00:00:17.040 Hello again and this time welcome to chapter 7, part 2.
00:00:23.040 This is artificial creativity. The first part is very much about artificial general intelligence,
00:00:27.600 so we've got creativity which is from our minds and that is the creativity that allows us to
00:00:32.400 create explanations as well as create all the things that appear to be rather uniquely human.
00:00:38.560 So for example art and poetry and music. But importantly,
00:00:42.800 explanatory knowledge like science and philosophy and mathematics, the things that allow us to
00:00:48.480 really make progress into the future and of course that does include the products of our creativity
00:00:53.680 like art, art requires explanations in order to improve towards some better standard,
00:01:00.160 objectively better standard. We're going to get there as well in a later chapter.
00:01:04.800 So that first form of artificial creativity is very much about trying to replicate
00:01:11.920 artificially the creativity of our minds, what our minds can do. It's always surprising to me how
00:01:19.680 people think that there is a genetic propensity for things like mathematics let's say.
00:01:26.240 Your mathematical ability, as often said, is genetic in some way.
00:01:31.200 This is what psychology at the moment, the prevailing view is that IQ intelligence
00:01:37.440 somehow has a genetic component and this flows through to things like
00:01:41.840 capacity to do mathematics. There's some inherent thing about brains that allows them to either
00:01:47.760 be good at mathematics or not so good at mathematics, which would mean if that's true if
00:01:52.880 they're truly as a genetic component to intelligence or to maths in particular that would mean
00:01:57.760 that have to be genes for that capacity. This can't be true. Although there are genes for
00:02:04.240 brains there are not genes for minds or what minds turn out to be.
00:02:10.160 Minds of course are universal explainers. There's unfortunately an entire field of academia
00:02:16.960 devoted to a misconception. It's called evolutionary psychology. Evolutionary psychology
00:02:23.440 purports to be about how we have inherited certain mental characteristics.
00:02:29.280 If we've inherited certain mental characteristics then this means that those mental characteristics
00:02:34.240 must somehow be in the genes. To me this seems like a category error. It's saying that abstract
00:02:42.240 mental capacities are somehow coded in the DNA. This is extremely unlikely because vast
00:02:49.040 is the DNA molecule is. It's not vast enough to contain all of the information that is inside a
00:02:55.200 mind as the mind begins to learn. And so in the same way that we cannot expect that there would
00:03:02.400 be a gene for the capacity to say speak English, there cannot be a gene for the capacity to
00:03:09.680 do mathematics. More or less all English speakers have the same proficiency. Yes there are differences.
00:03:17.680 Some people write books, some people are great orators. Other people kind of mumble a little.
00:03:23.680 But overall we understand each other. We're about the same level. I don't think there are vast
00:03:29.840 differences. There's probably like a 5% difference in vocabulary between the people who have
00:03:34.880 have the greatest linguistic dexterity and between the people who struggle to form a sentence.
00:03:41.360 But generally adults can understand one another. No matter the language that they're using,
00:03:46.080 unless it's a specialization. But again if it's a specialization such that you understand a lot
00:03:50.480 of medical or legal terminology, that's not in your genes, you have to learn that.
00:03:55.920 In the same way that it cannot be the case, that the capacity to speak English is in the
00:04:02.240 genes because it doesn't matter what culture, nation, you come from. If you come to an English
00:04:07.120 speaking country and you stay there for long enough, you will learn the language. Similarly,
00:04:12.160 if a person born in an English speaking country moves to China, they will eventually learn
00:04:16.560 Mandarin. In neither case can the language be encoded in the DNA. The capacity to speak either
00:04:23.120 language be coded in the DNA. What is coded in the DNA is a capacity for language generally.
00:04:28.960 And so it is with mathematics or for any other activity of the mind. That particular
00:04:36.320 activity of the mind, mathematics, doing poetry, understanding science, the list is very long.
00:04:43.840 None of these individual things can be coded in the DNA. Only the general, the most general
00:04:50.480 capacity can be coded in the DNA. Namely, the code for a brain which can run a mind which is universal.
00:04:56.560 That's it. That's it. Once you're universal, then you have a capacity to do everything else.
00:05:02.640 So why are some people better at maths than others? Or better at speaking English than others?
00:05:07.120 Because they show a great interest in doing those things than others. Is that interest
00:05:11.520 coded in the DNA? No, I don't think so. It's a combination of the way in which people are
00:05:16.480 brought up, what grabs their attention as they grow into maturity. People's interest change over time
00:05:22.000 as well. So artificial creativity has to be about trying to find the algorithm for whatever a
00:05:30.080 universal explainer is. This very, very special capacity that only, so far as we know, human beings
00:05:37.360 possess in the entire universe. Okay. So that's artificial creativity in terms of mental intelligence
00:05:44.400 or something like that. The second part that I'm going to talk about here and now is artificial
00:05:51.040 evolution. There's two sorts of knowledge, remember, that David has distinguished here in the
00:05:56.000 beginning of infinity. There's the explanatory type knowledge, and that's the knowledge that we are
00:06:02.000 able to create as human beings. And there's knowledge in the DNA, the knowledge that codes for
00:06:07.600 organisms. We are not able to, at this moment, because of our own lack of knowledge, artificially
00:06:14.960 create either. And the reason we can't artificially create either is because we don't understand
00:06:20.960 either sufficiently well. And I mentioned in part one that there were parts of this chapter
00:06:26.640 that I didn't fully understand the first time I read the book, or the second time, or the
00:06:31.920 tenth time I read the book. It really took me until last year, many, many years after the book had
00:06:39.600 been published for me to really grasp what was being said here. And it really was exciting once
00:06:46.320 I did figure it out. I didn't figure it out. I had to speak to someone else about it. I totally
00:06:50.880 understood that we didn't understand consciousness. We didn't understand how explanations were
00:06:56.400 generated. And I knew that because if you can't program it, you haven't understood it. I knew
00:07:01.200 that. But I just couldn't figure out why we couldn't understand evolution. I understood we had
00:07:06.480 evolutionary algorithms. And David even talks about evolutionary algorithms in the beginning of
00:07:11.200 infinity. So I was missing something. I was missing something big. I myself played games with
00:07:17.840 simulating evolution by natural selection. And I'll put a link in the bottom all about this.
