WEBVTT - The Machine Lords of Barnard 68, Part 2 0:00:03.040 --> 0:00:05.360 Welcome to Stuff to Blow Your Mind production of My 0:00:05.480 --> 0:00:14.480 Heart Radio. Hey, welcome to Stuff to Blow Your Mind. 0:00:14.520 --> 0:00:16.919 My name is Robert Lamb and I'm Joe McCormick, and 0:00:16.960 --> 0:00:19.040 we're back with part two of our talk about post 0:00:19.079 --> 0:00:23.400 biological intelligence. Now, in the last episode, we talked about 0:00:23.840 --> 0:00:26.480 let's see, we talked about some work by the SETI 0:00:26.560 --> 0:00:29.880 researcher Seth show Stack, and we talked about the philosopher 0:00:29.920 --> 0:00:33.519 Susan Schneider, who had both written about UH the idea 0:00:33.600 --> 0:00:37.080 of looking for signs of alien intelligence elsewhere in the 0:00:37.080 --> 0:00:40.599 Milky Way, and UH the the proposition that if we 0:00:40.600 --> 0:00:44.040 were to encounter such an intelligence, it would probably be 0:00:44.120 --> 0:00:50.360 more likely the machine descendants of a previous biological intelligence 0:00:50.600 --> 0:00:55.080 than it would be biological entities themselves. That that overtime, 0:00:55.640 --> 0:00:58.959 organisms like us will tend to sort of turn themselves 0:00:59.000 --> 0:01:01.760 into machines, or at least create a techno culture that's 0:01:01.800 --> 0:01:05.559 dominated by machines, and that these are the types of 0:01:05.760 --> 0:01:08.919 intelligences that we should really be looking for and trying 0:01:08.920 --> 0:01:12.720 to predict in terms of their characteristics and things like that. 0:01:13.120 --> 0:01:15.080 So we can jump right back into the middle of 0:01:15.120 --> 0:01:18.480 this conversation where we left off last time with talking 0:01:18.520 --> 0:01:23.200 about post biological intelligence. Now another big question here and 0:01:23.400 --> 0:01:26.919 and this will we'll go back to Schneider. Is it 0:01:27.040 --> 0:01:30.480 is that the question of would a machine culture like 0:01:30.520 --> 0:01:34.000 this if you encountered it, would this machine artificial intelligence? 0:01:34.000 --> 0:01:37.280 Would it be conscious? And what would that mean? And 0:01:37.360 --> 0:01:39.440 would it make a difference even, Yeah, this is a 0:01:39.440 --> 0:01:41.960 really good question. The way she puts it is would 0:01:42.040 --> 0:01:46.279 the processing of a silicon based super intelligent system feel 0:01:46.319 --> 0:01:49.600 a certain way from the inside. Now, I'm going to 0:01:49.680 --> 0:01:52.560 go into less detail on this argument than than I 0:01:52.600 --> 0:01:54.960 did on the other half of her argument, but I 0:01:55.000 --> 0:01:57.040 did want to try to give a few highlights. This 0:01:57.160 --> 0:02:02.040 question is inherently difficult to answer because, according to some philosophers, 0:02:02.080 --> 0:02:04.160 you know, some people would say that this question is 0:02:04.200 --> 0:02:07.040 impossible to answer because there is no way to test 0:02:07.080 --> 0:02:11.040 for consciousness beyond our first our own first person experience. 0:02:11.080 --> 0:02:13.000 I mean, we can't even test to see if other 0:02:13.040 --> 0:02:16.320 people are conscious. We just assume they are. It seems 0:02:16.400 --> 0:02:18.240 like they are, they claim to be, and there's no 0:02:18.280 --> 0:02:21.120 reason to assume they're not. But of course you have 0:02:21.240 --> 0:02:24.280 ideas like the philosopher David Chalmers, you know, he famously 0:02:24.360 --> 0:02:27.080 framed this idea of the easy problems versus the hard 0:02:27.120 --> 0:02:31.000 problems of consciousness, and so the easy problems that they're 0:02:31.040 --> 0:02:34.799 not actually easy, but they are, they're in principle solvable. 