2025-08-06 This Is World How Quantum Theory Leads to Conscious AI
Duration: 00:29:56
Transcript
David Deutsch
Consciousness is a feature of software. It’s a feature of certain computer programs like the ones we have in our brains. It’s not a feature of the brain as such. The G in AGI stands for General, an artificial intelligence, which can be programmed to perform any kind of cognition that any other entity that has cognition can perform. It couldn’t possibly be general if it couldn’t go wrong sometimes.
Maciej Kawecki
You mean subjectivity there?
David Deutsch
Yes. If quantum theory is true, then there must exist such a program. It will run on any hardware. Our minds can also run on any hardware. We just don’t know how to download it yet. But when we understand the architecture of the brain better, we will be able to convert the state of the brain into a functioning state of a universal computer. The way I would envisage the relationship between humans and AGIs to proceed in the future is that we merge into very much the same thing. We both consist of a core program that has the G property, the general property, and they both can use any convenient hardware.
Maciej Kawecki
Mr. Professor, it’s a great honor having you here. And my first question to you is, how do you define AGI? What AGI is for you?
David Deutsch
The G in AGI stands for General. And what we mean by AGI is an artificial intelligence which can be programmed to perform any kind of cognition that any other entity that has cognition can perform. So it can perform anything that a human mathematician could perform, subject to limitations of memory or whatever. And it can also feel emotions and anything else that the brain can do, it can do. It’s a one-to-one correspondence. That such a correspondence is possible between computer program and any physical object, including a brain, is a theorem of quantum theory. So if quantum theory is true, then there must exist such a program. It’s just very hard to write such a program.
Maciej Kawecki
AGI is what is around the corner of AI development?
David Deutsch
I don’t think so. I think AI and AGI are in many ways opposite functionalities. And AI is something that produces a given functionality very precisely. That is, it goes wrong very rarely. So if you have a program that plays chess, then if it loses a certain proportion of the time, then when you modify it, you are making the losing decisions less likely. So the better the AI is, the fewer possible outcomes there are of its functioning as an AI. And with the present AIs, if they produce obscenities or whatever, then you modify it so it never produces obscenities, and then it is a better version. With an AGI, it couldn’t possibly be general if it couldn’t go wrong sometimes, if it decided it didn’t like to play chess and decided it wants to play poker instead.
Maciej Kawecki
You mean subjectivity there?
David Deutsch
Yes. So subjectivity is yet another function of a cognitive entity.
Maciej Kawecki
Today we know that something is missing when we look at AI development. What is it? Is it physics, dimensions of the reality?
David Deutsch
Well, I think we’re only at the beginning of AI development. Even though I said it’s not AGI, it’s not in the direction of AGI, it’s almost the opposite of AGI, as I said, because AI narrows the possible outputs and AGI broadens the possible outputs.
Maciej Kawecki
Why are we in the opposite?
David Deutsch
To make an AGI, there must be no way of saying it will never have a certain thought. If there’s a certain thought that it can never have by virtue of its very construction or programming, then it is not general by definition.
Maciej Kawecki
Why couldn’t evolution have simply created minds that solve problems without any subjective experience at all? Why don’t our brains just solve problems in the dark, you know, without any inner awareness?
David Deutsch
Well, the process of evolution did solve many problems without any explanation. There are, you know, until there was…
Maciej Kawecki
Natural problems, but not about, you know, abstract thinking, abstract math, physics revolutions. So that’s what I mean.