00:07:22.800 You can play games where you set, like, for example, the number of rabbits in a particular
00:07:26.320 environment and how often they breed and how much food they have. And you allow them to produce
00:07:31.600 random mutations now and again, such as their fur color changes. Then you can introduce wolves
00:07:36.640 into the environment. And the wolves will eat the rabbits or not eat the rabbits depending upon
00:07:40.880 what color their fur is. And so you can simulate this kind of evolution by natural selection
00:07:45.360 in a computer. So I thought, isn't that programming evolution? And aren't the people who design
00:07:51.600 evolutionary algorithms? Aren't they programming evolution? No, not really. So I really thought
00:07:59.200 that not only we, I thought that I had a good handle on this evolution by natural selection
00:08:04.560 thing. I even remember in second year maths, I did a subject called continuous dynamical systems,
00:08:10.080 which was basically about something called differential equations. That aside, there was
00:08:14.000 maths applied to biology and it involves something called the logistic equation and more complex
00:08:20.000 versions thereof. And the logistic equation allows you to determine the growth of a population
00:08:25.520 given the amount of food in an environment. If there's too much, if there's just the right
00:08:29.920 amount, or if it's being restricted. So you can make predictions using mathematics and graphs.
00:08:35.760 And I thought we understood all that how populations grow into client. Again, I thought he's
00:08:40.960 biology being predicted by mathematics. I thought we understood Darwinism because didn't
00:08:45.840 Dawkins himself figure out the rest of the genetic details? Indeed, didn't Richard Dawkins himself
00:08:50.800 write a little program that simulated evolution. Now, this is going to be terribly self-indulgent
00:08:55.680 for a moment, but I just wanted to lay with a point and be terribly self-referential for the moment.
00:09:01.440 I certainly didn't major in biology, but I was always really keen at university when I had
00:09:05.520 the opportunity to take biology subjects, or at least subjects related to biology. I took a
00:09:11.200 brilliant subject called astrobiology. Astrobiology was great. I took a subject about astrobiology twice,
00:09:18.000 I was once in undergraduate, and then later on in postgraduate, when I used a book like this,
00:09:23.600 an introduction to astrobiology, which is a great book. One time I took a subject with great
00:09:30.560 astrophysicist to now lives in Australia. He began an America's called Charlie Line Weaver.
00:09:34.960 And if you're interested in astrophysics, or or astrobiology, look up Charlie Line Weaver on Google
00:09:43.920 and have a read through some of his papers. He's a great polymath. He visited a degree in history,
00:09:49.360 which is unusual for a physicist. And the subjects he's written about professionally in astrophysic
00:09:54.800 cover dark energy and dark matter. Why planets are the size that they are? That one's called
00:09:59.680 the potato radio. She can look that up. Why did life appear as soon as it possibly could on earth?
00:10:04.720 It appeared much faster than might otherwise have been expected. He created something called
00:10:09.600 the nasalization quotient and analyzed the length of animal trunks. And he did that to make a
00:10:14.400 point about how brain size is in a convergent feature of evolution. Anyways, the list is long and
00:10:19.200 interesting and his papers are really easy to read for any layperson, compared to typical physics
00:10:23.040 papers I mean. So anyways, Charlie taught me astrobiology and evolution. I got confident that I knew it.
00:10:30.000 Then I did our biophysics from another guy called Joe Wolf at the University of New South Wales.
00:10:34.400 And if you're interested in learning physics, well, look up Joe Wolf, University of New South
00:10:39.360 Wales, because he's got some brilliant pages, web pages on the basics of physics all
00:10:44.240 out to modern physics, including quantum theory and relativity. And what Joe said of the subject
00:10:50.400 of astrobiology was basically that astrobiology was the science closest to theology,
00:10:56.640 because after all, they're both yet to demonstrate the existence of their own subject matter.
00:11:00.880 So in biophysics, I learned even more biology and really thought I was getting across
00:11:04.960 this biology stuff. And I was excited to learn more about biology and I finally took a
00:11:09.600 philosophy subject called the philosophy of biology with Professor Michael Michael,
00:11:14.080 again at the University of New South Wales. He had a degree in zoology,
00:11:17.280 but he inspired me to take lots of subjects and logic and classical philosophy, which is how I
00:11:21.120 got devoted off into philosophy. And so in the philosophy of biology course, I read lots of
00:11:26.080 Dawkins and I read lots of Stephen J. Gould and about their debates. And so I thought, this is,
00:11:31.760 this theory, I mean, there's a few unknowns around the edges, but surely, surely we know it
00:11:36.960 as well as we know Newtonian mechanics or general relativity or quantum theory, it is the biological
00:11:43.120 equivalent of that, surely. But despite reading the selfish gene and the extent of phenotype,
00:11:49.200 I didn't know what I didn't know. And this was the wonderful thing.
