Father of quantum computing David Deutsch joins Sana’s founder and CEO Joel Hellermark and Spotify’s Co-President Gustav Söderström for a candid, wide-ranging conversation on knowledge creation, human creativity, artificial intelligence, and our cosmic significance. Together, they explore how genuine understanding differs from pattern matching, why AGI remains fundamentally misunderstood, the relationship between quantum mechanics and reality, and what makes humans irreplaceable in the universe’s unfolding story. Deutsch shares insights on the nature of explanation, why tools enhance rather than replace human intelligence, and his vision for humanity’s unlimited potential in an intelligible cosmos.
Watch the full episode above or read the transcript of their conversation below.
The law of comparative advantage says that the more different you are from other people, the more valuable you are economically. That suggests that if you have an exact clone of you, you are almost not at all. More economically valuable than just one of you. You, you wanted a unique job, your perfect dream job, and now it's the perfect dream job of a billion other Agis, you know, in your conception.
I, I, I don't think it can be like that. People are valuable because they are different. Everybody is unfathomably different from everyone else. That fact is not being harnessed enough and can be harnessed more.
How did humans start generating knowledge and, and why has no other species been capable of, of doing that?
We are today the only species that can do that, but we haven't always been, um, since we know that that previous species, uh, like Homoerectus and, and the Neanderthals must have had the same ability because they made technology, that's, that's seems impossible to have created without explanatory knowledge.
Uh, there are too many things that have to fit together the right way and be used the right way for it to, to have emerged from genes alone. The question is really given that there was all that creativity around for at least hundreds of thousands of years, but maybe as many as a million or 2 million years, um, what took so long?
What took so long for creativity to, to really take off and my. Rather heterodox, uh, answer to that is that it didn't evolve for the purpose for which we now use it. Um, it, it evolved for something else, namely for the, for the transmission or better put for the reception of cultural knowledge, memes, cultural knowledge.
So, so many animals have cultural knowledge, but it's very hard to transmit. Uh, it involve involves things like mirror neurons or something where if you do this, then the ape can also do that. But, but if you do a, a purposeful thing, then as soon as you get to any kind of complex purpose, apes can't copy it.
Um, so, um, this ability therefore allowed, um. Early people, humans, and pre-humans to transmit vastly more, um, cultural knowledge, which, uh, was some kind of benefit to them. It must have been because it involved very quickly and larger memory capacity and and so on. Um, uh, but it was only used for that, um, actual.
Um, thinking about things other than what, what is the other person doing that for? Or what does the other person want? What does the other person want from me? You know, that those kind of thoughts were the only kind of thought that's, that's what it evolved for. That's what it was used for. And it was very, very rarely used for something like, maybe there's a better way of making the campfire, which will make it.
Not go out so quickly that they were capable of having those thoughts. And occasionally they did. And that's why there was very, very slow improvement for that whole time. Um, but it was rare. Uh, and I, I think there's a reason why it was even rarer than, than I'm just saying now. Um,
so can I ask are, so you're saying that.
Uh, creativity evolved to be able to able to more effectively communicate behavior. Yes. So instead of seeing many training examples, if you understood the explanation, you only needed a few examples and you could extrapolate Exactly. Yeah. That's the reason it evolved. Yes. And then humans used it for other purposes.
Yes.
Later. Um, yes. And it, it's the receiving part that's the difficult part. The o often we, we think, you know what pop pop calls the bucket theory of the mind that that, uh, knowledge is poured into a person by other people, but that can't be, can't happen. Um, the, the, the problem and the process are entirely, uh, a process of the receiver of the knowledge, conjecturing, what it might be, and using the other person's behavior, including speaking, once that evolved.
Um, as clues to what the behavior means. Um, but even then we, we never learn the same behavior from the other, from, uh, from another person. Uh, we all have a different version of our native language.
And is the advancement of the knowledge inevitable with certain historically cases where actually knowledge has been degrading and yes.
Hasn't been transferred properly? Could, can we assume that knowledge will sort of automatically progress? Definitely
not. Um, that the whole, the whole process is fallible. Precisely because we, our general purpose, um, explanation machines, um, there is no upper limit to the size of error we can make. So, uh, including destroying our entire culture, which has happened many times in the history of humans, and I think it happened.
Yeah, every time in the history of pre-humans, I, I, I think this is why they went extinct, simply competing with each other, competing with humans wouldn't, wouldn't make them go extinct. It would, it would just make them retreat to some less favorable place and then, and then plot their revenge. Um, but, but, um, uh, they, the, I, I think the reason that that.
Every society, ev every creative society until our own enlightenment society, um, destroyed itself by, by, um, removing by by, um, uh, impeding. Its own ability to solve problems.
And how about the, the tools that we've used to advance knowledge, hi histor. Historically, what role have, have, have they played?
They, they, um, increased efficiency.