0:02:34.800 --> 0:02:37.760 There are things like what parts of the brain are 0:02:37.840 --> 0:02:41.119 necessary in order to generate conscious experience, Like you could, 0:02:41.480 --> 0:02:43.440 you know, you could do research on that and have 0:02:43.480 --> 0:02:45.840 people report back when different parts of the brain are 0:02:45.840 --> 0:02:47.919 disabled or something. You know, you can figure out things 0:02:48.000 --> 0:02:51.200 like that, but it's much more difficult to or Chalmers 0:02:51.200 --> 0:02:53.959 would argue, ultimately impossible to get to the bottom of 0:02:54.000 --> 0:02:57.800 the question why does all this information processing in the 0:02:57.880 --> 0:03:01.960 human brain under certain conditions have a felt quality to it? 0:03:02.080 --> 0:03:05.560 Like why is their consciousness in the first place? And 0:03:05.639 --> 0:03:08.360 if we do not know, or possibly even cannot know, 0:03:08.600 --> 0:03:11.680 why we possess a felt subjective experience, how could we 0:03:11.760 --> 0:03:15.200 ever reason backwards to know if alien machines would also 0:03:15.280 --> 0:03:19.000 possess it? Now, Schneider responds to all this thinking, and 0:03:19.240 --> 0:03:22.880 I'm oversimplifying here, but her main point is that the 0:03:22.960 --> 0:03:25.960 activity of the brain is, according to her argument, primarily 0:03:26.040 --> 0:03:30.639 computational and in the absence of compelling evidence for what 0:03:30.720 --> 0:03:34.960 she calls biological naturalism, and that's the idea that consciousness 0:03:35.120 --> 0:03:38.880 is or is likely to be unique to biological carbon 0:03:38.920 --> 0:03:43.120 based organisms. Daniel Dennett ridicules this point of view by 0:03:43.280 --> 0:03:45.480 calling it the belief that the brain possesses what he 0:03:45.520 --> 0:03:48.720 calls wonder tissue. You know that there's just something in 0:03:48.760 --> 0:03:52.440 the brain that, like magic, allows it to generate consciousness 0:03:52.440 --> 0:03:56.560 while other types of things can't generate consciousness. I don't 0:03:56.560 --> 0:03:58.640 know the answer to this question whether things other than 0:03:58.720 --> 0:04:02.360 brains can or cannot gem great consciousness. It Uh, I'm 0:04:02.400 --> 0:04:05.360 sort of skeptical of both sides of the argument. But 0:04:05.360 --> 0:04:09.880 but anyway, Schneider argues that we should conclude by analogy 0:04:09.880 --> 0:04:13.880 that other computational agents, because our brains are computational agents 0:04:13.880 --> 0:04:17.680 and they generate consciousness, that other computational agents are also 0:04:17.760 --> 0:04:22.159 capable of possessing consciousness unless there's some kind of evidence 0:04:22.200 --> 0:04:25.800 that biological naturalism is is necessarily true, and she says 0:04:25.839 --> 0:04:28.560 there's not, And I agree that there is not evidence 0:04:28.600 --> 0:04:30.800 of that. You know, what's interesting about the way that 0:04:30.880 --> 0:04:32.919 you just lay this out, though I can easily imagine 0:04:32.920 --> 0:04:37.480 a situation where an advanced AI is forced to, uh, 0:04:37.520 --> 0:04:40.720 sort of ponder the situation, well, is having a conscious 0:04:40.880 --> 0:04:44.839 um this conscious experience? Is it important. Well, let's let 0:04:44.839 --> 0:04:48.200 me create like a programming or a subset of myself 0:04:48.640 --> 0:04:51.960 that has at least as close of an approximation of 0:04:52.000 --> 0:04:54.919 consciousness as as as as it is understood at that 0:04:55.000 --> 0:04:59.560 time in order to evaluate you know, right, Um, so 0:04:59.640 --> 0:05:02.239 then it perhaps has sort of it's its main mind, 0:05:02.320 --> 0:05:05.000 but then it has sort of a subset of quote 0:05:05.040 --> 0:05:09.880 unquote conscious minds just in case it is important. Yeah, obviously, 0:05:09.920 --> 0:05:11.840 I mean huge question, like how would it know how 0:05:11.880 --> 0:05:14.760 to do that? But assuming could that, Yeah, that's really 0:05:14.760 --> 0:05:18.400 funny that, like it could try to iterate consciousness in 0:05:18.400 --> 0:05:20.920 in an experimental way to see if to see if 0:05:20.960 --> 0:05:23.440 it makes a difference, because that's another big thing, like, 0:05:23.480 --> 0:05:26.040 you know, the biological question about consciousness. We at least 0:05:26.080 --> 0:05:28.520 know that biological brains can be conscious. We don't know 0:05:28.560 --> 0:05:31.640 if computers can be or not. But since biological brains 0:05:31.640 --> 0:05:36.360 are conscious, is that an adaptive evolutionary trait? Does consciousness 0:05:36.600 --> 0:05:40.799 do something? Or or could you have a an animal 0:05:40.960 --> 0:05:45.479 that is absolutely functionally identical to a human but not conscious. 