00:11:53.360 It's possibly a problem with taking formal courses of this sort and you gaining some small amount
00:11:58.480 of so-called expertise in an area. You overestimate yourself, especially if you've never challenged
00:12:03.280 yourself. And this is where the beginning of infinity came in. It challenged me about what I thought
00:12:08.560 I knew. I didn't know. And to be fair, no one does. But at least David Deutsch knew what others didn't,
00:12:15.360 or at least he had a way of diagnosing how to know that no one else understands either.
00:12:19.840 As a bit of a footnote for what it's worth, my lecturer, Michael, that I mentioned, who did the
00:12:25.520 philosophy of biology subject, he must have known that we didn't know much about evolution
00:12:30.720 by natural selection. He must have known that what was known about evolution by natural selection
00:12:35.760 contained huge gaps. Because he published a book just a couple of years ago, I'll put it on the
00:12:41.920 screen now. And in it, he states clearly, the big gaps we have in our understanding of evolution
00:12:47.600 by natural selection that comport with what is said in the beginning of infinity. So I guess I
00:12:52.560 should have known. I didn't listen well enough during the lectures, apparently. I might mention
00:12:56.960 something else here in terms of not learning the right lesson. It's kind of like I suppose how
00:13:03.600 people can go through an entire physics to ground learners, much as possible about quantum mechanics
00:13:08.880 and how to do all the calculations and never quite accept or appreciate that it does imply
00:13:14.240 the many worlds interpretation. It implies that there are parallel universes, that there is a
00:13:18.640 multiverse. The parallel to that situation is me studying what's called the milliuri experiment.
00:13:27.200 I apologize, this is clearly a very long introduction, but bear with me just for one more moment
00:13:33.920 while I mention, again, why I didn't realize what I didn't realize. If you study astrobiology,
00:13:40.880 so astrobiology is about the conditions required in order for there to be life out there in the
00:13:45.840 universe somewhere. And so we might be best to look for life out there in the universe somewhere.
00:13:50.960 It would be a good idea to have a place to start a cable, probably in a start on Mars,
00:13:54.480 but beyond that, we might be a good place, would it be a warm pond, would it be beneath the surface,
00:14:00.080 etc. One of the first things in any astrobiology course is usually something about the milliuri
00:14:07.200 experiment. The milliuri experiment was done back in the 50s and the idea is pretty simple.
00:14:13.040 You take a flask and you fill it with chemicals inorganic chemicals that you think were present
00:14:19.280 in the early earth. So things like oxygen, nitrogen, carbon dioxide,
00:14:25.760 perhaps some methane, perhaps sulfur, etc. You can make up your own list of chemicals and this
00:14:30.640 experiment has been repeated again and again, so I think they're just generally called
00:14:34.320 milliuri type experiments now. And what milliuri did, these were the first guys that did it,
00:14:40.240 was to take this flask, seal it, heat it up a little, and pass electricity through it. That was
00:14:45.760 to simulate lightening and to leave it for some time. And then after a certain amount of time has
00:14:51.360 passed you, open up the flask and you investigate what's inside. Now at first, what was found
00:14:56.800 inside of this flask offered great hope. It was quite exciting. What they had managed to create
00:15:02.640 out of the nitrogen and oxygen, methane, etc were amino acids. They got people got very excited about
00:15:08.800 the fact that amino acids had been produced inside of this flask. They're excited about producing
00:15:13.520 amino acids because amino acids are the very building blocks of proteins and proteins are what
00:15:19.920 living organisms are made out of. They're the things that DNA actually produces. DNA or genes
00:15:26.080 code for proteins. So if you're creating amino acids, surely leaving it for even longer, perhaps
00:15:30.960 changing or fiddling with the conditions a little bit inside of your milliuri flask, you might get
00:15:35.760 proteins. And so you get proteins, then you might get nucleic acid. You might get something that's
00:15:39.520 self replicating. You might get DNA. You might get something crawl out of the flask eventually.
00:15:44.400 This is what the milliuri type experiments are all about. Taking something that is definitely
00:15:49.120 not alive, not living inorganic, leaving it for some time in a warm environment with energy,
00:15:57.280 in other words, perhaps with electricity. And then at the end, having something alive, something
00:16:01.680 organic. Surely the production of amino acids indicates that we're on the way to producing life.
00:16:10.640 Well, no. As many other people have pointed out, I might just mention
00:16:16.720 Paul Davies. Paul Davies wrote a next one called the Goldilocks Enigma,
00:16:20.480 all about how the conditions in the universe seem to be just right for life. But he will admit,
00:16:25.760 as many other people will, that these milliuri type experiments that produce amino acids
00:16:30.720 aren't showing very much. We find amino acids with our powerful telescopes looking at
00:16:36.880 interstellar gas clouds. There's amino acids out there in interstellar space.
00:16:41.600 So amino acids seem to arise pretty spontaneously, given the right elements with a little bit of
00:16:47.280 energy. Other people have described finding amino acids as to finding a pile of bricks.