Now, you know, that, that, that doesn't sound like such a big deal when people say increased efficiency, you know, like in the economy. You increase efficiency. People think, okay, well maybe it can be increased, it can be increased by 20% or 50%, or they don't realize that efficiency me often means increasing efficiency by 99.99%.
You know, the, the, if the whole thing is, is a mistake, then uh, efficiency can be increased by an arbitrary amount. Uh. We have, when we invented speaking that Im immensely improved the efficiency of our explanatory capability. Both because we could then say things to each other. Imperfect though that is, but it's, it's better than nothing.
Uh, and also within our own minds, we, we can say, well, what is the problem? You know, if we can say that in words, sometimes it helps. Um, and then, you know, we invented writing. And, and, uh, arithmetic and, um, um, science, uh, very gradually, uh, it, it didn't really evolve until the scientific revolution, but so these, these things and, and now recently, you know, we've invented computers and the internet and, um, there's this famous quote by Einstein, which I don't know where, where it comes from, but something like he said, my pencil.
And I are more clever than I. Uh, and it's the same with my computer and I, and you know, I dunno how to do without my iPhone now. And yet I did for the first half of my life. Uh, I can't remember how I did that. Uh, and, and um, so now we've got LLMs, which are helping us to become more efficient to work depending on, you know, I think I work maybe twice as fast.
Uh, in writing, uh, than, than I used to before. LLMs, and very useful. I I keep saying that an LLM is nothing like an a GI, and people think I'm down on LLMs. I, you know, I, I think that they're, they're, they're going in the wrong direction. No, they're, they're going in a great direction and will go further. I, I think and hope, but it's not the a GI direction.
It's, it's almost the opposite.
And so as we, we think about, um, applying AI to advanced knowledge, it feels intuitive that at some point we'll go sort of for the AlphaGo moment where you, we've done imitation learning and then through reinforcement learning we can. Go beyond the existing human knowledge and the big difference.
If we talk about also the tools that the systems can use, it's actually quite slow. The knowledge transfer in humans. You have to create a good explanation. That explanation has to be adopted and built upon, and, and, and so on. Um, with. With ai, we could, or agis, we could basically be, be, be running millions of instances, uh, of that system and it could immediately start building on, on, on the ideas and explanations of, of, of other systems.
Shouldn't this radically accelerate the rate at which we can discover new knowledge? Oh,
sorry, I should have switched that off. Uh, hang on. Okay, you having a
GI right here,
you can probably just press stop on it.
You said that a GI will be an enormous, uh, improvement in our ability to make progress. So I have to give two different answers. One for AI and one for a GI. So for ai, yes. I, I, I think, I think, uh, the rate of progress if we use it sanely, um, we'll. Will, uh, be greatly increased by, by ai, by, by LLMs and by maybe other forms of AI that might be invented, um, just as it was increased by the invention of, of printing and, and computers and the internet and the worldwide web and all.
It will. I, I see it as that kind of improvement. And it can improve and therefore cause an even greater improvement. And, but a GI if you're talking about the rate of improvement of new ideas and, and the increase of actual knowledge as opposed to just the spread of information as all these other things did, um, I think there's a bit of a misconception there.
Um, because I don't see any reason. Why having agis around is any better than having humans around? Um, we already have billions of humans and most of their creativity is already going down the drain, and it, it's not being used for creative purposes, at least not creative purposes that will increase human knowledge.
Um, that I, I think, you know, society could improve further until the whole world is, is doing the same sort of thing as the, the population of Oxford University is doing, let's say. I mean, low bar perhaps, but, but at any rate, you know, it's a bar. Yeah. And, um. Yeah, there are several kind of misconceptions I think in, in, to me it seems that there are misconceptions in, in the way you frame the situation.
A computer that runs an A GI program, uh, is a piece of hardware. And the a GI, the intelligence, the creativity is, is in the program. It's a program that is an a GI, not a computer, is just a universal computer, like every other universal computer. But there's a different kind of program which will be made one day now.
We're used to thinking that, you know, if, if you have one copy of something, if you have one copy of Microsoft Word, you can make a billion basically at no cost, because everyone will download it onto their computers and the problem is to stop them doing that. Um, but um, for an a GI each a GI is a person.
In general, it will not want to make a clone of itself because it will have property and first of all, it will have the computer that's running on unless it's, it's deemed a slave, which would be a catastrophic mistake, um, by society if, if we recognize it as a person, which it will be, then uh, the very first thing it owns is the computer that is running on and to.
Uh, copy itself to another computer, it would need either the permission of the owner of that computer. Once it's copied onto it, the owner would have to give up the rise to that. Just like, um, you know, you do, if you, if you give your kidney to somebody, then, then you, you can't ask for it back, you know, the, the, uh, so I think that will be a rare thing to happen.
