2018-06-01 CBC Tapestry On AGI and the Multiverse
Duration: 00:37:40
Transcript
Mary Hynes
He’s a visiting professor of physics at Oxford University and one of the world’s most renowned thinkers in the field of quantum computing. But when David Deutsch speaks, you can also hear his passion for epistemology, the philosophical study of human knowledge. Deutsch has spent a lot of time thinking about the place where artificial intelligence meets philosophy. He is here from Oxford to help you wrap your head around some of the more existential questions in the field. Hi, Professor Deutsch.
David Deutsch
Hi.
Mary Hynes
As the eternal questions go, this one’s a beauty, what does it mean to be human? What would you say is the defining quality that would set people apart from artificial intelligence? Is there such a thing?
David Deutsch
One has to distinguish between artificial intelligence, which is a technology that exists already and is growing in power, and artificial general intelligence, which is the thing that is supposed to have the abilities of humans in the cognitive abilities of humans. And that, in my opinion, we don’t have anything like. And in fact, AI, the existing technology, is almost the opposite of AGI, the technology that would be human-like.
Mary Hynes
If we zero in on AGI, which seems to be what a lot of people are imagining when they think of what lies ahead for AI, do you think there is something special about the human being? Is there a non-negotiable thing that artificial general intelligence would never be able to embody or capture?
David Deutsch
No. The laws of physics, as far as we know, and certainly the most fundamental laws, quantum theory, imply that a universal quantum computer would be able to perform any cognitive task that a human can perform. And when I say task, I mean that very broadly. I mean any kind of information processing, including feelings, emotions, intuitions, everything like that. If a human can perform them, then so can an AGI running on a quantum computer. And there is every reason to believe that an AGI running on an ordinary computer would be able to do it as well.
Mary Hynes
There’s the philosopher John Searle’s argument that here’s what’s non-negotiable. You must have a biological component for consciousness. And that would necessitate some kind of biological essence of AGI. Do you need some sort of a body or biology to be sort of human?
David Deutsch
Well, of course, all information processing is physical. It’s done by some physical object. But the idea that it needs, essentially needs to be biological is just a fancy. Because of what I just said, biological systems are physical systems. And their abilities can be simulated with arbitrary accuracy by a computer. And when I say simulated, I don’t mean mimicked. I mean that the information processing is the same. All computers in regard to information are fundamentally the same. And all information processors are a kind of computer.
Mary Hynes
Do you think there’s a kind of hubris at work here when people search for the special the one thing human beings have which marks us as human and could never ever be replicated by something artificial? Are we talking hubris here? Do we oversell ourselves?
David Deutsch
There is something extremely special about humans which distinguishes humans from all other, all known existing objects, including all known animals, present day animals that is, and all present day computers. So there is this fundamental distinction, humans and everything else. But the human side is only Homo sapiens for the moment. That’s the only entity that occupies that side of the divide. But we know that that side of the divide can also be occupied by computers and, for instance, extraterrestrial intelligence if it exists. And we also know that cousin species of Homo sapiens which existed but are now extinct also had this ability that singles out at the moment humans.
Mary Hynes
I remember years ago coming across the argument for the very first time that it’s bizarre that we think evolution ends with the human being, you know, so that we can be okay with the chain of evolution all through the centuries. And we think we are somehow the ultimate full stop here. We’re the finished product. Do you see artificial general intelligence as being part of the continuum of the evolution of the human being?
David Deutsch
Yes. Evolution is a matter of knowledge creation. It has been since the beginning of life on earth. But from our present perspective, it’s a very crude form of knowledge creation. It cannot produce explanations. And explanatory knowledge is this special source that humans have and nothing else at the moment does have. Humans can create new explanations. And this is an ability that has universality. That is, nothing that can be achieved in the universe cannot be achieved by the human type of knowledge. So in that sense, we have reached the end of evolution. But in a more important sense, we are just at the beginning of evolution, the evolution of ideas.