0:05:45.480 --> 0:05:48.039 This is actually the concept of a philosophical zombie or 0:05:48.040 --> 0:05:52.080 a p zombie, a being that is that is indistinguishable 0:05:52.160 --> 0:05:55.200 from a normal human except it has no inner experience. 0:05:56.480 --> 0:05:59.480 So in like this scenario where the super AI creates 0:05:59.480 --> 0:06:03.440 like account soul of quote unquote conscious iterations of itself, 0:06:03.640 --> 0:06:06.520 like maybe they're just faking consciousness. How would it know? Yeah, 0:06:06.680 --> 0:06:08.320 and then how would it know? How would we know? 0:06:09.240 --> 0:06:11.400 And yeah, again if you're and then if you're dealing 0:06:11.440 --> 0:06:14.000 with an AI, like suddenly we make contact with an 0:06:14.000 --> 0:06:18.120 AI from another world? Um, is it important that it 0:06:18.240 --> 0:06:20.359 be conscious or not conscious? Like? There are lots of 0:06:20.400 --> 0:06:23.760 things that are important and even beneficial that are not conscious, 0:06:23.800 --> 0:06:26.400 like the you know, the Bill of Rights is not 0:06:26.480 --> 0:06:30.640 a conscious entity. Um, but you know, I think most 0:06:30.680 --> 0:06:32.880 would argue that it is. It is important. It does 0:06:32.920 --> 0:06:35.800 good things. I mean, you could argue that it is 0:06:35.880 --> 0:06:40.000 only important in that it has effects on conscious on 0:06:40.040 --> 0:06:43.280 things that are conscious. Yeah, Like in a universe where 0:06:43.320 --> 0:06:45.960 there was nothing that was conscious, would the Bill of 0:06:46.040 --> 0:06:48.800 Rights be useful? But I don't know. I mean, I 0:06:48.839 --> 0:06:50.560 guess there are some theories of value that would that 0:06:50.600 --> 0:06:53.240 would say like, yeah, things could still be of value 0:06:53.279 --> 0:06:56.960 even if they weren't conscious, right, uh yeah, And then 0:06:56.960 --> 0:06:59.520 again it just it also kind of becomes pointless because 0:06:59.560 --> 0:07:02.359 once you're talking to that AI. Um, yeah, Well what 0:07:02.440 --> 0:07:04.279 does what does it mean if it's conscious or not? Like, 0:07:04.279 --> 0:07:07.359 how does that change the way you interact with it? Um? 0:07:07.440 --> 0:07:10.360 Unless you're you know, actively saying hey, stop, think about 0:07:10.400 --> 0:07:12.960 what you're doing, think about what you're thinking about. Uh. 0:07:13.160 --> 0:07:15.120 I don't know. So I don't know what to think 0:07:15.120 --> 0:07:18.280 about the consciousness question for for alien machines. I mean, 0:07:18.320 --> 0:07:21.600 I think I think Schneider makes the best argument that 0:07:21.640 --> 0:07:23.680 I could imagine for it, But I still don't know 0:07:23.720 --> 0:07:26.040 if I'm convinced, just because this whole realm to me 0:07:26.120 --> 0:07:30.240 just seems so uncertain. Um. But but then she goes 0:07:30.280 --> 0:07:31.720 on to some other things that I think are some 0:07:31.760 --> 0:07:35.960 really interesting ideas. Actually, she talks about what would be 0:07:36.120 --> 0:07:41.400 the predictable characteristics of super intelligent machines, the minds we 0:07:41.440 --> 0:07:44.320 would encounter out there if we did encounter them. Well, 0:07:44.360 --> 0:07:46.120 she admits that there's not a lot we can know, 0:07:46.360 --> 0:07:48.520 at least certainly not that much that we can say 0:07:48.560 --> 0:07:51.200 with with too much confidence, But we can make some 0:07:51.440 --> 0:07:55.360 educated guesses about the broad strokes of of alien intelligence. 0:07:55.720 --> 0:07:58.040 And to do this she cites the work of again 0:07:58.040 --> 0:08:00.800 philosopher Nick Bostrom, who is fame is for writing about 0:08:00.800 --> 0:08:03.520 AI risks, and I believe he actually coined the term 0:08:03.560 --> 0:08:05.960 super intelligence, though it could be wrong about that, but 0:08:07.040 --> 0:08:10.040 Bostrom says, yes, it is hard to predict the goals 0:08:10.120 --> 0:08:13.120 of a future AI, you know, alien intelligence is very 0:08:13.120 --> 0:08:16.640 difficult to understand. But he identifies what he thinks are 0:08:16.720 --> 0:08:20.480 several intellectual tendencies that are likely