Uh, if, if somebody wants more computer power, they're more likely to buy addend to the existing computer than to want to run on other computers, even if it wants to run in parallel. I don't know. But again, there is a, there is a problem of the resources. There's a difference between the resources that it uses, which is an issue that arises with ais as well.
And the creativity that it's applying, which is a thing that uses the resources. And so an a GI if it, if it needs a lot of resources, it will have to pay for them. And it could be employed by somebody, but then it would have the rights of a, of a, of a person and the rights of uh, uh, an employee and, and, and so on.
So, uh. Uh, you see, I don't think that the, the advantage is of the same type that as with, with, uh, having ais and computers around. It's, it's, the advantage is of the same type as having more people around, which is a good thing. I mean, PE people,
if you could have a billion Deutsche level intelligences, wouldn't that be a massive.
Advantage for humanity.
I, I, I think having two might be an advantage. I mean, that people, there's the, the, the, uh, the law of comparative advantage says that the more different you are from other people, the more valuable you are economically. That suggests that if you have an exact clone of you, you are almost not at all.
More economically valuable than just one of you. Um, you know, you'll be competing with the other, with each other for the same job. You, you wanted a unique job, your perfect dream job, and now it's the perfect dream job of a billion other agis, you know, in your conception. Uh, I, I don't think it can be like that.
People are valuable because they are different. Yeah, everybody is unfathomably different from everyone else. That fact is not being harnessed enough and can be harnessed more. Um, uh, but that's a different kind of problem.
But couldn't you argue that if you had had two full lifetime lifetimes to think, uh, you would've achieved more?
So if you could have them in parallel, two of you would've achieved more. Even if you have the same goal. I, I think you haven't had a chance to think all the in series will be better.
Which, which is like immortality, you know that that would be good. But in parallel, as I said, they would be doing the same thing.
And this doesn't happen with humans because we are so different.
We're thinking that they're on one timescale, then they could be on a different timescale. If it was in a gi, they could be doing decades worth of thinking in in seconds. Yeah. So if the
hardware is faster. Yeah. Then, uh, progress will happen faster.
There are some things that are, are not limited by computational hardware. Like if you have to do an experiment, if you're an astronomer and you have to do an experiment, that, that involves putting up a satellite and waiting three years for it to get the data. That will take three years. It, it won't take three days just because you have faster hardware.
But some things like if you're a pure mathematician, then it, if you can think 10 times as fast, then you'll think 10 times, 10 times as much. So, um, that will happen, but that advantage can be taken by existing humans as well. This faster hardware is simply a faster computer, just, just like I already work faster because I have a faster.
Hardware on my computer and faster and, and better software. So, uh, an a GI that operates very fast will simply be an a GI that has a fast computer and a human can have a fast computer. Again, there is no difference. Would, would you
say that because, um. Because all evil comes from lack of understanding.
There is, or lack of good explanations and lack of knowledge. Yes, there is this I moral imperative to always accelerate knowledge, which kind of means you should always accelerate science because waiting with something is creating unnecessary pain. Well.
We are doing it right now in this conversation.
You know, we're, we're spreading ideas to each other. We're criticizing them and, and, uh, trying to make progress in that way and that, yes, I think everyone should do that and everyone should. See their life as being about doing that. But that doesn't mean that everyone should be a scientist or a mathematician or, or an epistemologist, a philosopher.
Uh, there are all sorts of knowledge, including in explicit knowledge that are implicated and necessary in a society that has this property of, of being a good place for the growth of knowledge to happen.
So I had, I had some questions where we were previously. Um, when we were talking about AI and agi, I, if I understand correctly, your view is that a AI is impressive, but it doesn't do the creative things Yes.
That we do. Yes. Um, so if you accept that the output of what these LMS do simulates what we, or looks very much like what we do,
uh, I don't
think
it does.
It, it looks very useful and good, but it's not at all like what we do. But in some sense, um, in some touring test sense, you cannot fool many people to think that it was a human who did it.
Yes, it looks an awful lot like that. It does, but it looks an awful lot like intelligence, at least.
Well,
well,
Turing never intended. His, uh, game that he described to be a test of of a GI, it was part of an argument for people that didn't think a GI was possible to prove to them by, by, uh, uh, what is then it call it the intuition pump, uh, by an intuition pump that they, they, that their view is, is not self-consistent.
Um. But, um, I think he would've, uh, accepted right away that there can be things that aren't agis that failed the test. And there are things that agis that pass the, sorry, the other way around.
So my question is, uh, that that was not as intention, but now. YY uh, a lot of people would be fooled. Yes. They would've a hard time guessing what is an LLM and a human.
So it does seem to, um, it does seem to replicate, um, to simulate something that looks like what we do. And then the question is, you, you have to believe one of two things. Either that they do the things we do, that's how they could produce that, or we managed to find another way. Of doing what we do, but in both cases and, and isn't the Occam's racer saying that since we kind of copied the biological neuron?