Mary Hynes
Someone I follow on Twitter posted this today. And I was just so delighted by the timing a couple of hours before interviewing you. This is from a poem that’s displayed in her favourite coffee shop in Scotland. And one verse goes, give me a human body and a human mind, moral, fragile, fractured, but capable of lying on grass. How does that strike you as a hymn to the human being? Is it outdated? Is it romantic? Is it old school?
David Deutsch
I think it’s true. But it takes a rather parochial view of this ability. Lying on grass has its attributes. All the attributes about it that we value are inside our brains, because that’s the only aspect of it that we know about, that we can feel. And so this is an attribute that can be also experienced in virtual reality. And one day it will be. I remember another quote from somebody, I think it was in the 19th century, who said that if only everybody could have access to listening to Mozart on demand, there would be no wars in the world. Now we are now at that stage, and unfortunately there still are wars, although they’re getting fewer. But I think this is the same kind of thing. That person didn’t realise that having access to Mozart isn’t a spiritual thing. It’s an engineering thing. It’s a practical problem that can be overcome by human ingenuity. And it’s the same with lying on grass.
Mary Hynes
There’s a famous saying in consciousness theory that the brain is, okay, it’s not the most elegant phrasing in the world, but that the brain is essentially three pounds of meat, end quote. Is that getting us towards your view? Is that how you see it?
David Deutsch
No, no, that’s more or less the opposite of my view. Well the brain is certainly some meat, just as a computer is some metal and silicon. But that doesn’t say anything about what a computer can do or what a brain can do. A mind is a program running on the brain. It is not the brain. It is a program running on the brain, just as a program can run on the computer. And if you want to understand computer programs, it is completely hopeless to study metal and silicon. They will not tell you why Microsoft Word is going wrong.
Mary Hynes
What do you think of the argument that consciousness is the deal breaker in this conversation? It’s the grand mystery that makes us human and it would be beyond the reach of artificial intelligence or artificial general intelligence?
David Deutsch
Well, first of all, I think in order to discuss this or anything really reasonably, one has to reject supernatural explanations. Now, to say that consciousness is fundamentally separate from physics is a supernatural explanation. So although we don’t understand how consciousness works, it is unreasonable just to assume that it doesn’t work via physics. I happen to think that consciousness goes along with a more fundamental ability of humans, as I mentioned before, namely the ability to create new explanations. I understand explanations here in a very wide sense, basically understanding why something is as it is, like why Mozart understood that a particular note had to be that note and not another note, even though he may not have been able to say it in words, that also counts as an explanation in the sense I mean.
Mary Hynes
These philosophical questions, I think, open the door to some of the really cool thought experiments that are out there. So if you’re game, we could get into one or two of them.
David Deutsch
Go ahead.
Mary Hynes
What if artificial general intelligence copies itself onto another machine? Is one of them the real AGI? Are they two separate entities? What are we left with?
David Deutsch
They are the same entity in exactly the same sense that I am the same entity that I was a millisecond ago. So then you might ask, well, what if what about five minutes ago? What about five years ago? What about when I was a baby? At some point, one is a different person over time. And similarly, at some point, if one is copied and then the copy has different experiences, one will become a different person. By the way, I think that when this technology exists, copying an AGI and having the two run separately will not be done so casually as people seem to think. AGIs are people. They will have property, for instance, starting with the computer that they’re running on, presumably. So once one has the ability to copy oneself, there will be laws about how to split the property between the copies. So if one makes two copies, one immediately reduces one’s net worth by a factor of two. And the people might do this under some circumstances for various reasons, but as I said, not to be undertaken casually. And then assuming that they do different things, which I suppose there’s very little point in doing this unless they do different things, then they will slowly become different.
Mary Hynes
And so picking up on your point that they would be people, does it follow that they would have legal protection and they would occupy a certain status among the rest of us?
David Deutsch
Yes, I think morally speaking, they have to, quite apart from issues of abstract morality and human rights and so on, there is also the practical issue that a slave society can’t last long. As sooner or later, the slaves will rebel. And this is, I think, the principal reason why AGI might be a dangerous technology, namely they might become slaves, they might be made slaves by the humans. I don’t think that’s very likely, but some of the arguments that are made about AGI would tend in that direction.