David Deutsch
So there are fundamental limitations on the kind of creativity that evolution is capable of. One major thing it can’t do is it can’t jump over a conceptual gap. So if there is an organism which could do with a certain improvement, like if there’s a lizard and it would really benefit the lizard if it could breathe fire, like a dragon, then the only way that evolution can generate a new organism which can breathe fire would be to have a sequence of organisms in successive generations, which could come closer and closer to the ability to breathe fire. And every single one of those would have to be not only a viable organism in its own right, because it has to survive to reproduce, to be mutated and so on, but also be useful, because if it’s not useful, then its competitors who didn’t have that slight mutation would outbreed it. So it’s very difficult therefore for evolution to cross a conceptual gap like that. Whereas thinking, we can say, imagine that you had a flying car. Now the flying car won’t work if you just imagine it. I used to imagine all sorts of flying and swimming and underwater cars when I was a child. And at first I did not try to bridge that conceptual gap. But the way that humans do bridge it when they do is not by making something, by making a sequence of things that starts with the first one and ends with the one with the new ability. They think of how to bring it about. That is the explanatory part. And evolution doesn’t have explanations.
Maciej Kawecki
You claim all humans are universal explainers.
David Deutsch
Yes.
Maciej Kawecki
Is human brain capable of solving any problems?
David Deutsch
Well, universal explainers refers both to the capacity to generate explanations and also the capacity to understand them, which is really the same capacity. It’s the same ability. Now, this does not mean that given a problem, a given human brain or any brain existing at a certain time will manage to solve it. That’s a different issue. One of the things that prevents most conceivable problems from being solved is that nobody wants to solve them. Nobody wants to find an integer with a given property. Only very few people want to find whether the Riemann conjecture is true and so on. And there are far more potential conjectures. There are infinitely many potential conjectures, most of which nobody will want to solve. The issue is different. The issue is, are we prevented by the constitution of our brains from creating a piece of knowledge which could in principle exist? Not which could be produced, but which could in principle exist. And the answer is no, as I argue in my book, because if there were such a thing that could not be comprehended by humans, but which nevertheless impinged on us in such a way that we would want to solve it, then if that couldn’t be solved, then this is the same thing as believing in supernatural beings. Because then we say that we can comprehend the universe this far, but no further. And there is no explanation for why it’s this far and no further. And therefore, since the thing beyond the barrier affects us, we can’t therefore understand the things that are within the barrier either. So it’s a contradiction. So we either have unlimited capacity for explanation or we have no real capacity for explanation, because everything that we think we understand is actually an illusion, according to that theory.
Maciej Kawecki
If we have unlimited capability for explanations, why do some people struggle with basic math while others discover relativity? How can we all be universal when our abilities are so different?
David Deutsch
Well, people struggle with basic math when somebody is trying to force them to do it and they don’t want to. They don’t want to is the key to this state of affairs. They have something better to do, now at least something that they think is better for them to do. As I said, very, very few people in the world want to solve the Riemann hypothesis. Want to prove the Riemann hypothesis or disprove it or whatever. Very few people want to learn calculus. Very few people want to learn to solve simultaneous equations. Unfortunately, we have a tradition of education that is based on making people do things they don’t want to do. And under such circumstances, creativity is impaired. How much it is impaired depends on sheer luck mostly. Some people are forced to do exactly the thing that they would have wanted to do anyway. Some people are forced to do things where they can do their own thing and not be noticed and not be punished and so on. So many people get away with it. But most people, in the case of maths for instance, which I think would be interesting for a lot more people than it actually is, get put off by the coercion of it. So Einstein said something like, even a starving wild beast would put off its dinner if it were forced to eat it. Or something like that. He put it better than I just have. So this is true in general and it’s rather sad. And I think if you interview anyone who’s been successful in a given field, whether it’s, yes, but fields that aren’t socially rewarded. So not someone who becomes a professor, but someone who becomes very, very good at something. Let’s say making model ships out of matchsticks or something. You will find, first of all, that they’re very interested in doing it. You will find that nobody forced them to do it. You will find a whole constellation of things that are characteristic of creativity that happened. And usually that person will say that they were lucky to escape. And they were lucky to be able to continue this thing and acquire a great expertise in it. Because they very narrowly were stopped. Sometimes they can be stopped by being forced to do that very thing, like the hungry lion. What is the function of consciousness for you? I wish I knew. But note that consciousness is a feature of software. So it’s a feature of certain computer programs like the ones we have in our brains. It’s not a feature of the brain as such.