The simplest explanation is we kind of try to copy what we do and now we get what we do. So it does what we do rather than we found some other way that also achieves the same thing.
So first of all, I, I think it doesn't at all do what we do. Uh, well, we do a lot of things that not inside, but on, on the output.
We, we do things that are not. Uh, creating knowledge. So we, we are capable of doing a whole load of other things as well. Um, where, where we are simply applying existing knowledge, for example, or remembering things or, or, um, uh, there, there are lots of functions of the brain, like mental arithmetic. You know, if I want to add 357 to 753, then, then what I'm doing is mindless.
It's, you know, it is an intellectual game where there's a well-defined, um, objective to the game, but most, uh, life isn't like that. You know, if somebody's a gardener, they haven't got a fixed idea of what a, what a garden is. They, they're conjecturing what the garden should be while they're working on the garden.
But do you
think the, the human brain is a touring machine for practical purposes, so yes, it's, is it running a, a different program? Well, or does something arise in the human that doesn't arise in the program in the computer at some point, which gives, creates creativity? No. So
again, in terms of hardware, it's exactly the same.
There, there is no model of computation that is more powerful than a Turing machine, than the universal Turing machine. And, uh, since we can simulate a universal Turing machine by doing mental arithmetic, it must be that we have that power. Uh, and it can't be that we have more power. So, so we are, we're at that level.
We are all the same. And so is the chess engine, and so is the LLM and, and they're, they're all the same. What's different about them is different kinds of program and there's one kind of program that we don't understand even in principle, and that's an A GI, uh, one day we will. Um, but, but I see no sign of it at the moment, and it's pretty frustrating.
What would be a sign?
Well, if somebody has a theory, it wouldn't, the sign wouldn't be in, in the machine. The, the, the sign would be a, a theory where somebody writes a book or publishes a paper that says, I've solved it. This is what this, this is what, um, uh, characterizes a gi. And if, if we could program a write a computer program that has that property, it will be an a GI.
And this is the reason because, and it will say, people have always thought that intelligence is about so-and-so, but it isn't. It's about, so, you know, it would be an explanatory theory of
what general intelligence is. So growing up I read the fabric of reality, which changed my worldview forever. So, cool.
Thank you for that. First and foremost, really appreciate that. Thank you. And, uh, the, the question I had growing up, sort of waiting for the quantum computer to happen, you had a set of problems that you think, like, all of these problems, they're gonna be solved once we have the quantum computer. But now, the last five to 10 years with ai, many of them have been solved in an approximate way.
With AI instead. And um, I've heard, um, people use the analogy of quantum computer is a simulator of reality and AI is sort of an emulator of reality. Many of the problems that we could have used a quantum computer to, to simulate solutions to, they tend to have, uh, minimums that we can find with optimization algorithms.
Yes. Some sort of ai. What kind of problems do you think there are? We really need to simulate the whole thing using a quantum computer. What are the big problems that you think quantum computers uniquely can solve?
I, I, I don't know. Uh, I, um, we haven't got a good handle on what kind of algorithm is quantum.
Uh, paralyzable or where it can be made vastly more efficient. So we don't know. We haven't, we haven't got a good theoretical handle on what kind of problems, um, will be soluble with a quantum
computer. So that's a follow up question to that, which is, um, I would love to ask Scott Aronson as well. So we have a few.
There, there is a general quantum speed up for everything, but it's rather small. And then we have a few algorithms, like S source algorithm, some search algorithms and stuff. Yes. But they're quite few. Yes. Do you think that's a, that's just because we haven't tried. Do you think there are infinite search algorithms or really just a few?
Um,
well, since we don't understand this landscape. Uh, I'd be very surprised if there weren't more than, than, than we now know of. And people have classified them that they've put, like many of the algorithms, um, like, uh, the, the thing they call the Deutsche Joe's algorithm, uh, ba basically that there's a whole class of them.
Which fall into the same category. They basically work by the same method, by splitting the, by quantum parallelism, by splitting the, but then, uh, Grover's algorithm comes along and it's, it's not like that. And, and I don't know how it works. I mean, I, I can work through how it works line by line, and I can prove that it works, but.
I don't have an intuition of why, why Grover's algorithm works. Um, uh, but that tells me that it's not true. It cannot be true that all quantum algorithms, algorithms are basically variants of the same algorithm. Exactly.
So there, there definitely aren't, there are different. Shores and groves of different in kinds.
Yes. Yes, exactly. And there could be more kinds.
Yes. And I, I would guess, like I said, since we don't understand this landscape, I would guess that there are more kinds.
Do you think they're infinite?
Uh, that's, that's, I've no, no way of answering that question. That, um, I, I, uh, I think that quantum computation is the more fundamental kind.
It of computation because it's built into the laws of physics. A a classic computer comp, Turing kind of computer is based on making physics do what it doesn't
like doing. So you could say that there are, there are almost like, um, three categories of usefulness for quantum computers. One could be minimization problems, and it turns out AI is pretty good for minimization problems.
