0:00:00.000,0:00:09.044 preroll music 0:00:09.044,0:00:14.049 Herald: Our next talk is going to be about AI and[br]it's going to be about proper AI. 0:00:14.049,0:00:17.730 It's not going to be about[br]deep learning or buzz word bingo. 0:00:17.730,0:00:22.590 It's going to be about actual psychology.[br]It's going to be about computational metapsychology. 0:00:22.590,0:00:25.750 And now please welcome Joscha! 0:00:25.750,0:00:33.050 applause 0:00:33.050,0:00:35.620 Joscha: Thank you. 0:00:35.620,0:00:37.710 I'm interested in understanding[br]how the mind works, 0:00:37.710,0:00:42.640 and I believe that the most foolproof perspective[br]at looking ... of looking at minds is to understand 0:00:42.640,0:00:46.600 that they are systems that if you saw patterns[br]at them you find meaning. 0:00:46.600,0:00:51.700 And you find meaning in those in very particular[br]ways and this is what makes us who we are. 0:00:51.700,0:00:55.239 So they way to study and understand who we[br]are in my understanding is 0:00:55.239,0:01:01.149 to build models of information processing[br]that constitutes our minds. 0:01:01.149,0:01:05.640 Last year about the same time, I've answered[br]the four big questions of philosophy: 0:01:05.640,0:01:08.510 "Whats the nature of reality?", "What can[br]be known?", "Who are we?", 0:01:08.510,0:01:14.650 "What should we do?"[br]So now, how can I top this? 0:01:14.650,0:01:18.720 applause 0:01:18.720,0:01:22.849 I'm going to give you the drama[br]that divided a planet. 0:01:22.849,0:01:26.470 Some of a very, very big events,[br]that happened in the course of last year, 0:01:26.470,0:01:30.080 so I couldn't tell you about it before. 0:01:30.080,0:01:38.489 What color is the dress[br]laughsapplause 0:01:38.489,0:01:44.720 I mean ahmm... If you have.. do not have any[br]mental defects you can clearly see it's white 0:01:44.720,0:01:46.550 and gold. Right? 0:01:46.550,0:01:48.720 [voices from audience] 0:01:48.720,0:01:53.009 Turns out, ehmm.. most people seem to have[br]mental defects and say it is blue and black. 0:01:53.009,0:01:57.500 I have no idea why. Well Ok, I have an idea,[br]why that is the case. 0:01:57.500,0:02:01.170 Ehmm, I guess that you got too, it has to[br]do with color renormalization 0:02:01.170,0:02:04.720 and color renormalization happens differently[br]apparently in different people. 0:02:04.720,0:02:09.000 So we have different wireing to renormalize[br]the white balance. 0:02:09.000,0:02:12.650 And it seems to work in real world[br]situations in pretty much the same way, 0:02:12.650,0:02:18.000 but not necessarily for photographs.[br]Which have only very small fringe around them, 0:02:18.000,0:02:20.600 which gives you hint about the lighting situation. 0:02:20.600,0:02:27.000 And that's why you get this huge divergencies,[br]which is amazing! 0:02:27.000,0:02:29.660 So what we see that our minds can not know 0:02:29.660,0:02:33.250 objective truths in any way. Outside of mathematics. 0:02:33.250,0:02:36.340 They can generate meaning though. 0:02:36.340,0:02:38.760 How does this work? 0:02:38.760,0:02:42.010 I did robotic soccer for a while,[br]and there you have the situation, 0:02:42.010,0:02:45.150 that you have a bunch of robots, that are[br]situated on a playing field. 0:02:45.150,0:02:48.480 And they have a model of what goes on[br]in the playing field. 0:02:48.480,0:02:52.050 Physics generates data for their sensors.[br]They read the bits of the sensors. 0:02:52.050,0:02:55.900 And then they use them to.. erghmm update[br]the world model. 0:02:55.900,0:02:59.020 And sometimes we didn't want[br]to take the whole playing field along, 0:02:59.020,0:03:03.380 and the physical robots, because they are[br]expensive and heavy and so on. 0:03:03.380,0:03:06.480 Instead if you just want to improve the learning[br]and the game play of the robots 0:03:06.480,0:03:07.800 you can use the simulations. 0:03:07.800,0:03:11.200 So we've wrote a computer simulation of the[br]playing field and the physics, and so on, 0:03:11.200,0:03:15.210 that generates pretty some the same data,[br]and put the robot mind into the simulator 0:03:15.210,0:03:17.040 robot body, and it works just as well. 0:03:17.040,0:03:20.590 That is, if you the robot, because you can[br]not know the difference if you are the robot. 