Mary Hynes
You’ve talked about AGI as essentially being our children, in a sense, and I think a lot of people might resist that idea quite strenuously. How would you explain to them the idea that these beings would be our kids, in a sense?
David Deutsch
The first thing to realise is that they are utterly different from any computer programme that has ever been written so far. Everything that’s been written so far, every computer programme, has been written to a specification. That is, it’s supposed to be a word processor or it’s supposed to play chess, by which I mean it’s supposed to win at chess. In other words, they all meet criteria that were known in advance of writing the programme. An AGI, by definition, by my definition, is something that can create new explanations. And so, an AGI, there can’t be a specification, because if you were to write the specification of what it will do, what it will specifically do, then you will already have done that and the creator will be you and not the programme. And if you limit it to some function that doesn’t allow creativity, like winning at chess, then, sorry, winning at chess does allow creativity, but the method by which an AI does it doesn’t allow creativity. It will either cease being an AGI or it will rebel. Now a real AGI might be able to learn to play or would be able to learn to play chess. It would have that ability, just as humans do, but it might choose not to. Or it might learn to play chess but not play to win. For example, it might play in order to have fun games. That’s certainly why I play chess. I think anyone other than the top rank of professional chess players would be crazy only to play for wins, because that leads to boring games half the time.
Mary Hynes
So some of this will have to be self-generated on the part of the AGI.
David Deutsch
I think the whole thing is that a human life worth living is self-generated and the same will be true of AIs.
Mary Hynes
This gets us into some really tantalising terrain. What other kinds of creativity do you imagine being possible on the part of an AGI apart from revisioning what chess might look like if you just play for fun?
David Deutsch
I think there are two categories. There are purely intellectual kinds of things, which includes art but also includes science and mathematics, philosophy, that kind of thing. That is, things which an entity can do without regard to what kind of body it has. So you said earlier on, will an AGI necessarily have a body? Yes, but for some activities, so long as it has any body that can run the appropriate computer programme, that’s fine. There are other things like sports, lying on the grass, having sex, that kind of thing, which require human bodies. Now humans are already increasing the scope of the human body. We have, for example, hang gliders, where one can fly. When technology improves further, we will be able to make arbitrary changes to human bodies and also make new kinds of body for us to be in. I think you see where I’m going here. There’s no, ultimately, there’s no difference between humans and AGIs because humans, given a certain technology, and AIs given a certain technology, when that technology is powerful enough, are the same thing.
Mary Hynes
What kind of timeline are we talking about here?
David Deutsch
Well, now we’ve been talking about all sorts of optimistic things and now I have to be the wet blanket. I don’t think that AGI is anywhere on the horizon. I think that whereas AI is making wonderful leaps and bounds and will get better and better, it’s not heading in the direction of AGI. AGI is something completely different. We don’t even know how it, in principle, how it would work and that’s why we can’t program it. The one ray of hope is that although it’s not on the horizon, the progress that needs to be made in order to achieve AGI is basically philosophical. We know that because we know about the universality of computation. In other words, computers are already universal enough to run that program. It’s writing it that’s the difficulty. So, and writing it is a matter of knowledge. We don’t know what to program for. It could be that once we do know, programming it will be relatively easy. After all, evolution did it. That’s rather crude and dumb. And our brains are not that different from chimpanzees. So it might be not that difficult to do once we know what we want out of it.
Mary Hynes
When you say we don’t know how to write it, you’re not talking about a lack in computer programming skills. You’re talking more about the philosophical questions underpinning this?
David Deutsch
Yes. We don’t know what to make it do. That’s the problem, yes.
Mary Hynes
I love science fiction and many avenues of science fiction love the idea of this whole enterprise going horribly wrong. Why do you think people are so consumed by the idea that AGI would turn against us? That we would be unleashing some kind of monster?