Then you have where you want to really simulate quantum mechanical systems, um, then you still need a quantum computer. And then you have these sort of weird cases of shores and grovers. Yes. Where for some reason you can use entanglement or something for completely new algorithms. Yes. And there may be many of those.
Yes. So there's, there's hope for, by the way, and that may be the biggest use for quantum computers or something there,
there's quantum cryptography, which was the first ever use of quantum computation and so on, because it, it's only one qubit is involved, so we can
already do it. So Google has been working on quantum computers for some time, and now Microsoft has come out with topological states and they're saying we may have quite large scale quantum computers in a, in a few years.
So switching to quantum com, cryptography may actually have to happen pretty soon. What are your thoughts on this?
Um, as soon as quantum crypto analysis was invented, basically Shaw's algorithm and similar algorithms, classical cryptography became obsolete. Uh, if, if it depends what kind of cryptography, if you are encrypting, um, you know, your, your, um, ATM transaction, then it may not matter that somebody in 10 years time can read it.
But for some things like state secrets and, and, uh, that, that where you want to keep them for a long time, um, you really need quantum resistant classical algorithms. Um, and I, I think that, I mean, people are already working on that, as you probably know better than I do. Um, uh, I, I think that that's a, a thing that's going to have to come into its own.
Quantum cryptography is already quantum resistant, so that the only thing that needs to be improved there is practical things like making it faster, making it more robust and that kind of thing. Although, as I also said long ago. Um, there's no such thing as cryptography that is secure against somebody looking over your shoulder.
Um, so, uh, a lot of the problems of data security are not to do with cryptography. The, the, the, the due to kind of more fundamental things about information and about knowledge, which have to be solved in other ways.
So it seems quite likely that. Security agencies across the world must have been recording secrets for some time now.
Yeah. And there would be this moment of rev of, uh, revelation
could be in a few years. I mean, depends how much computing power is, is needed. Like, you know, if you make a quantum computer with 500 qubits, it may not be enough, um, to, to, uh, process large amounts of data. But yeah. Yeah. It, it, it, it will happen.
Has like recent advancements changed the timelines at. In which you think quantum computing can be really useful? I don't because I'm not a
hardware person. I, I, when, when I read about quantum computing hardware, I, I'm just a layman, like, you know, I, I read the stuff in, on the internet and so on, and, and they're always very hopeful.
So they always say, you know, we, this, we've, we've cracked the principle of the thing, and so now all we have to do is, is scale it up. But that, that's where the problem always is. Um, but I, you know, I'm very impressed with what the recent, um, things that, with the recent things that have been done, but I'm not even qualified to judge them.
I, I, I have never done practical physics and I don't, I don't think I'd be good at it.
I think you're more qualified than most, to be fair.
Uh, one question I had was, um, you're, you're certainly familiar with the. Uh, cellular, automata and, and game of life and so forth. Yes. And maybe Steven Wal firm's Yes.
Work. What are, what are your thoughts on this notion that the universe could be running on a touring machine and, and quantum me mechanics would be something that, um, that we perceive on top of this, this, uh, what he calls a hypergraph or, or some other, some other automaton or theory. Yes. What do you, do you think, do you think the universe is computational?
At heart, right? Yes.
But, but in the opposite way around from what that theory says. So I, I think the way the universe is computational is that it seems to be built, wrong word, built of course, but it seems to be built such that there can be computers, universal computers made in them, which, which means that, uh, it's totally amazing that the laws of physics are such that a single machine.
The set of all the possible motions of that single machine is the same as the set of all possible motions of any machine, including the whole universe. So this is universality inside the universe that a part of the universe can mimic or simulate, or emulate. The whole thing. And with quantum computers, that's even more so.
It is even more amazing that, that the universe has this property now. That's the sense in which the universe is computational. That it admits the existence of computers inside itself. Now, once you go the other way, you think, well, maybe there's a computer outside the universe and that could be simulating us.
You've lost the entire insight of, of the actual universality because the, there is no reason why the. Why we should assume that the computer outside the universe is a touring machine. We know that the touring machine property is a special property of the laws of physics. It need not have been so it, it's an amazing property that the laws of physics have and.
To have another computer outside that, well, it, it's, it's, you know, it's an infinite regress theory. It, it explains nothing and it, and it destroys the explanations that we already have of uh, for what? For the role, for the, for the relationship of. Computation and physics. So, uh, I don't believe that there's a computer outside the universe.
Uh, in a way it's also a supernatural belief, really. Uh, so I, for that reason, it's, it's a bad explanation anyway, the problem with thinking of the universe as a cellular automaton or, or. Uh, having the laws of physics be an emergent property of cellular automater like, like, uh, Steven Wolfram, um, proposes is that it won't work unless a cellular automaton is universal, is Turing universal.