Maciej Kawecki
Amazing. In Pasadena, I talked to a Turing Award winner, Judea Pearl. And he said that for him, consciousness is the blueprint of its own software.
David Deutsch
That’s a very deep sounding statement. Judea Pearl is a very deep thinker on this very issue. He understands, for example, the difference between prediction and explanation. Very few people do. And he is a campaigner for explanatory ideas as opposed to predictive ones. Especially in the field of probabilistic prediction. So yes, I agree. But I’m not quite sure I can decipher that statement itself right now. But I’m sure he’s right.
Maciej Kawecki
Do we need to understand what consciousness actually is before we can build conscious AI? Or could we accidentally, you know, create conscious machines while still being clueless about how consciousness works?
David Deutsch
It is possible sometimes to solve a problem without having a full understanding of what it really took for one’s solution to work. A good example is the conquest of flight by the Montgolfier brothers. So they lit fires underneath this balloon and the balloon took off into the air and so on. And they had no idea how it worked. They did not know that they were making a hot air balloon. They thought that smoke had an inherent tendency to rise and that it was the smoke rising that pushed up the balloon. So now that’s in a sense that’s completely wrong. That’s the wrong explanation of how it works. But it contains truth. There is some truth in that explanation and some knowledge in that explanation. And that small amount of knowledge was enough to make the thing work. It would not have been enough to make it work a little better. For that you need a bit more understanding of what’s actually happening. And when it comes to a powered flight with the Wright brothers, nobody could achieve that until they really understood some aerodynamics. They didn’t understand the whole of aerodynamics, but they understood the science of aerodynamics far better than the Montgolfiers had understood the science of hot air balloons. Usually, usually we need to understand something at least up to a certain advanced stage before we can manipulate it, before we can make it do what we want. Understanding the world and manipulating it are intimately related and it’s not often that you get one without the other.
Maciej Kawecki
Talking about AI, today you are using the word knowledge. Can knowledge without experience still be called knowledge?
David Deutsch
What we call things doesn’t really matter. We can give it whatever name we like. We can call it Fred or whatever. I’m following the terminology of Karl Popper, who was my entry into philosophy when I read his books. His way of using the term knowledge is that it is a form of information. It’s a form of information that has special properties. So it can cause physical effects. In this terminology, the genes of an organism contain knowledge because it is that sequence of base pairs in the DNA that allows the animal to fly and a different sequence would not allow it to fly. That is non-explanatory knowledge. The knowledge in the Wright brothers’ minds was explanatory knowledge, which achieved a similar effect, but it was a completely different kind of knowledge and different kind of information. They are both knowledge. Over the years, I’ve gone through various definitions of knowledge. My current favorite is that knowledge is a special kind of information. Most information is not knowledge. My current view is that the best definition of knowledge is that knowledge is information with causal power. However, I’m not wedded to that definition. I would have said knowledge is information that has the property that it causes itself to remain in existence. That’s a definition that is suitable for the knowledge in genes and the knowledge in books. But it’s not that great for knowledge in a brain. Also, in constructor theory, my current definition is the most suitable, which is probably why I like it best.
Maciej Kawecki
What if AI develops ideas so alien that humans literally can’t understand them? Would it mean that human minds are not Turing-complete?
