and we ask them, okay, well, we’ve been curious about quantum gravity and about qualia and so on. And they say, oh, yeah, well, we understand that. But we can’t explain it to you because your brains aren’t suitable. So, we would say, well, what would you need to make our brains understand it? Now, they couldn’t possibly say, well, either they would say you need more memory and more speed. Well, we already have prosthetics, as it were, such as pen and paper or computers that can vastly increase our speed. And if we had faster ones, we could make it faster. And faster ones must be possible if that’s what the alien’s ability depends on. So, it can’t be that. So, then they could say, well, no, it’s not memory and speed. The only other thing it could possibly be is that your program isn’t good enough. Well, what program? Write the program down, and we’ll execute it the hard way. And then we will understand whatever you have understood the hard way. Well, they might say, well, it will take you all your lives to understand the least thing. Okay, well, make us immortal then. There’s nothing fundamental imposing an absolute limit on the number of computations we can perform in a lifetime. So, it can’t be that. So, these hypothetical aliens can’t exist. They would violate Turing universality, and therefore, we can assume that we have explanatory universality. That’s assuming, I’ve assumed in this whole argument, what you might call physicalism, that the only relevant features of us are that we are physical objects obeying certain laws, laws of nature. Now, it could be that you postulate supernatural objects that are not subject to laws of nature, but we can rule those out because all of them are bad explanations. So, QED, I don’t think that that is a flawed argument.
Coleman Hughes
All right, David Deutsch, thank you so much for coming on my show. Your books are the fabric of reality and The Beginning of Infinity. This has been a fascinating conversation. And if you have anything else you’d like to point my audience towards in terms of your recent work or a website or Twitter handle, now is the time.
David Deutsch
Well, you can, anyone who wants to can easily find me by searching for my name. I suppose the spelling D-E-U-T-S-C-H. Yes. So just find that. And the only other thing I have to say is it’s been really fun talking to you.
Coleman Hughes
Likewise, David. Thanks.
Markdown