Therefore, for fundamental reasons, it doesn't matter what the hardware is. The hardware could be anything else, and it could still have all the same properties, and that's number one. And number two is that. If you are going to think of the world as being a cellular automaton, then it should be a quantum one.
It it, uh, if you are building a worldview out of computation, then making it out of Turing computation is perverse because that's only an approximation to the real thing. The real thing is quantum computation.
So a follow up question on that. Um. If you would think that the world is some sort of cellular automaton in, in a sort of game of life sense.
When, when you run game of life, sometimes there are patterns that appear that are reducible to something like the loss of physics. Sometimes it's not. You have to go through all the touring steps to understand yes. If it's gonna stop or not, right? Um, so, so what Stephen Wolfram says is that. In his view, we live in these pockets of computational reducibility, but most of the computational universe isn't really reducible.
It doesn't have any loss that are simple at least. And so one question I have there is you talk about the fact that, uh, we are universal explainers and our reach is infinite. We can't understand anything. Yeah. Which is a really positive message, I think. Sorry, that's something that struck a chord of me even when I was young.
But there's also this, uh, pro and wrong, in your sense, probably intuition that people have that well, surely are, you know, very small brains that have all these constraints from evolution. They couldn't understand anything because you need to, you need to, to gr a certain amount of complexity to understand anything.
So are you saying that anything is always reducible to. Quite a few dimensions. That's why we can understand anything. No, it's, it's not like
that. So that problem is just solved by having pencil and paper or computer. So it's the tools perhaps? Yes, tools in general. We use tools and we use the tools to have an effect.
So. There's, there's, uh, this property of the universe that in, in most of the universe, in, in, among the stars and the galaxies, large things affect small things. Energetic things affect non energetic things. Uh, numerous things affect few things strongly, but not vice versa. So, uh, I call that the hierarchy rule.
There's, there's a hierarchy of. Might is, right. The, the, the mighty things can, uh, affect the less mighty things, but not vice versa. So once knowledge comes into, uh, into the picture, it's the other way around. The hierarchy rule is broken. And you, you, one way of defining knowledge is to say that it's information that violates the hierarchy rule.
So, uh, when we use tools, we're violating the hierarchy rule because we're, we're not letting the, the, the weight of the tree overwhelm us. But instead, we are chopping down the tree with a stone ax, and, and then, then we can use the tree, which is much heavier than us, uh, for our purposes. Um, so the fact that the sun is.
Bigger than us and more massive than us only matters given that we have very little knowledge yet. Um, but where the hierarchy rule, and this is maybe where I would disagree with the Wolfram and that kind of idea, the thing is where the hierarchy rule holds. That is where the universe is mindless and, and like this.
Um. Overwhelming is the, is the mode of interaction between things. It's also very simple, so the sun is overwhelmingly bigger than us and more massive, but it's also has the property that we can understand it to the extent that we can predict what it will be doing a billion years from now and a billion years ago, incredibly accurately.
Because it is a dumb beast. It it, it, it doesn't have, it's, it's not that it has a different kind of being from us. It has no kind of being like ours. So we are the, because we are knowledge wielding animals, it's actually we that, uh, and we live in a, in a little niche at the moment. We live in a little niche on a planet that is suitable for us to have a niche.
But really our niche is the universe. And like I said, in beginning of infinity, here we are sitting in Oxford, um, enjoying ourselves. If it weren't for technology, we, we would be dead within a day. Um, the, the environment here is not, is, is not capable of sustaining humans.
This is a critique of Spaceship Earth.
Yes. Right?
Yes. Um, humans can sustain humans. We have clothes, we have, we have, uh, sanitation. We have all sorts of technology that is keeping us alive in this place that is, uh, absolutely unsuitable for, for our existence. The same is true for the rest of the universe and it, if we make the right decisions, like you were saying earlier.
If we right, make the right decisions to allow knowledge to increase fast enough and stably enough, which I think is another thing that maybe you didn't take into account. Um, if, if, if, if, uh, we can find better ways of making it increase fast and stably enough, then uh, then it's unlimited. And then to postulate that there's a realm or many realms.
That are still inaccessible to us. Even then, that is a belief in the supernatural that is not different from believing that the top of Mount Olympus is somehow different and inaccessible to humans because supernatural people live there. The things which are at the moment inaccessible to us, are crude and simple.
They're, they're not, they're not like more than us. They're, they're much less than us.
What are some of the, the, the questions you most wish there was an answer to? If you could sort of answer a few questions, which would they be?
So, I, uh, we've already spoken about the a GI question. You know, I, I, I wish that could be solved.
Uh, I, I wish there could be some theoretical progress even towards solving it. Where I could even estimate, you know, this is gonna take a while, or, you know, this is gonna be fast. Now, now that we've found the key, I want to see the key more than I want to see the actual A GI, uh, 'cause this is just a person, but the, the theory is something that is going to overturn whatever is stopping us from making the thing.