00:16:53.120 If you walk around the corner and find a pile of bricks, you don't expect that the next pile
00:16:58.720 of bricks that you're going to find is going to self-assemble itself into something like the
00:17:02.960 opera house, which will be the equivalent of thinking that just because you found amino acids,
00:17:07.200 you're on the way to finding a living organism. So just because milliuri were able to create
00:17:14.720 amino acids inside their flask, because there's absolutely nothing about whether or not
00:17:19.360 left for a certain amount of time you will get life. So the remarkable thing is that
00:17:25.600 milliuri experiments have been repeated again and again since the 1950s, with basically the same
00:17:32.000 results, I think proteins might have been created at some point. We do not know how inorganic
00:17:39.360 material becomes organic material, by organic material, I mean material itself replicating
00:17:45.280 that to life. Many popularizers of science today talk about how we do not know how geochemistry
00:17:51.840 becomes biochemistry. There's a massive gap in our understanding and this comes to bear on this
00:17:58.400 whole question of why we do not understand how the knowledge that's in DNA got there at all.
00:18:06.160 How it gets there. How the universality of DNA, which means that the DNA can actually create
00:18:11.600 any living organism that's out there. How that came to be in the first place. We don't know.
00:18:17.920 We can't create DNA from inorganic materials. It must have happened. It must have been a spontaneous
00:18:25.200 thing. We just don't know how. I'll just mention it passing one final thing. The milliuri
00:18:30.400 experiment seems to suggest that life is very, very hard to create because our smartest
00:18:37.600 biologists working for decades at a time on these kind of experiments have not produced
00:18:43.840 artificial life in the lab. They don't know how. It must be really hard. We've repeated
00:18:51.440 what we think of the conditions in the early earth over and over again. I think it must be
00:18:55.760 thousands of times, thousands of different experiments by now and nothing is crawled out of that
00:18:59.840 flask. So that's an argument that life is very hard to spontaneously arise. But the universe is
00:19:07.520 such that it's very difficult for life to spontaneously arise. If smart people working in labs
00:19:13.600 can't do it, then is it going to happen by chance? If people with knowledge can't seem to do it,
00:19:19.920 then how on earth could an environment without any knowledge cause it to happen spontaneously?
00:19:25.600 We don't know. There's a great mystery there. So on the one hand, we have this argument that
00:19:30.960 life must be very difficult and therefore be very rare in the universe as well.
00:19:36.080 One of the reasons, one of the answers to the fermi paradox might be it's just too difficult.
00:19:41.200 It's highly unlikely that however many planets there are out there in the universe,
00:19:46.800 there's simply not enough planets out there in the universe. Even if you would have covered them
00:19:50.960 all in inorganic material and to warm them just like in the military experiment and to make them
00:19:56.960 perfectly bio-friendly, nonetheless, it might be the case that it is exceedingly unlikely
00:20:03.760 for inorganic material to become alive. And so that might be an answer to the fermi paradox.
00:20:09.520 And the reason for books like this one, which I would also recommend, rare earth,
00:20:14.720 why complex life is uncommon in the universe by ward and brownly, that's a great book. Now,
00:20:21.760 standing in stark contrast to that, I might just mention a paper by the great Charlie Line Weaver
00:20:26.720 who I mentioned earlier, and he wrote a paper looking at how quickly life arose here on the planet,
00:20:35.200 here on earth. There was a period called the late heavy bombardment here on earth,
00:20:39.920 where all of these comets from asteroids, material, from the very far reaches of space,
00:20:46.560 I think, out in the ought cloud, or maybe it's the kyper belt. One of those places where there's
00:20:52.480 lots and lots of asteroids and the far reaches of the solar system. This material was kicked in
00:20:58.320 towards the earth, and it bombarded the earth. And so this is called the late heavy bombardment.
00:21:03.120 And it caused the temperature that surfaces the earth to rise very, very high. I think to
00:21:08.400 molten rock temperatures, so thousands of degrees Celsius sterilizing the earth. So if there
00:21:13.600 wasn't any life there, it was certainly wiped out at the late heavy bombardment. Probably there
00:21:17.680 was no life there to begin with, because prior to that, the conditions weren't any more friendly.
00:21:22.560 They became exceedingly hostile during the late heavy bombardment, as the point, however.
00:21:27.280 But then what happened? Well, then the earth cooled. And if we look through the fossil record,
00:21:32.320 if we dig down deep, we find dinosaurs. And if we dig even deeper so that the strata become
00:21:38.800 older and the organisms are more ancient, then we find fish. And if we keep going down further,
00:21:44.240 then we find nothing but bacteria. And if we keep going down further, we find nothing.
00:21:48.160 We find no living organisms. We find the period at which the late heavy bombardment happened,
00:21:52.720 and there's geological evidence for the late heavy bombardment. But as soon as the late heavy
00:21:57.360 bombardment was over, almost as soon as it was over in geological time, life appeared.
00:22:03.200 Life appeared straight away. It seems on a geological time scale. So that is an argument that,
00:22:10.240 life will arise as quickly as it can, given favorable conditions. Because here on earth,
00:22:16.160 as soon as the conditions were favorable, life arose. That's remarkable. But now we have a
00:22:21.600 contradiction. The military type experiments seem to be suggesting that life is very, very difficult
00:22:27.360 to create. No matter what the conditions, because the scientists keep trying to create friendly
00:22:31.600 conditions and nothing's crawling out of the flask, on the other hand, the actual experience of
00:22:36.160 earth was life arose as soon as it possibly could, as soon as the conditions were just right.