0:03:20.590,0:03:24.460 You can not know what's out there. The only[br]thing that you get to see is what is the structure 0:03:24.460,0:03:27.530 of the data at you system bit interface. 0:03:27.530,0:03:30.090 And then you can derive model from this. 0:03:30.090,0:03:32.960 And this is pretty much the situation[br]that we are in. 0:03:32.960,0:03:38.180 That is, we are minds that are somehow computational, 0:03:38.180,0:03:40.700 they are able to find regularity in patterns, 0:03:40.700,0:03:44.530 and they are... we.. seem to have access to[br]something that is full of regularity, 0:03:44.530,0:03:46.630 so we can make sense out of it. 0:03:46.630,0:03:48.930 [ghulp, ghulp] 0:03:48.930,0:03:52.800 Now, if you discover that you are in the same[br]situation as these robots, 0:03:52.800,0:03:56.180 basically you discover that you are some kind[br]of apparently biological robot, 0:03:56.180,0:03:58.530 that doesn't have direct access[br]to the world of concepts. 0:03:58.530,0:04:02.140 That has never actually seen matter[br]and energy and other people. 0:04:02.140,0:04:04.890 All it got to see was little bits of information, 0:04:04.890,0:04:06.270 that were transmitted through the nerves, 0:04:06.270,0:04:07.870 and the brain had to make sense of them, 0:04:07.870,0:04:10.470 by counting them in elaborate ways. 0:04:10.470,0:04:12.720 What's the best model of the world[br]that you can have with this? 0:04:12.720,0:04:16.530 What will the state of affairs,[br]what's the system that you are in? 0:04:16.530,0:04:20.920 And what are the best algorithms that you[br]should be using, to fix your world model. 0:04:20.920,0:04:23.310 And this question is pretty old. 0:04:23.310,0:04:27.750 And I think that has been answered for the[br]first time by Ray Solomonoff in the 1960. 0:04:27.750,0:04:30.840 He has discovered an algorithm,[br]that you can apply when you discover 0:04:30.840,0:04:33.540 that you are an robot,[br]and all you have is data. 0:04:33.540,0:04:34.870 What is the world like? 0:04:34.870,0:04:40.990 And this algorithm is basically[br]a combination of induction and Occam's razor. 0:04:40.990,0:04:45.710 And we can mathematically prove that we can[br]not do better than Solomonoff induction. 0:04:45.710,0:04:51.380 Unfortunately, Solomonoff induction[br]is not quite computable. 0:04:51.380,0:04:54.450 But everything that we are going to do is[br]some... is going to be some approximation 0:04:54.450,0:04:55.820 of Salomonoff induction. 0:04:55.820,0:04:59.400 So our concepts can not really refer[br]to the facts in the world out there. 0:04:59.400,0:05:02.380 We do not get the truth by referring[br]to stuff out there, in the world. 0:05:02.380,0:05:07.960 We get meaning by suitably encoding[br]the patterns at our systemic interface. 0:05:07.960,0:05:12.270 And AI has recently made a huge progress in[br]encoding data at perceptual interfaces. 0:05:12.270,0:05:15.900 Deep learning is about using a stacked hierarchy[br]of feature detectors. 0:05:15.900,0:05:21.280 That is, we use pattern detectors and we build[br]them into a networks that are arranged in 0:05:21.280,0:05:23.030 hundreds of layers. 0:05:23.030,0:05:26.500 And then we adjust the links[br]between these layers. 0:05:26.500,0:05:29.380 Usually some kind of... using[br]some kind of gradient descent. 0:05:29.380,0:05:33.220 And we can use this to classify[br]for instance images and parts of speech. 0:05:33.220,0:05:37.950 So, we get to features that are more and more[br]complex, they started as very, very simple patterns. 0:05:37.950,0:05:41.290 And then get more and more complex,[br]until we get to object categories. 0:05:41.290,0:05:44.199 And now this systems are able[br]in image recognition task, 0:05:44.199,0:05:47.480 to approach performance that is very similar[br]to human performance. 0:05:47.480,0:05:52.040 Also what is nice is that it seems to be somewhat[br]similar to what the brain seems to be doing 0:05:52.040,0:05:53.740 in visual processing. 0:05:53.740,0:05:57.570 And if you take the activation in different[br]levels of these networks and you 0:05:57.570,0:06:01.430 erghm... improve the... that... erghmm...[br]enhance this activation a little bit, what 0:06:01.430,0:06:03.500 you get is stuff that look very psychedelic. 0:06:03.500,0:06:09.620 Which may be similar to what happens, if you[br]put certain illegal substances into people, 0:06:09.620,0:06:13.650 and enhance the activity on certain layers[br]of their visual processing. 