to be found in 0:08:20.520 --> 0:08:23.520 any super intelligent AI, and they're likely to be found 0:08:23.560 --> 0:08:25.560 in any of them because he says, these traits are 0:08:25.680 --> 0:08:29.840 useful in attaining almost any goal. And so these goals 0:08:29.840 --> 0:08:35.040 he identifies our resource acquisition makes sense. You need resources 0:08:35.080 --> 0:08:39.600 in order to like keep your processes going. Technological perfection, right, 0:08:39.640 --> 0:08:44.040 you want yourself to work efficiently, Cognitive enhancement, you always 0:08:44.080 --> 0:08:47.120 want to be smarter, self preservation, you want to be 0:08:47.160 --> 0:08:50.160 able to keep doing things. And then what he calls 0:08:50.400 --> 0:08:55.400 goal content integrity, and Schneider summarizes that by saying, I e. 0:08:55.920 --> 0:08:59.720 That a super intelligent being's future self will pursue and 0:09:00.080 --> 0:09:03.720 tin those same goals. Uh. And this one was really 0:09:03.720 --> 0:09:06.679 interesting to me actually thinking about the idea that a 0:09:06.800 --> 0:09:10.480 machine would need to try to make sure that as 0:09:10.600 --> 0:09:15.480 it iterates to to improve itself, it doesn't change what 0:09:15.520 --> 0:09:18.600 it was trying to do. In the first place, sort 0:09:18.600 --> 0:09:22.000 of a prime directive sort of situation, right, Yeah, Or 0:09:22.040 --> 0:09:23.880 to come back to the culture, the idea that like, 0:09:24.000 --> 0:09:27.959 if you're created, if your original design is to aid 0:09:28.120 --> 0:09:31.040 humans and make their life easier, than you keep doing 0:09:31.080 --> 0:09:34.679 that even if you are ultimately the calling all the 0:09:34.720 --> 0:09:36.760 shots now and you know, are in charge of all 0:09:36.800 --> 0:09:40.760 the interactions with other civilizations, etcetera. Yeah, And that that 0:09:40.840 --> 0:09:43.800 actually comes into the next thing she says about Bostrom's 0:09:43.840 --> 0:09:47.959 ideas on on these super intelligence is uh. She she writes, quote, 0:09:48.160 --> 0:09:53.040 he underscores that self preservation can involve group or individual preservation, 0:09:53.440 --> 0:09:56.040 and that it may play second fiddle to the preservation 0:09:56.080 --> 0:09:59.800 of the species the AI was designed to serve. So 0:10:00.040 --> 0:10:01.800 could be that these AI s, if they ever do 0:10:01.880 --> 0:10:03.679 come to exist, would yeah, that they would be the 0:10:03.920 --> 0:10:08.480 custodians or caretakers, thinking mainly about the preservation of the 0:10:08.520 --> 0:10:11.240 species that created them, and then when they come to us, 0:10:11.280 --> 0:10:16.839 they ultimately just want to serve man. But then one 0:10:16.920 --> 0:10:19.680 last thing that Schneider argues that I thought would be 0:10:19.720 --> 0:10:21.680 interesting to mention is uh, and I think I said 0:10:21.679 --> 0:10:25.000 this earlier. But she also argues that perhaps the most 0:10:25.080 --> 0:10:28.280 common form of super intelligence we could expect to encounter 0:10:28.520 --> 0:10:32.760 would be what she calls biologically inspired super intelligent aliens, 0:10:33.320 --> 0:10:36.400 and that if this argument is correct, this could also 0:10:36.480 --> 0:10:40.559 tell us some things about intellectual characteristics that we would 0:10:40.559 --> 0:10:44.280 expect to find in these super intelligences. So, to read 0:10:44.320 --> 0:10:47.520 from Schneider's chapter, she says, uh, it may turn out 0:10:47.520 --> 0:10:51.320 that of all super intelligent ai s, biologically inspired super 0:10:51.360 --> 0:10:55.400 intelligent ais, they're the most resemblance to each other. In 0:10:55.400 --> 0:10:59.880 other words, visas maybe the most cohesive subgroup because the 0:11:00.040 --> 0:11:02.880 other members are so different from each other. And there 0:11:02.960 --> 0:11:06.360 she's talking about members of the galactic community. Basically that 0:11:06.400 --> 0:11:09.360 the biologically inspired ones would have the most in common 0:11:09.400 --> 0:11:12.440 with each other. So what kinds of things could they 0:11:12.440 --> 0:11:15.080 have in common? She says, noticed that besas have two 0:11:15.120 --> 0:11:19.760 features