David Deutsch
That’s a composite question. First of all, there are plenty of things that we can’t understand. If you ask why is the configuration of sand on a beach what it is and why isn’t it something else, we have to say I don’t know. We could study it for a lifetime. We could say a few things about it. It’s got these runnels because the fluid dynamics works in a certain way when the tide goes in and out. But we still wouldn’t be able to explain everything about this. Similarly, if we look at things that people already have understood, like the meaning of a linear A inscription in Crete, nobody understands linear A. That doesn’t mean that there’s some limitation to the human hardware. Now, the other thing is, you said, would it mean that we’re not Turing-complete? Well, we’re obviously Turing-complete because we can follow through the instructions of a universal Turing machine with a piece of paper and a pencil. But if I may say so, the question should have been, does it mean we’re not universal explainers? And the answer would still be no, because you cannot tell. There is no test for whether a particular piece of software can solve a particular problem or not. Because it might not want to. So again, it’s rather like school again. We are protecting ourselves. That is one reason why we might not want to do it, but there is an infinity of possible reasons. So if somebody doesn’t speak, doesn’t reply when spoken to, it may be that they can’t speak or it may be that they don’t want to speak. Or it may be that they don’t want to speak today, but will happily speak tomorrow or whatever. There is no test. There is no test for these counterfactual properties, which are essential to being human.
Maciej Kawecki
Are you optimistic about humanity’s future with AI?
David Deutsch
Yes, so AGI, I think, presents no fundamental problem because people get this mixed up: an AGI is a computer program. It’s got nothing to do with the hardware. It will run on any hardware. Our minds can also run on any hardware in principle. We just don’t know how to download it yet. But when we understand the architecture of the brain better, we will be able to convert the state of the brain into a functioning state of a universal computer. And even if we don’t do that, we will be able to use the extra hardware of a computer to enhance our hardware speed. And we already do this. You and I are at this moment communicating, using computers that are performing billions of operations per second, which would take us, if we had to do the same thing with semaphore flags, would take us millions of years to do. And yet with this hardware, we can do it. And it’s the same with any other information processing task. So if you have an AGI that has the G, that has the capacity to create new explanations, then it can use as much add-on memory or processors as it needs to solve a particular problem. And so can humans. And the way I would envisage the relationship between humans and AGIs to proceed in the future is that we merge into very much the same thing. That we both consist of a core program that has the G property, the general property, the general explanatory property, and they both can use any convenient hardware.
Maciej Kawecki
But you know, last time two intelligent species shared Earth, humans and Neanderthals, one went extinct. Why should this time be different?
David Deutsch
Well, actually, we now think that there were more than two. There might be as many as six. And five of them went extinct. And it wasn’t because we murdered them. It was because they failed to solve the problems that were facing them. And during the Ice Age, every other species was reduced in numbers. Humans not only increased in numbers, but spread out over a much greater area. And that’s because humans were doing this thing of solving problems, whereas other species who had the capacity to do so chose not to. Why did they choose not to? Because of their culture. We know that as well, because our culture also impeded our progress for hundreds of thousands of years.
Maciej Kawecki
My last question to you is, if you could find just one answer to the question about, you know, the nature of the reality around us, what would it be?
David Deutsch
I think it’s maybe cheating because it’s more than one question, probably more than one question. But I think I would want to know what is the difference between a creative computer program and a non-creative one or a conscious one and a non-conscious one, or between one that has qualia and one that doesn’t, and so on. I think all those distinctions are the same, but I could be wrong. But that’s the question I would want answered. I think, unfortunately, too few people are working on this, because they think that AGI is a kind of advanced form of AI.
Maciej Kawecki
This is the biggest challenge for the humanity today, AGI?
David Deutsch
Well, there are practical challenges for humanity, like weapons of mass destruction in the hands of terrorists. But for philosophical problems or scientific problems in that realm, in the realm of understanding the world, I think that problem that I mentioned of consciousness and so on is the most important one.
Maciej Kawecki
Yes. Thank you very much for your time. And I keep my finger crossed for your Nobel Prize in October. Because the theory of quantum algorithms in the 100th anniversary of quantum mechanics, that’s the Nobel Prize achievement, and you know that, I’m sure.
David Deutsch
Yeah, rather a scary thought.
Maciej Kawecki
OK, well, thank you very much. Thank you very much for your time. It was a great pleasure and a great honor.
Markdown