0:06:13.650,0:06:21.540 [BROKEN AUDIO]If you want to classify the[br]differences what we do if we want quantify 0:06:21.540,0:06:33.030 this you filter out all the invariences in[br]the data. 0:06:33.030,0:06:36.360 The pose that she has, the lighting,[br]the dress that she is on.. has on, 0:06:36.360,0:06:38.020 her facial expression and so on. 0:06:38.020,0:06:42.900 And then we go to only to this things that[br]is left after we've removed all the nuance data. 0:06:42.900,0:06:47.410 But what if we... erghmm[br]want to get to something else, 0:06:47.410,0:06:49.850 for instance if we want to understand poses. 0:06:49.850,0:06:53.240 Could be for instance that we have several[br]dancers and we want to understand what they 0:06:53.240,0:06:54.400 have in common. 0:06:54.400,0:06:58.330 So our best bet is not just to have a single[br]classification based filtering, 0:06:58.330,0:07:01.199 but instead what we want to have is to take[br]the low level input 0:07:01.199,0:07:05.180 and get a whole universe of features,[br]that is interrelated. 0:07:05.180,0:07:07.220 So we have different levels of interrelations. 0:07:07.220,0:07:08.960 At the lowest levels we have percepts. 0:07:08.960,0:07:11.580 On the slightly higher level we have simulations. 0:07:11.580,0:07:16.920 And on even higher level we have concept landscape. 0:07:16.920,0:07:19.300 How does this representation[br]by simulation work? 0:07:19.300,0:07:22.229 Now imagine you want to understand sound. 0:07:22.229,0:07:23.669 [Ghulp] 0:07:23.669,0:07:26.710 If you are a brain and you want to understand[br]sound you need to model it. 0:07:26.710,0:07:31.070 Unfortunatly we can not really model sound[br]with neurons, because sound goes up to 20kHz, 0:07:31.070,0:07:36.660 or if you are old like me maybe to 12 kHz.[br]20 kHz is what babies could do. 0:07:36.660,0:07:41.240 And... neurons do not want to do 20 kHz.[br]That's way too fast for them. 0:07:41.240,0:07:43.250 They like something like 20 Hz. 0:07:43.250,0:07:45.590 So what do you do? You need[br]to make a Fourier transform. 0:07:45.590,0:07:49.650 The Fourier transform measures the amount[br]of energy at different frequencies. 0:07:49.650,0:07:52.500 And because you can not do it with neurons,[br]you need to do it in hardware. 0:07:52.500,0:07:54.180 And turns out this is exactly[br]what we are doing. 0:07:54.180,0:07:59.860 We have this cochlea which is this snail like[br]thing in our ears, 0:07:59.860,0:08:06.669 and what it does, it transforms energy of[br]sound in different frequency intervals into 0:08:06.669,0:08:08.009 energy measurments. 0:08:08.009,0:08:10.479 And then gives you something[br]like what you see here. 0:08:10.479,0:08:12.550 And this is something that the brain can model, 0:08:12.550,0:08:16.210 so we can get a neurosimulator that tries[br]to recreate this patterns. 0:08:16.210,0:08:21.370 And we can predict the next input from the[br]cochlea that then understand the sound. 0:08:21.370,0:08:23.410 Of course if you want to understand music, 0:08:23.410,0:08:25.160 we have to go beyond understanding sound. 0:08:25.160,0:08:29.340 We have to understand the transformations[br]that sound can have if you play it at different pitch. 0:08:29.340,0:08:33.599 We have to arrange the sound in the sequence[br]that give you rhythms and so on. 0:08:33.599,0:08:35.889 And then we want to identify[br]some kind of musical grammar 0:08:35.889,0:08:38.799 that we can use to again control the sequencer. 0:08:38.799,0:08:42.529 So we have stucked structures.[br]That simulate the world. 0:08:42.529,0:08:44.319 And once you've learned this model of music, 0:08:44.319,0:08:47.309 once you've learned the musical grammar,[br]the sequencer and the sounds. 0:08:47.309,0:08:51.779 You can get to the structure[br]of the individual piece of music. 0:08:51.779,0:08:54.399 So, if you want to model the world of music. 0:08:54.399,0:08:58.279 You need to have the lowest level of percepts[br]then we have the higher level of mental simulations. 0:08:58.279,0:09:01.910 And... which give the sequences of the music[br]and the grammars of music. 0:09:01.910,0:09:05.149 And beyond this you have the conceptual landscape[br]that you can use 0:09:05.149,0:09:08.249 to describe different styles of music. 0:09:08.249,0:09:12.130 And if you go up in the hierarchy,[br]you get to more and more abstract models. 