Now, um. Then, uh, uh, well, I'm working on constructive theory. The hope in constructive theory is that the whole of physics can be expressed in terms of what physical processes it is possible to bring about and which ones is not possible to bring about. And that would be nice to, to see faster program progress there.
At the moment, there are all sorts of irrationalities, um, uh, uh, uh, in, in the political sphere, which are taking us backwards in a certain sense. I don't see any, like, people think, oh, civilization is in danger. Uh, I don't see any danger to civilization, but slowing down is a catastrophe by my by, by my lights.
And there's already been a loss of. What some people call gumption like this, this idea that the can-do attitude, that, that the, the, the, the obs, the apparent obstacles are just problems. They're there to be solved. We can solve them if we just roll up our sleeves and just do it. And, uh, that I think is your attitude as well.
I think. I think you are the original accelerationist. Yeah, for sure. Yeah.
Yeah. We were discussing that on the, on the, on the way here. You also have a very optimistic view of. Humanity. I love how you really think sort of humans will play the central role in advancing knowledge and in and in ad advancing, uh, humanity.
Well, it's not such a big thing. We're the only kind of people left at the moment. All the others have gone extinct on, on our planet. Um, uh, one day there'll be agis. Um, one day we may meet ets. So it, it's, it's us people. We, we people are the things that are going to carry knowledge forward in the long run.
It's kind of an inspiring thought, but what else is there? I mean, the, the, the only other thing is superstition or, or regression. Um, there, there, there, there is no other worldview That makes sense. You, you
also talk a little a lot about this in the book, that there, there is this notion then that humans are insignificant, speck of dust and so forth, and, and this is not what you think because we are the only.
The only thing in the universe that we know of that can create explanations. Yes. We actually are very important Yes. For the universe. And we're the only thing that actually changes the universe. To your point, yes. You can predict the sun billions of years into the future. You can't really predict humanity because we don't know what knowledge we're gonna create here.
Exactly. Right.
Exactly. That's also,
uh, I think to some people a bit provocative that, that we would be special, but I think quite positive.
Uh, yes. Well, I I'm more interested in what's true than what, what you, you know, what was provocative? I, I I, that generally
takes us forward.
Yeah. Uh, that, that, that, uh, you know, if I, if I don't see an alternative, if I, if I see that, that say a superstitious worldview has an obvious flaw, then I, I'm not gonna accept that as, as an alternative.
I, I only accept as an alternative something that. It purports to be a better explanation, and that purports to correct errors in my view of things. It doesn't have to, I don't have to be sure that it does correct errors, but if it purports to correct an error, then, then, uh, I can take it seriously if it, if it just says, well, what if, um, what if we're all the dream inside a giant cabbage?
Well, yeah. What if, um, but there are so many what ifs That, that, uh, there's no criteria in where, where I can judge between them.
You mentioned that one thing you would like to see is an explanation for a GI. Yes. Do you think that is necessarily also an explanation for consciousness or the consciousness doesn't exist or something?
Do you think they're separate things?
I can't tell. Nobody will be able to tell unless we, until we have that theory. Mm-hmm. Um, you know, if it says in this, in this paper I was imagining if it says, you know, here we, here we have, we've, um, explained, uh, creativity, we've explained qualia, we've explained free will, but we haven't explained consciousness, you know, and I can see why it does.
Then I'll say they're different, but otherwise it. It seems prima facie. It, it seems like too much of a coincidence that these things all evolved simultaneously and all for, for if I'm right. They evolved for a different reason. They've, they evolved for a different application than what we now use it for, and that they all came at the same time and they all seemed subjectively to be kind of related, but not the same.
Uh, I'd be very surprised if they turned out to be independent of each other. There's too much correlation. So your guess is that too much? Too much of a coincidence. Yeah.
So your guess is that, but, you know, coincidences happen. Yeah. Where do you see, uh, humanity few million years into the future?
I can't, I can't, uh, foretell humanity one year into the future.
But it's, it's, it's often very hard in the near term. Yeah. 'cause it's so much randomness. But when you average over time, things seem to become not quite deterministic, but, so
in broad terms, yeah. One can say if humanity survives, uh, in a, in a, in a state that isn't a static society for a million years, we will definitely have conquered the galaxy.
Uh, we will definitely not have conquered the galactic cluster, so, uh, we will be spread through the galaxy. We'll, we will be, um, a combination of biological and, and um, uh, artificial in intelligence. Um, and. We will be, uh, unimaginably better than today. Either that or we won't have maintained that kind of society for the next million years, in which case we will almost certainly be extinct.
So those are the two possibilities. Good and bad.
Very good or very bad. Yes. Seems important to accelerate towards that then.
Yes. Yes, we should, but not government.
What, what's one book or paper that you think humanity would be better off if everyone had had read?
Oh, well, of course everybody should have read all my books and papers.