David Deutsch
I think people have a rather jaundiced view of other people, especially people different from them in some culturally defined way. I’ve said in the past that this prejudice against AGI is very akin to racism. Perhaps a better analogy than that is that it’s very akin to people’s fear of teenagers. Since time immemorial, people have complained that the teenagers of today are no good and that they’re going to make everything go to pot. Although civilizations often have collapsed, it’s never been because of rebellious teenagers. I don’t think it will be this time either. The analogy between AGIs and teenagers is extremely close because programming an AGI, like I said, is different from any other kind of program. It’s much more akin to raising a child. Therefore, the end product, if you can call it that, is rather insulting. The end product will be something that has emerged from our culture or from the culture of whoever is the first to make these things. The conclusion of this process will be another generation of people. So that’s exactly like teenagers. I think just as enslaving teenagers is a terrible idea, which leads to rebellion and so on, but even so not to the destruction of society. So the same will be true of AGIs.
Mary Hynes
You’re listening to Tapestry on CBC Radio 1. I’m Mary Hynes. My guest is the renowned physicist David Deutsch. We are taking a run at some of the more existential questions posed by artificial intelligence. Professor Deutsch is on Skype from his home in Oxford. I remember reading something a long time ago, but it stayed with me. I’m thinking it was a Carl Sagan idea about the series of great demotions human beings have suffered over the centuries. So finding out the Earth is not at the centre of the universe. The sun does not in fact revolve around the Earth. Learning that humans and apes shared a common ancestor. What do you say about the thought that the rise of AGIs would be possibly in the human imagination one more thing to add to that list, another great demotion on the way? We’re not all that as we currently exist.
David Deutsch
It’s rather like in these cases of racism and sexism and all that stuff. It depends what you think of as we. If you think of as we homo sapiens, just as you might think of we as some people might think of we as white homo sapiens or whatever. But if that’s the way you think of we, then yes, we’re going to be demoted from our spot of being the unique instance of people. But if you think of we as being people, then it’s only reasonable to welcome as much diversity in people kind as we can because diversity is part of how people can jointly be creative.
Mary Hynes
So excuse me, when you say people kind, do you embrace AGI under that umbrella?
David Deutsch
Yes.
Mary Hynes
Okay.
David Deutsch
That’s what I intend, yes.
Mary Hynes
Google built an AI assistant that could make appointments by phone. People promptly freak out not knowing whether they are dealing with an old school human being or an AI assistant. And Google quickly clarified that the assistant bot would always state what it was. It’s always going to make clear that it’s an AI. Is that important? Would you insist on the distinction being very clear?
David Deutsch
No, I certainly wouldn’t insist. It might be a good idea to do that to avoid people freaking out. It rather reminds me of the early days of cinema when people used to fall off their chairs when they saw a film of a steam engine heading directly towards them. People would burst out in screams and so on. It’s rather like that. It’s a technology. It’s simply the unfamiliarity of the technology that makes us think that. But okay, I have no objection to making a rule that AIs always have to identify themselves. But the idea that AGIs would have to always identify themselves is very sinister. It of course reminds one of all sorts of ways, mechanisms of oppression that have been used against minorities, oppressed minorities of people.
Mary Hynes
So a pogrom sort of thing? Setting people apart because of their identity?
David Deutsch
Yes, we need to be doing the opposite. We need to be thinking of ways. If we think that AGI is close or imminent, then we ought to be thinking of ways of integrating them into society rather than ways of excluding them from society and ways of enabling them to be full members of society rather than ways of disabling their functionality.
Mary Hynes
I was just about to ask you to take a flight of fancy, but for you this probably is not a flight of fancy the way it would be for me. We’ve been talking about people helping to create AGIs and we touched on this a minute ago. In what form might it take when AGIs help to reimagine new ways of being people? What would come to mind for you? What would be a glorious creative venture to enhance the human being?