that may give rise to common cognitive capacities and goals. One, 0:11:20.040 --> 0:11:24.640 visas are descended from creatures that had motivations like find food, 0:11:24.880 --> 0:11:29.920 avoid injury, and predators, reproduce, cooperate, compete, and so on. 0:11:30.600 --> 0:11:33.880 And then second, she says, the life forms that beasts 0:11:33.880 --> 0:11:37.479 are modeled from have evolved to deal with biological constraints 0:11:37.520 --> 0:11:42.520 like slow processing speed and the spatial limitations of embodiment. So, 0:11:42.559 --> 0:11:46.200 she says, could these two principles one and two yield 0:11:46.320 --> 0:11:50.600 traits common to members of many super intelligent alien civilizations? 0:11:50.640 --> 0:11:53.439 I suspect so, and she gives a bunch of examples, 0:11:53.440 --> 0:11:55.360 But I mean, a very simple and easy to grasp 0:11:55.400 --> 0:11:59.920 one would be that since intelligent biological life is primarily 0:12:00.040 --> 0:12:04.559 be concerned with its biological imperatives, mainly survival and reproduction, 0:12:05.000 --> 0:12:07.440 she says, it is more likely that visas would have 0:12:07.520 --> 0:12:11.200 final goals involving their own survival and reproduction, or at 0:12:11.240 --> 0:12:15.120 least the survival and reproduction of the members of their society. 0:12:15.679 --> 0:12:17.920 And I was just thinking this can be extrapolated to 0:12:17.960 --> 0:12:22.240 other ideas. For example, why wouldn't a superintelligent AI just 0:12:22.240 --> 0:12:25.960 just reprogram itself until it is no longer anything like 0:12:26.120 --> 0:12:30.480 its biological ancestors. So is it still really reproducing the 0:12:30.520 --> 0:12:33.360 original version of itself at all? Well, if you think 0:12:33.360 --> 0:12:37.960 back to Bostrom's idea of of goal content integrity, I 0:12:38.000 --> 0:12:40.320 wonder if this could in a way entail a kind 0:12:40.320 --> 0:12:44.240 of halting of the evolutionary process of life that has 0:12:44.280 --> 0:12:46.680 gone on throughout all of history, Because suddenly, once you 0:12:46.760 --> 0:12:51.280 reach this level of intelligence, a a machine iterating itself 0:12:51.320 --> 0:12:54.320 may just want to preserve the idea that it is 0:12:54.360 --> 0:12:58.679 still its original self. That's an inherently motivating goal for it, 0:12:59.080 --> 0:13:01.800 and thus it would prevent changes to itself that would 0:13:01.840 --> 0:13:05.480 make it feel too different from what it once was. Huh. 0:13:05.960 --> 0:13:07.719 You know it reminds me of like when you when 0:13:07.760 --> 0:13:09.760 you hear a really great song for the first time, 0:13:09.880 --> 0:13:11.760 or you you start playing a video game and it 0:13:11.760 --> 0:13:14.800 really grabs you, or you know, you get super into 0:13:14.920 --> 0:13:17.720 you know, some fandom or another. There's, at least for me, 0:13:17.720 --> 0:13:20.240 there's sometimes that point where you realize, like, wow, this 0:13:20.360 --> 0:13:23.600 is really fulfilling for me right now, and the day 0:13:23.640 --> 0:13:26.080 will come when it won't be Like, as much as 0:13:26.080 --> 0:13:28.360 I enjoy this game or this book or this song 0:13:28.480 --> 0:13:30.160 or whatever it is, there will come a day when 0:13:30.160 --> 0:13:32.840 I will set it aside because and I will there 0:13:32.840 --> 0:13:35.200 will be something else I'm into. So I guess the 0:13:35.280 --> 0:13:39.480 question is it's like, if if we or this machine 0:13:39.840 --> 0:13:42.880 that we're imagining here, if it could decide no, I 0:13:42.920 --> 0:13:45.400 will always be into this album. This album is great 0:13:45.440 --> 0:13:47.840 and it shall always be that way. Would it do that, 0:13:47.880 --> 0:13:50.400 would it set itself in time, or would it like 0:13:50.440 --> 0:13:52.480 assume that it would always be in this and just 0:13:52.760 --> 0:13:54.600 it kind of gets back to that vampire scenario you've 0:13:54.640 --> 0:13:56.760 brought up before. You know, you don't know what you're 0:13:56.760 --> 0:13:59.559 going to want when you become the vampire, and it's 0:13:59.559 --> 0:14:01.640 hard to iagine what your mindset is when you reach 0:14:01.679 --> 0:14:04.720 that