00:22:42.720 What is going on? This is really exciting. There must be an answer to this. But we don't
00:22:47.280 understand how life arose. We do not understand these processes of life. I suppose this is kind of
00:22:54.560 only tangentially related precisely to what the beginning of infinity is about, which is more about
00:22:59.520 once you've got life, once you've actually got life, how can you artificially simulate the process
00:23:04.640 of evolution by natural selection? Although the process of evolution by natural selection
00:23:08.400 probably predates any of the simple living organisms that we have. It probably goes all the way
00:23:12.640 back to when we had an RNA world, so it was probably evolution by natural selection going on then.
00:23:18.160 We don't even know how to get RNA out of this inorganic material. That's enough for me from the
00:23:23.200 moment. Let's go to the book. So David writes, When discussing Lamarckism in Chapter 4, I pointed out
00:23:31.440 the fundamental difference between a muscle becoming stronger in an individual's lifetime,
00:23:35.440 and muscles evolving to become stronger. For the former, the knowledge to achieve all the
00:23:40.160 available muscle strengths must already be present in the individual's genes before the sequence
00:23:44.880 of changes begins. And so must the knowledge of how to recognize the circumstances under which to
00:23:50.240 make the changes. This is exactly the analog of a trick that a programmer has built into a chatbot.
00:23:56.080 The chatbot responds as though it had created some of the knowledge while composing it
00:24:00.080 response. But in fact, all the knowledge was created earlier and elsewhere. We're just going to flag
00:24:06.000 that as me talking. We will come back to that later. That is the key fact in this chapter that I
00:24:14.480 think I missed. I'll say it again. When it comes to a chatbot, when it comes to knowledge and the
00:24:23.360 chatbot appearing to say something original or something creative or to give you a piece of knowledge
00:24:28.880 that it has produced itself. David says, in fact, all that knowledge was created earlier
00:24:37.440 and elsewhere. Okay, I'll continue reading. The analog of evolutionary change in a species
00:24:44.800 is creative thought in a person. The analog of the idea that AI could be achieved by an accumulation
00:24:50.640 of chatbot tricks is Lamarckism. The theory that new adaptations could be explained by changes
00:24:56.880 that are in reality just a manifestation of existing knowledge. There are several current areas
00:25:02.400 of research in which that same misconception is common. In chatbot-based AI research,
00:25:08.080 it sent the whole field down a blind alley. But in other fields, it has merely caused researchers
00:25:12.880 to attach over ambitious levels to genuine, albeit relatively modest achievements.
00:25:18.400 One such area is artificial evolution. Recall Edison's idea that progress requires alternating
00:25:26.000 inspiration and perspiration phases. And that, because of computers and other technology,
00:25:31.440 it is increasingly becoming possible to automate the perspiration phase.
00:25:36.720 This welcome development has misled those who are overconfident about achieving artificial
00:25:40.960 evolution and AI. For example, suppose you are a graduate student in robotics, hoping to build
00:25:46.880 a robot that walks on legs better than previous robots do. The first phase of the solution must
00:25:53.120 involve inspiration. That is to say, creative thought, attempting to improve upon previous
00:25:58.960 researchers' attempts to solve the same problem. You will start from that, and from existing
00:26:04.960 ideas about other problems that new conjecture may be related. And from the designs of walking
00:26:10.320 animals in nature, all of that constitutes existing knowledge, which you will vary and combine in
00:26:16.320 new ways, and then subject to criticism and further variation. Eventually, you will have created
00:26:21.760 a design for the hardware of your new robot. It's legs with their levers, joints, tendons and motors.
00:26:27.200 It's body which will hold the power supply. It's sense organs through which it will receive
00:26:32.000 the feedback that will allow it to control those things effectively, and the computer that will
00:26:35.920 exercise that control. You will have adapted everything in that design as best you can to the
00:26:42.080 purpose of walking, except the program in the computer. So I just paused there. So we're already
00:26:50.000 getting a hint of where this argument is going. The knowledge was created earlier and elsewhere.
00:27:00.240 With evolution by natural selection, organisms don't come into being just with legs. Those legs
00:27:07.440 have evolved over time. From things that weren't legs, maybe like fins or flippers or something
00:27:13.040 like that. And then legs evolved. In fact, in the case of flippers, and of course they evolved
00:27:17.840 from legs, you get my point. Here, the graduate student has already built a theme, a robot,
00:27:27.680 with the limbs. So already, we don't have anything resembling artificial evolution.