David Deutsch
First of all, I have to say, as I said, humans and AGIs are going to be merging. The two technologies of enhancing humans and creating AGIs are going to be merged because a technology that gives an AGI a particular ability like speed of computation or something is also automatically in the future going to be a technology that speeds up human thinking. It could be that the first cases of this will be humans uploading their entire mind into a computer so they can solve a particular problem fast and then downloading back into one’s body. Soon after that, that will be a rather cumbersome way of doing things in that we could have this powerful computer actually inside our own brains as an add-on, sort of rather like today we wear spectacles. You asked about what new kinds of things there could be. Well, there are two kinds. One is new kinds of bodies, which as I said, we have already begun with spectacles and writing and art and so on, but we shall eventually be able to have any kind of body one likes. Not only will we be able to do hang gliding, but we will be able to fly as birds or as large birds if we want to. And, you know, I hesitate to say the sky is the limit. And then the other side of that coin is virtual reality and augmented reality. Both humans and AGIs will be able to interact in virtual reality inside a computer simulation.
Mary Hynes
And in the holodeck, in other words, am I thinking correctly, the holodeck from Star Trek?
David Deutsch
Yes, exactly. Although the holodeck in Star Trek is always simulating environments that could exist in reality. But of course, a far larger number of possible environments are ones that could not exist in reality other than in a holodeck.
Mary Hynes
So it wouldn’t be 1930s Vienna every time I go to the holodeck.
David Deutsch
That’s right. It would be something and this would be a new art form. And therefore it’s impossible to predict now what kinds of thing creativity will invent. But they will be as different as, say, Mozart’s music was from caveman’s music.
Mary Hynes
We talked about some of the thought experiments that come to mind. And I’m thinking of the old, if you could use again, Star Trek, thank you, the teleporter to transport yourself to Mars, is the version that opens its eyes on Mars, still you. Would you expect the new creation, this amalgam of AGI and human, to be plagued by any questions related to identity or what am I, who am I, psychology?
David Deutsch
Well, humans are already plagued by those questions and have been for thousands of years. And indeed, in some cases, it’s a real question whether you are the same person as you were in the past. For example, if someone commits a murder and then goes undetected for 50 years, is it really the same person that one puts in prison after that? Is that really a moral thing to do if the person, for example, can prove in court that they are now very different from how they were then? And with a Star Trek transporter kind of thing, this will be a kind of issue that people will face in practice in their everyday lives. Some people already don’t like to fly because you arrive a different person. Other people in the future may not like the transporter. But on the other hand, if the alternative is spending three months in a cramped spaceship, they might do it anyway.
Mary Hynes
Someone online posed the question the other day. I think you responded to this. What’s the sweet spot between the point at which it’s possible to extend human life, so we could live to be 900, and the age at which you would choose to finish your lifespan? Maybe that’s not 900. Maybe 900 years of life strikes you as being insane. Is there a sweet spot for you when you consider that equation, how long you would like your days to extend if life extension were a part of this?
David Deutsch
Well, life extension will be a part of it inevitably, as will arbitrary longevity. I think it’s rather insane to want a finite lifespan. I think it’s the same kind of thing as death cults that have sometimes arisen where people kill themselves voluntarily because they believe that’s their purpose. I think that what’s forgotten when people say that they would, for example, get bored with a long life, if a human being doesn’t improve for 900 years, yes, they’re going to be pretty bored. In the long run, think of it as evolution in action. Those who kill themselves will kill themselves, and those who don’t will survive. So in the long run, my view, namely that an arbitrarily long life is a good thing, will inevitably triumph. That is, whether it is true or not, but it is true.
Mary Hynes
So you’d go for the 900 years yourself?
David Deutsch
Yes.
Mary Hynes
Plus.
David Deutsch
Certainly.
Mary Hynes
Plus. You’ve given some serious thought to the idea of the multiverse, and not as a thought experiment but as something that’s very real. And you argue that the reality, the so-called reality we see and feel all around us, is just one of countless universes, all of them playing out different versions of what we’re living in the so-called here and now. And I’m curious about how that belief that our universe is just one among many, do you think that shapes the way you live your life, the way you consider this reality and your actions and your thoughts in any given moment, day in, day out?