point. Yeah, yeah, that's a really good point. Certainly 0:14:04.760 --> 0:14:08.360 applies to becoming some kind of machine or merging with it, 0:14:08.480 --> 0:14:17.240 or remodeling yourself if you already are machine than now. 0:14:17.320 --> 0:14:20.240 Schneider makes a number of other arguments about the types 0:14:20.480 --> 0:14:23.640 of post biological intelligences that we would be likely to 0:14:23.760 --> 0:14:26.920 encounter again, derived from the idea that there is some 0:14:27.000 --> 0:14:33.760 kind of ancestral biological inspiration behind these hypothetical superintelligences. And 0:14:33.840 --> 0:14:37.560 the thing she zeroes in on is that some limitations 0:14:37.760 --> 0:14:42.440 from original biological organisms are things that aies would probably 0:14:42.480 --> 0:14:45.040 want to engineer out of themselves. Right now, you can 0:14:45.080 --> 0:14:47.600 think of plenty of things about your brain that if 0:14:47.760 --> 0:14:50.160 you know, your brain were to evolve into some kind 0:14:50.160 --> 0:14:53.040 of computer that was always perfecting itself, it might want 0:14:53.040 --> 0:14:56.000 to leave by the wayside over time. You know, maybe 0:14:56.000 --> 0:14:58.600 some of your obsessions and anxieties and stuff like that. 0:14:59.200 --> 0:15:01.640 But what's left if you take all that out right, Yeah, 0:15:01.720 --> 0:15:05.000 that's a good point. But then she also says that 0:15:05.080 --> 0:15:10.000 there are quote cognitive capacities that sophisticated forms of biological 0:15:10.000 --> 0:15:13.400 intelligence are likely to have and which enable the super 0:15:13.440 --> 0:15:17.040 intelligence to carry out its final and instrumental goals. We 0:15:17.040 --> 0:15:19.680 could also look for traits that are not likely to 0:15:19.720 --> 0:15:22.680 be engineered out, as they do not detract the visa 0:15:22.760 --> 0:15:26.280 from its goals. So there are some traits of biological 0:15:26.320 --> 0:15:29.520 intelligence that probably have inherent advantages. There are just some 0:15:29.560 --> 0:15:32.080 ways that brains work really good, and it would want 0:15:32.120 --> 0:15:35.720 to replicate that and just refine it across time. And 0:15:35.720 --> 0:15:39.200 then there are other traits of biological intelligences that might 0:15:39.240 --> 0:15:42.280 not have clear advantages, but they at least wouldn't detract 0:15:42.440 --> 0:15:45.080 from the attainment of goals. So just you know, why 0:15:45.120 --> 0:15:47.960 why not keep them around? Yeah, sort of the lukewarm 0:15:48.000 --> 0:15:51.760 stuff that's not detrimental to their goals. But all that 0:15:51.960 --> 0:15:54.160 doesn't maybe help it all that much, but isn't isn't 0:15:54.200 --> 0:15:57.320 using a lot of energy, et cetera. Right, So to 0:15:57.360 --> 0:16:01.640 get into Schneider's explicit predictions for bio logically inspired superintelligence 0:16:01.720 --> 0:16:03.760 is the first one. I'm not going to get deep 0:16:03.800 --> 0:16:06.200 into because it's a little dry. But this is a 0:16:06.280 --> 0:16:09.720 fair point, I guess. She says, learning about the computational 0:16:09.800 --> 0:16:12.800 structure of the brain of the species that created the 0:16:12.880 --> 0:16:17.840 visa can provide insight into the visas thinking patterns. Okay, 0:16:17.880 --> 0:16:21.640 So basically you can start to gain some insights into 0:16:21.680 --> 0:16:25.880 the computational structure of an animal's brain or or nervous 0:16:25.880 --> 0:16:29.840 system more broadly, by studying the brain's connect tome. A 0:16:29.840 --> 0:16:33.880 connect to home is a map of the connections between neurons, 0:16:33.920 --> 0:16:36.880 which at least in theory, would help you understand which 0:16:36.920 --> 0:16:39.640 cells and structures in the brain or the nervous system 0:16:39.680 --> 0:16:43.920 broadly share information with which others in order to better 0:16:44.000 --> 0:16:47.720 understand how information is processed as a whole. Yeah, I 0:16:47.720 --> 0:16:50.000 mean this makes me think, for instance, like when we 0:16:50.040 --> 0:16:54.080 think of an artificial intelligence, we are often loosely thinking 0:16:54.080 --> 0:16:56.480 of like that like that single entity. But what have 0:16:56.600 --> 0:16:58.480 you had, What have you had an alien life form 0:16:58.520 --> 0:17:01.880 that had sort of a pronoun bounced bicameral mind situation 0:17:01.960 --> 0:17:06.239 going on, where like the actual organic organism had uh 0:17:06.720 --> 0:17:09.639 like two houses of thought going on that kind of 0:17:09.680 --> 0:17:12.160 communicate with each other, and therefore that ends up being 0:17:12.200 --> 0:17:15.280 reflected in the ai they create. Oh, that's very interesting. 