00:27:35.040 We have a creature that already has the required hardware. So let's go back to the book,
00:27:42.080 we'll just reread that last section. Last sentence. You will have adapted everything in that
00:27:49.920 design as best you can to the purpose of walking, except the program in the computer. The function
00:27:55.840 of that program will be to recognize situations such as the robot beginning to topple over,
00:28:00.560 or obstacles in its path, and to calculate the appropriate action and to take it. This is the
00:28:05.200 hardest part of your research program. How does one recognize when it is best to avoid an obstacle to
00:28:10.000 the left or to the right or to jump over it or to kick it aside or ignore it, or lengthen one
00:28:14.560 stride to avoid stepping on it, or judge it impossible and turn back. In all those cases,
00:28:20.480 how does one specifically do those things in terms of sending countless signals to the motors
00:28:24.880 and the gears as modified by feedback from the sensors? You will break the problem down into sub
00:28:30.160 problems, veering by a given angle is similar to veering by a different angle. That allows you to
00:28:35.040 write a subroutine for veering that takes care of the whole continuum of possible cases,
00:28:38.800 once you have written it, all other parts of the program need only call it whenever they decide
00:28:43.840 that veering is required, and so they did not have to contain any knowledge about the messy details
00:28:48.480 of what it takes to vee. When you have identified and solved as many of those subproblems as you can,
00:28:54.000 you will have created a code or language that is highly adapted to making statements about
00:28:58.960 how your robot should walk. Each call of one of its subroutines is a statement or command in that
00:29:04.800 language. So far, most of what you have done comes under the heading of inspiration. It required
00:29:11.200 creative thought, but now perspiration looms. Once you have automated everything that you know
00:29:16.560 how to automate, you have no choice but to resort to some sort of trial and error to achieve
00:29:21.520 any additional functionality. However, you do now have the advantage of a language that you have
00:29:26.480 adapted for the purpose of instructing the robot how to walk. So you can now start with a program
00:29:31.600 that is simple in that language, despite being very complex in terms of elementary instructions of
00:29:37.280 the computer, and which means, for instance, walk forwards and stop if you hit an obstacle.
00:29:43.280 Then you can run the robot with that program and see what happens, or you can run a computer
00:29:47.040 simulation of the robot. When it falls over, anything else undesirable happens, you can modify
00:29:51.520 your program, still using the high level language you have created to eliminate the deficiencies
00:29:56.480 they arise. That method will require ever less inspiration, and ever more perspiration.
00:30:03.360 But an alternative approach is also open to you. You can delegate the perspiration to a computer
00:30:09.600 by using a so-called evolutionary algorithm. Using the same computer simulation, you run many
00:30:15.440 trials. Each with a slight random variation of that first program, the evolutionary algorithm
00:30:21.200 subjects each simulated robot automatically to a battery of tests that you have provided.
00:30:26.560 How far it can walk without falling over. How will it cope with obstacles and rough terrain
00:30:30.560 and so on? At the end of each run, the program that performed best is retained and the rest of
00:30:34.640 discarded. Then many variants of that program are created and the process is repeated.
00:30:39.680 After thousands of iterations of this evolutionary process, you may find that your robot walks
00:30:44.000 quite well according to the criteria you have set. You can now write your thesis.
00:30:49.120 Not only can you claim to have achieved a robot that walks with a required degree of skill,
00:30:53.760 you can claim to have implemented evolution on a computer.
00:31:00.240 This sort of thing has been done successfully many times. It is a useful technique.
00:31:03.680 It certainly constitutes evolution in the sense of alternating variation selection.
00:31:07.760 But is it evolution in the more important sense of the creation of knowledge by variation
00:31:12.800 selection? I'll just repeat that. This is David Deutsch's emphasis on what Darwin and Dawkins have
00:31:21.760 said about evolution by natural selection. David emphasizes how DNA contains knowledge,
00:31:31.440 knowledge of how to create organisms. Therefore, evolution by natural selection is really
00:31:36.240 a theory about how knowledge has evolved in DNA. I have a knowledge got into that DNA.
00:31:44.080 So I'll just repeat that. But is it evolution in the more important sense of the creation
00:31:49.600 of knowledge by variation in selection? This will be achieved one day, but I doubt that it has been
00:31:54.720 yet. For the same reason that I doubt chatbots are intelligent, even slightly. The reason is that
00:31:59.200 there is much more obvious explanation of their abilities, namely the creativity of the programmer.
00:32:04.400 The task of ruling out the possibility that the knowledge was created by the programmer in the
00:32:08.800 case of artificial evolution has the same logic that checking a program is an AI, but hard to
00:32:13.920 be because the amount of knowledge that the evolution purportedly creates is vastly less.
00:32:19.920 Even if you yourself are the programmer, you are in no position to judge whether you created
00:32:23.680 that relatively small amount of knowledge or not. For one thing, some of the knowledge that you
00:32:27.760 packed into the language during those many months of design will have reach because it encoded
00:32:32.080 some general truths about the laws of geometry, mechanics, and so on. For another, when designing
00:32:38.320 the language you had constantly in mind, what sort of abilities it would eventually be used to
00:32:44.400 express? Okay, me talking. So the programmer who says they've successfully simulated evolution
00:32:53.280 by natural selection in a computer, well, they can't have because evolution is blind,
00:32:59.520 evolution is blind, we know this, we know this from the theory of evolution by natural selection.
00:33:04.880 It's not directed towards a particular goal. This is a huge misconception, by the way.
00:33:09.920 Okay, yes, yes. I admit, lots of people do have this misconception.
00:33:14.400 Lots of people think there is an arrow of evolution that organisms move towards ever greater
00:33:20.560 complexity. Not true. Bacteria have been around since the beginning of life on earth for
00:33:26.960 billions of years. They have remained there. They're just as evolved as we are,
00:33:31.520 like Ricky Gervais's fond of saying things like that, which is true in a sense.
00:33:36.480 They've been evolving and they've found their niche and they have been perfectly suited to their
00:33:40.320 niche, as have moths and other insects, as have fish. They have all evolved to fill their niche.