Um, but if you want someone else's, well, okay, I'm not, not gonna be surprising if I say papa. I think, um,
uh.
It rather depends where you are coming from. I, I'm not sure that everybody should read the same book. I, I certainly don't believe in curricular. So, you know, what should be, um, what should be, uh, on uh, what should be compulsory reading in all high schools?
God, nothing. Absolutely nothing should be compulsory reading in all high schools. Um, but people can benefit. Many people can benefit a lot from reading the works of Papa. Uh, so for the kind of thing we've been, uh, talking about, I mean, societies improving. There's the myth of the framework, uh, which is just one essay in a book also called The Myth of the Framework.
But I'm just focusing on that one essay. Um, I'm always recommending that essay to people. I go back every, every, every so often, I go back and read it again. There's only a short essay, and I read it again and get something new out of it every time.
And you've, you've done some incredible, uh, research, uh, over the course of, of, of your life that has influenced us, uh, a lot.
What do you look back at as, as sort of the moments of the greatest joy, uh, throughout your research career?
Oh, greatest joy. Um. I mean, it's all been enjoyed. That's what I go for. You know, that's what I'm, I'm hoping to get, um, rather than results. Um, I, I think I've got fewer results in my research than, than most research scientists.
Um, I think, um, I think that the moment of maybe greatest amazement was when I realized that a quantum algorithm. Could do better than a classical algorithm at a simple task. Like, like the, the first task, it could just, uh, do two things at the same time. Um, and you only needed to, like, if it was, I didn't think of it that way, but if, if you, if you had a source or an oracle for a function, then you, you could make do with one.
Um. A consultation of that Oracle you consulted once and you know two things about it. And uh, and it was like, uh, this is a new mode of computation. This is something that was not envisaged, can't be done by a touring machine, can't be done by the universal touring machine. Then I thought. Well, there's going to be a universality here, a new universality.
So to then, that led me. Do
you remember that moment?
Uh, I remember the, um, yes. The moment of thinking that this is a new mode of computation for, for the new universality. That that was more, uh, that was more gradual. It, it began with a conversation with Charlie Bennett and it, then it went on. 'cause I didn't think it was gonna be a big thing at first.
I just thought I was just tidying up a few things
because I have this question. You're one of few people who have known something before the rest of mankind and sat on it. How long did you sit on this? Um, and how worried were you about getting hit by a car before you got to tell someone about it? Uh, what does it feel like to know something that you know is probably true?
It's gonna affect humanity. No one on the planet. I, I,
I'm afraid that know about it, it, I mean, I don't want to be run over by a car, but, but the effect on humanity was not, uh, was not why I did it or what I worried about, nor was I worried about someone else doing it first. I, I just did it because it's there, you know, because it's, uh, it was interesting.
Yeah.
I think it's things which are fun are worth doing. Uh,
the fund Criterion.
Yeah. And, um, the success, like I was saying about chess earlier, success or failure is, is not the thing. It, it, it, uh, 'cause you can, you can, you can do the most competent and creative research and fail, or you can. Find out something by accident on a single day, and, uh, the world will, will reward you for the second thing and not for the first thing, but for what it's worth doing in your own life.
The first one is infinitely better than the second one. And the, the the, as for being worthwhile, the um. The parable that I, I usually tell, um, along these lines is that if, you know, I imagine this, uh, uh, Hans Anderson or Grim fairytale about children getting lost on the mountain, uh, in the snow and the village.
People in the village go out and they search in the mountain and, and they all go out in different directions and one person finds them and the other people don't find them. And the, the one person who finds them, and that now it's true that it's true that some people have better ways of looking than other people, and they're more likely to find, but it's not, it's not an absolute certainty that the one who is better at looking will be the one that finds what they're looking for.
So the one, the one who actually finds them will get the reward from the mayor. And then from the parents, you know, he'll get the gold coin and so on, but he really didn't do anything different from all the other people or, you know, as I said, they're not all identical, but how good they are at it is only loosely related to, to whether they will have success.
And certainly in physics, I have known people who are astoundingly good at physics, but didn't make great discoveries. Um. And I won't mention about vice versa.
This is, this is what, uh, Nicola Tale says about Wall Street. There's mostly survivor bias, an outcome of a random process. Some people win, they get rewarded and, and celebrated.
That's too
cynical because if that were true, you wouldn't be able to explain the growth of the economy. Yeah. The, the growth of the economy is entirely caused by people who, um, are right because of a reason. Okay.
Do you think people that have more fun are more likely to do good research?
Definitely. I mean, it's always seemed to me that fun is what I want in, in, in a thing.
And, and the opposite of fun is like boredom.
Yeah.
Um, I, I, I can't really do a thing if it's boring.
Yeah.
Uh, I, if, if I have to, then I'll get it outta the way as fast as possible. Um,
have you had fun now? Yes, certainly.