0:17:15.359 --> 0:17:17.879 That will actually come back to a question I have 0:17:17.920 --> 0:17:20.600 about one of the points she makes later on. But again, 0:17:20.640 --> 0:17:23.040 just the point she's making here is that if you 0:17:23.080 --> 0:17:26.840 can look at the physical structure of the original ancestral 0:17:27.000 --> 0:17:30.360 organism that the intelligence is evolved from, that can help 0:17:30.400 --> 0:17:34.800 you understand something about how the intelligence of its machine 0:17:34.800 --> 0:17:38.200 descendant works. Quote. While it is likely that a given 0:17:38.280 --> 0:17:40.640 visa will not have the same kind of connect dome 0:17:40.680 --> 0:17:43.639 as the members of the original species, some of the 0:17:43.640 --> 0:17:48.159 functional and structural connections may be retained, and interesting departures 0:17:48.200 --> 0:17:51.159 from the originals may be found. Now after that, she 0:17:51.240 --> 0:17:52.840 brings up a second point that I thought was a 0:17:52.960 --> 0:17:58.080 very interesting prediction. She writes, quote visas may have viewpoint 0:17:58.240 --> 0:18:02.680 invariant representation. Now what does that mean? Well, an easy 0:18:02.680 --> 0:18:05.119 way to think about it is this. If you're watching 0:18:05.119 --> 0:18:08.840 a movie and the camera suddenly cuts to a different 0:18:08.920 --> 0:18:11.240 angle in the middle of a scene, but it's still 0:18:11.280 --> 0:18:13.880 the same scene going on. How is it that you 0:18:13.960 --> 0:18:18.159 still understand you're watching a continuation of the same action 0:18:18.200 --> 0:18:22.159 as before. Everything looks completely different, but you understand that 0:18:22.200 --> 0:18:25.359 these are the same actors playing the same characters in 0:18:25.400 --> 0:18:29.320 the same room, even though it looks totally different. This 0:18:29.400 --> 0:18:32.680 is one of the ways that human intelligence still drastically 0:18:32.760 --> 0:18:36.320 outperforms artificial intelligence on Earth. You know, humans can look 0:18:36.320 --> 0:18:38.960 at an object it's a VHS tape of the Star 0:18:38.960 --> 0:18:41.720 Wars Holiday special, and you can look at it from 0:18:41.800 --> 0:18:44.720 completely different angles. Maybe the front cover of the box 0:18:44.800 --> 0:18:47.359 looks completely different than the back cover of the box, 0:18:47.680 --> 0:18:49.960 but you turn it around and you still understand that 0:18:50.000 --> 0:18:53.399 you're looking at the same object. Humans are able to 0:18:53.520 --> 0:18:57.240 form mental representations of objects in the world that can 0:18:57.280 --> 0:19:01.720 be isolated and recognized and manipul related within the mind's eye, 0:19:02.240 --> 0:19:05.440 and we humans are not typically going to be confused 0:19:05.560 --> 0:19:08.359 about what we're looking at because we took a step 0:19:08.400 --> 0:19:12.080 to the side and change the angle of observation. Even 0:19:12.080 --> 0:19:15.000 though the light reflecting off of the object and hitting 0:19:15.000 --> 0:19:18.400 our eyes will produce a very different pattern on the retina, 0:19:18.680 --> 0:19:22.200 we somehow still use our intelligence to know that we're 0:19:22.200 --> 0:19:25.400 still looking at the same object or scene. And this 0:19:25.480 --> 0:19:28.239 is a much much harder task for a computer. I mean, 0:19:28.280 --> 0:19:32.240 ask anybody who's been involved in visual object recognition. It's 0:19:32.280 --> 0:19:35.359 an incredibly difficult task for AI. And this is one 0:19:35.400 --> 0:19:38.879 of the many amazing fast and loose intellectual feats that 0:19:39.000 --> 0:19:41.879 humans do all the time so often that we we 0:19:42.040 --> 0:19:45.240 rarely appreciate how amazing our brains are in this regard. 