00:33:47.760 There is no arrow of evolution leading towards a particular form of complexity. Evolution
00:33:53.520 wasn't always directed towards us. So far as we can tell, at least that's what the theory says,
00:33:57.760 because evolution is blind, it doesn't know what the next best organism would be for a given environment.
00:34:04.640 So if evolution is blind, fact, it cannot be the case that this is an evolutionary algorithm
00:34:12.560 in the Darwinian sense, because the programmer, when designing the language,
00:34:18.080 had constantly in mind what sort of abilities it would eventually be used to express,
00:34:22.560 which is completely unlike what evolution does. Evolution does not have in mind what sort of
00:34:28.560 abilities it's going to be expressing. It's going to be expressing in the organisms of the future.
00:34:34.960 I'll continue reading. Next paragraph, I'm just going to skip the next paragraph and then David
00:34:41.200 writes, one thing that always seems to happen with such projects evolutionary algorithms
00:34:46.160 is that after they achieve their intended aim, if the evolutionary program is allowed to run further,
00:34:51.280 it produces no further improvements. This is exactly what would happen if all the knowledge in
00:34:57.280 this successful robot had actually come from the programmer, but it is not a conclusive critique.
00:35:02.000 Biological evolution often reaches local max mirror fitness. So again, I'll just pause there.
00:35:06.320 So what's David saying there is that in real life evolution, one wouldn't expect the
00:35:14.000 organism that's evolving to simply stop there. If it was true evolution, especially if you're
00:35:18.080 doing it to computer and you can simulate things such that they get better and better, very,
00:35:22.320 very quickly. So you can do billions of cycles of evolution in a few seconds or a few minutes
00:35:29.440 or whatever it happens to be, could you just simulate it? Why doesn't the organism,
00:35:33.520 organism, I say organism? Why doesn't the simulated robot, let's say, inside of the computer
00:35:38.320 that you've taught to, you haven't taught, you've allowed to evolve the capacity to walk,
00:35:43.920 why doesn't it then start jumping? Or better yet, why don't the legs evolve into wings and it
00:35:49.520 fly away? That's kind of what evolution does. You have improvements beyond the thing that it
00:35:56.560 becomes good at, but that doesn't happen. All that it evolves towards is precisely the thing
00:36:02.560 you expected to evolve before to towards. You expect it to evolve towards walking because that's
00:36:08.560 what you've programmed it to do. The only thing that you've added in is this so-called
00:36:12.480 evolutionary algorithm where instead of just solving the problem yourself, you allow the computer
00:36:17.440 to determine, to judge for itself, whether or not it's slightly improving or slightly getting
00:36:22.880 worse and to make a judgment on that and you're calling that evolution. Now, as David says there,
00:36:29.360 that is not a conclusive critique, biological evolution often reaches local maxima or fitness.
00:36:35.040 Also, after obtaining its mysterious form of universality, DNA, he's talking about here,
00:36:40.720 seemed to pause for about a billion years before creating any significant new knowledge.
00:36:46.320 But still, achieving results that might well be due to something else is not evidence of evolution.
00:36:52.000 That is why I doubt that any artificial evolution has ever created knowledge.
00:36:56.080 I have the same view for the same reasons about slightly different kind,
00:36:59.680 about the slightly different kind of artificial evolution that tries to evolve simulated
00:37:04.000 organisms in a virtual environment and the kind that pits different virtual spaces against each
00:37:08.560 other, which is the one that I'm going to put down in a link below. And you play around with
00:37:13.360 those. It's a good way to learn about evolution by natural selection, but yes, there's lots of
00:37:17.840 those things out there on the internet and there's lots of ways of simulating evolution, but as David
00:37:23.120 says there, they're not real artificial evolution. They're toys. They might be good pedagogical
00:37:30.400 tools to use, but these are not proper simulations of evolution by natural selection.
00:37:36.160 The next part is the real kicker. So let's read the next couple of paragraphs.
00:37:43.760 So he wants to, he's just asserted that no genuine artificial evolution has ever been simulated
00:37:54.320 and he writes, to test this proposition, I would like to see an experiment of a slightly different
00:37:58.880 kind, eliminate the graduate student from the project. Then instead of using a robot design to
00:38:06.480 evolve better ways of walking, use a robot that is already in use in some real life application
00:38:10.480 happens to be capable of walking. And then instead of creating a special language of subroutines
00:38:15.200 in which to express conjectures about how to walk, just replace its existing program in its existing
00:38:20.400 microprocessor, in its existing microprocessor by random numbers, permutations use errors of the type
00:38:26.720 that happen anyway in such processes, though in the simulation you allowed to make them happen
00:38:30.800 as often as you like. The purpose of all that is to eliminate the possibility that human knowledge
00:38:35.440 is being fed into the design of the system and that its reach is being mistaken for the product
00:38:39.680 of evolution. Then run simulations of that mutating system in the usual way, as many as you like.
00:38:45.120 If the robot ever walks better than it did originally, then I am mistaken.
00:38:48.640 If it continues to improve after that, then I'm very much mistaken.