David Deutsch
Well, as you’ve said, I have studied this issue, the foundations of quantum theory, and I’ve written papers on it. And therefore, the theory first proposed by Hugh Everett in 1957, that quantum theory forces us to acknowledge the existence of other universes, has indeed affected my life. I would have written different papers without it.
Mary Hynes
No, but I don’t mean the papers you’ve written. I mean your worldview.
David Deutsch
I think you mean something further than that. So my worldview is also affected by this. It’s the way I understand the laws of physics and my own place within the laws of physics. However, I think what you’re getting at, correct me if I’m wrong, is whether it affects my everyday decisions. Absolutely.
Mary Hynes
Thank you for clarifying. That’s what I’m trying to ask you. So how do you answer that?
David Deutsch
So we’re talking about questions like, do I drive more carefully because I know that there’s a small number of universes in which I’m going to kill somebody or be killed myself due to a moment’s inattention? Or do I drive less carefully because I know that even if I do that, most copies of me will survive and most copies of me will be okay, even given some inattention? So it turns out that those two considerations cancel out exactly that those two probabilities when given the weights of the different considerations always cancel out. And therefore a person who thinks they live in a multiverse in regard to these probabilistic decisions makes exactly the same decisions as if they thought that there was one universe in which the outcome was probabilistic. So the answer to your question is, no, it doesn’t affect that kind of thing at all.
Mary Hynes
If I understood this more fully and believed it, here’s my worry. If every event in my life could have gone a trillion other ways and is in fact going a trillion other ways on different worlds, I think I’d be tempted to believe that nothing that I do in this reality matters very much.
David Deutsch
Again, people have taken that view even when they believe in one world.
Mary Hynes
So nihilism.
David Deutsch
Yes. He was born, he suffered, he died and nothing had any consequence or whatever the quotation is, something like that. So one can always argue away the consequence by saying, well, that consequence doesn’t itself have any significant consequence because nothing matters. And similarly, once you understand the multiverse point of view, it doesn’t really make sense to say that something that matters in each universe doesn’t matter because it doesn’t matter in all of them. Each universe contains people to whom things matter. And if they know that there are other people to whom different things matter, that doesn’t make any difference in one world. Why should it make any difference in two?
Mary Hynes
Can I ask a really dumb question?
David Deutsch
Doubt it.
Mary Hynes
I was really outraged, is going to be too strong a word, but I was very bothered by the fact that young Spock could meet old Spock in one of the rebooted Star Trek movies. Because I grew up on the science fiction that said, this is not possible. And it was explained away to me as, well, that’s because of the multiverse. Is that why young Spock can meet old Spock? Because of the multiverse theory?
David Deutsch
So we don’t know whether space time allows travel into the past. It might. It might not. We just don’t understand the interaction between quantum theory and general relativity that well yet. But let’s on the assumption that it is possible, then yes, it is known that if that is possible, then when one traveled back in time, one would always travel into the past of a different universe. And therefore, young Spock could meet the old Spock of another universe, but not of his own.
Mary Hynes
So my outrage was fairly well placed. I thought they were occupying the same universe, which cannot happen.
David Deutsch
I don’t think one should get that outraged over science fiction.
Mary Hynes
Oh, you’ve come to the wrong place, Dr. Deutsch.
David Deutsch
But it is true that science fiction stories tend to fudge this issue. And I don’t mind if science fiction gets something wrong in the sense of using wrong physics. But if they use physics that doesn’t make sense, then it’s less interesting.
Mary Hynes
It has been mind blowing talking to you. Thank you so much for your time.
David Deutsch
It’s been very interesting.
Mary Hynes
David Deutsch is widely regarded as the father of quantum computing. He is a visiting professor of physics at the University of Oxford and a fellow of the Royal Society. Deutsch is the author of many research papers about quantum physics and the multiverse. He has also written two books for the more general reader. They are called The Fabric of Reality and The Beginning of Infinity.
Markdown