0:19:45.680 --> 0:19:48.520 Another example from a recent episode was, you know, recalling 0:19:48.520 --> 0:19:51.800 the Moses Illusion episode, we talked about how good we 0:19:51.840 --> 0:19:55.080 are at getting the gist of a statement or a question. 0:19:55.560 --> 0:19:58.760 Even if major pieces of information within the sentence are 0:19:58.800 --> 0:20:01.920 wrong and should be throwing you often completely the wrong direction, 0:20:02.320 --> 0:20:05.080 you still are able to very quickly get what the 0:20:05.119 --> 0:20:09.080 person was probably intending to say and operate on that basis. 0:20:09.560 --> 0:20:13.600 Now here's where it goes with viewpoint invariant representations, especially 0:20:13.640 --> 0:20:17.080 as it concerns like physical objects in the world. Schneider 0:20:17.160 --> 0:20:20.639 argues that you can expect any biologically inspired AI to 0:20:20.720 --> 0:20:25.680 have viewpoint invariant representations because they seem to be inextricably 0:20:25.800 --> 0:20:30.400 linked to the biological development of intelligence. And uh, just 0:20:30.760 --> 0:20:33.119 I'm expanding on her thoughts here, but I think the 0:20:33.160 --> 0:20:37.080 reasoning goes something like this, What is intelligence? That's actually 0:20:37.119 --> 0:20:39.920 kind of a difficult question to answer, right, Like, it's 0:20:40.000 --> 0:20:42.240 kind of hard to pin down. But I think one 0:20:42.359 --> 0:20:47.040 plausible answer has to do with speed. Intelligence has something 0:20:47.119 --> 0:20:51.600 to do with the ability to accelerate problem solving or 0:20:51.720 --> 0:20:55.000 goal acquisition. So you could have an organism that has 0:20:55.080 --> 0:20:59.240 essentially a random strategy for trying to get what it wants, 0:20:59.600 --> 0:21:03.320 and very step it goes above a random strategy is 0:21:03.359 --> 0:21:07.000 in a way an increase in intelligence. It's accelerating the 0:21:07.040 --> 0:21:10.359 solution of problems. Now, to follow the biological reasoning a 0:21:10.400 --> 0:21:14.600 little further there, why is it that animals in general 0:21:14.720 --> 0:21:19.159 need a speed of problem solving intelligence that most plants 0:21:19.320 --> 0:21:21.960 do not? Well? I think the answer there is that 0:21:22.400 --> 0:21:27.280 animals survival and reproduction strategies are usually based on movement. 0:21:27.440 --> 0:21:30.600 This wouldn't be true of all things in the Kingdom animalia. 0:21:30.600 --> 0:21:32.920 I'm not so much for sponges and stuff, but most 0:21:32.960 --> 0:21:37.520 animals move fairly rapidly, whether that's for foraging or evading 0:21:37.560 --> 0:21:41.879 predators or seeking mates or anything like that. If you 0:21:41.920 --> 0:21:46.040 are able to move fairly quickly, that means your body 0:21:46.119 --> 0:21:49.840 needs a system of deciding in what direction to move 0:21:49.920 --> 0:21:53.800 relatively quickly. And so I could be missing something here, 0:21:53.800 --> 0:21:55.600 but it seems to me that it's a pretty safe 0:21:55.640 --> 0:21:59.080 assumption that this is one of the major drivers of 0:21:59.119 --> 0:22:02.560 the development of biological intelligence. Coming up with better and 0:22:02.600 --> 0:22:08.639 better systems for adaptively optimizing strategies for rapid movement to 0:22:08.720 --> 0:22:11.200 fit the specifics of the situation you're in. So you're 0:22:11.200 --> 0:22:15.879 constantly faced with new situations, predator approaching from a different angle, 0:22:16.480 --> 0:22:19.520 food to be found in a different you know, orientation, 0:22:19.640 --> 0:22:22.600 or like in a different hard to reach space, and 0:22:22.680 --> 0:22:26.040 your body needs a way to adapt to whatever situation 0:22:26.119 --> 0:22:29.280 you're in to decide the best way to move. Yeah, 0:22:29.320 --> 0:22:32.800 it kind of comes down to a certain extent passive 0:22:33.080 --> 0:22:39.480 energy acquisition versus active energy acquisition. Yes, because you know, obviously, 0:22:39.480 --> 0:22:42.320 if you have passive energy acquisition, you don't necessarily need 0:22:42.359 --> 0:22:43.760 to move as much. You know, you can just sort 0:22:43.760 --> 0:22:45.880