00:38:52.000 One of the main features of the above experiment, which is lacking in the usual way of doing
00:38:56.960 artificial evolution, is that, for it to work, the language of subroutines, would have to evolve
00:39:02.640 along with the adaptations that it was expressing. This is what was happening in the biosphere
00:39:06.880 before that jumped to universality, that finally settled in the DNA genetic code. As I said,
00:39:11.920 it may be that all those previous genetic codes were only capable of coding for a small number
00:39:16.000 of organisms that were all rather similar, and that the overwhelming reach and that the overwhelmingly
00:39:21.200 rich biosphere that we see around us, created by randomly varying genes while living the
00:39:25.440 language unchanged, is something that became possible only after that jump. We don't even know
00:39:30.560 what kind of universality was created there. Why should we expect our artificial evolution to work
00:39:35.920 without it? David concludes by speaking about how we need to be a little bit more humble and
00:39:44.000 modest and face the fact there are huge unknowns here, huge unknowns with human intelligence,
00:39:51.680 human creativity, and trying to create artificial creativity inside artificial general intelligence.
00:39:59.280 And there are huge unknowns similarly with evolution by natural selection. We've never simulated
00:40:04.720 evolution. If we can't program it, we haven't understood it. And so I'll just emphasize that
00:40:10.400 again, just to put a cap on this about what I didn't understand. It's really the part where it
00:40:17.040 talks about taking away the graduate student, removing the goal, removing the goal, okay, I didn't get
00:40:23.120 that. Evolution doesn't have a goal. It's just survival of the fittest. But it can't possibly,
00:40:31.920 the evolution doesn't know what the fittest is. It has to test out what works. And what survives
00:40:37.680 is what survives. It becomes a little bit of a truism, the survival of the fittest. Darwin
00:40:44.560 never actually talked about survival of the fittest. And I think for a very good reason, he knew
00:40:47.920 that it was sort of a tautology because who were the fittest or the fittest of the ones who survived,
00:40:51.760 right? So anyway, taking away the graduate student from the experiment, where you've got this
00:40:56.560 robot that can't yet walk, removes the goal, removes the knowledge that could be put into that
00:41:02.400 robot by the graduate student, by the programmer. If you give the robot legs and let the software take
00:41:08.880 over, then what happens? Well, if it's just a random lot of numbers in there, then you're letting
00:41:14.560 the microprocessor just churn through these numbers, then that would be a true simulation of
00:41:20.320 evolution by natural selection. This random sequence of numbers. If the robot starts to walk,
00:41:27.120 if the robot starts to walk a little bit better, well, then okay, knowledge is somehow being
00:41:32.960 created by that robot in order to improve its walking. So if the robot was to start to walk better,
00:41:38.400 David admits he would be wrong. And if he continues to improve, he'd be very wrong. But wrong
00:41:42.960 about what? Well, then evolution, he would be wrong that evolution doesn't have a goal. Because
00:41:49.440 the robot, if it starts to walk, must have had a goal, must have had the goal of walking.
00:41:53.120 Now, of course, it's not just David who would be wrong. He's not just David who would be wrong.
00:41:58.480 He just happened to explain this thought experiment. Darwin and Dawkins would be wrong,
00:42:03.760 and it's what they were in fact getting it, and I didn't understand. There is literally no goal.
00:42:07.920 There's no thing walking otherwise to strive towards. So why? Nevertheless, is there an
00:42:13.600 improvement of the kind we actually see in nature? We do see things walking that couldn't walk
00:42:19.280 before kind of, you know, the fish evolved into land creatures and the land creatures walked,
00:42:24.960 but the individual fish didn't. That would be, again, Lamarchism. We do see increased complexity,
00:42:31.440 life evolving to fill niches. And there's sometimes leads to greater complexity, but why?
00:42:36.960 How does life fit so well in its niche? How does it adapt? We don't know. Except to say,
00:42:43.520 well, diffused survive. And the fittest are just those who survive and both who survive are the
00:42:50.640 fittest. In high school biology, students now learn all sorts of fancy details about genetics,
00:42:56.560 about replication and transcription, how genes on DNA code for proteins. The DNA is a kind of
00:43:02.240 software, and it constructs proteins, different parts of the DNA code for different proteins.
00:43:07.040 Indeed, it's all very computer-like. There's a readhead that moves along the DNA,
00:43:11.200 copying sections to RNA, which then gathers up the required materials for protein synthesis. We
00:43:15.920 know many details, but we cannot replicate the evolution of this in the lab. We simply do not know
00:43:22.560 enough details to simulate the evolution of this process. This is where it's not like general
00:43:28.800 relativity. If there's a true evolutionary algorithm out there, let's see an actual ecosystem
00:43:36.240 of hope online. So there's so much we don't know. And as I began this with the military experiment,
00:43:41.600 we don't know why all these military experiments continue to fail. There are big gaps here now
00:43:46.880 understanding about how complexity increases. In other words, how the knowledge comes to be
00:43:53.520 instantiated within the DNA, within the self-replicating molecules. This is really exciting for me,
00:44:00.160 this chapter. This was a brilliant one. It's the one, as I fully admit, I didn't appreciate until
00:44:05.920 last year. But now I think I do. I have a better idea, and it was all about the fact that
00:44:12.080 evolution doesn't have a goal, but evolutionary algorithms most certainly do. And so therefore
00:44:18.400 they're misnamed. They're not true evolution by natural selection algorithms. Call them
00:44:23.520 evolutionary algorithms if you like, but don't think that in any way it resembles the kind of
00:44:28.160 evolution we have in biology. Looking forward to doing the very next chapter. The next chapter
00:44:35.040 is our window and infinity. So I'm living away from biology and into the philosophy of mathematics.
00:44:40.960 I can't wait to do that one, and I'll see you then. Bye-bye.