0:09:12.130,0:09:13.860 More and more conceptual models. 0:09:13.860,0:09:16.449 And more and more analytic models. 0:09:16.449,0:09:18.160 And this are causal models at some point. 0:09:18.160,0:09:20.999 This causal models can be weakly deterministic, 0:09:20.999,0:09:22.980 basically associative models, which tell you 0:09:22.980,0:09:27.339 if this state happens, it's quite probable[br]that this one comes afterwords. 0:09:27.339,0:09:29.389 Or you can get to a strongly determined model. 0:09:29.389,0:09:32.730 Strongly determined model is one which tells[br]you, if you are in this state 0:09:32.730,0:09:33.879 and this condition is met, 0:09:33.879,0:09:35.589 You are are going to go exactly in this state. 0:09:35.589,0:09:40.110 If this condition is not met, or a different[br]condition is met, you are going to this state. 0:09:40.110,0:09:41.449 And this is what we call an alghorithm. 0:09:41.449,0:09:46.769 it's.. now we are on the domain of computation. 0:09:46.769,0:09:48.730 Computation is slightly different from mathematics. 0:09:48.730,0:09:51.179 It's important to understand this. 0:09:51.179,0:09:54.699 For a long time people have thought that the[br]universe is written in mathematics. 0:09:54.699,0:09:58.399 Or that.. minds are mathematical,[br]or anything is mathematical. 0:09:58.399,0:10:00.439 In fact nothing is mathematical. 0:10:00.439,0:10:04.529 Mathematics is just the domain[br]of formal languages. It doesn't exist. 0:10:04.529,0:10:07.300 Mathematics starts with a void. 0:10:07.300,0:10:11.939 You throw in a few axioms, and if you've chosen[br]a nice axioms, then you get infinite complexity. 0:10:11.939,0:10:13.679 Most of which is not computable. 0:10:13.679,0:10:16.270 In mathematics you can express arbitrary statements, 0:10:16.270,0:10:18.269 because it's all about formal languages. 0:10:18.269,0:10:20.369 Many of this statements will not make sense. 0:10:20.369,0:10:22.469 Many of these statements will make sense[br]in some way, 0:10:22.469,0:10:24.429 but you can not test whether they make sense, 0:10:24.429,0:10:26.740 because they're not computable. 0:10:26.740,0:10:29.929 Computation is different.[br]Computation can exist. 0:10:29.929,0:10:32.459 It's starts with an initial state. 0:10:32.459,0:10:34.739 And then you have a transition function.[br]You do the work. 0:10:34.739,0:10:38.449 You apply the transition function,[br]and you get into the next state. 0:10:38.449,0:10:41.249 Computation is always finite. 0:10:41.249,0:10:43.689 Mathematics is the kingdom of specification. 0:10:43.689,0:10:47.290 And computation is the kingdom of implementation. 0:10:47.290,0:10:50.629 It's very important to understand this difference. 0:10:50.629,0:10:55.329 All our access to mathematics of course is[br]because we do computation. 0:10:55.329,0:10:57.459 We can understand mathematics, 0:10:57.459,0:10:59.939 because our brain can compute[br]some parts of mathematics. 0:10:59.939,0:11:04.439 Very, very little of it, and to[br]very constrained complexity. 0:11:04.439,0:11:06.860 But enough, so we can map[br]some of the infinite complexity 0:11:06.860,0:11:10.410 and noncomputability of mathematics[br]into computational patterns, 0:11:10.410,0:11:12.279 that we can explore. 0:11:12.279,0:11:14.410 So computation is about doing the work, 0:11:14.410,0:11:16.939 it's about executing the transition function. 0:11:19.730,0:11:22.899 Now we've seen that mental representation[br]is about concepts, 0:11:22.899,0:11:25.670 mental simulations, conceptual representations 0:11:25.670,0:11:29.110 and this conceptual representations[br]give us concept spaces. 0:11:29.110,0:11:30.970 And the nice thing[br]about this concept spaces is 0:11:30.970,0:11:33.399 that they give us an interface[br]to our mental representations, 0:11:33.399,0:11:36.290 We can use to address and manipulate them. 0:11:36.290,0:11:39.119 And we can share them in cultures. 0:11:39.119,0:11:40.899 And this concepts are compositional. 0:11:40.899,0:11:43.639 We can put them together, to create new concepts. 0:11:43.639,0:11:48.230 And they can be described using[br]higher dimensional vector spaces. 0:11:48.230,0:11:50.319 They don't do simulation[br]and prediction and so on, 0:11:50.319,0:11:53.119 but we can capture regularity[br]in our concept wisdom. 0:11:53.119,0:11:55.220 With this vector space[br]you can do amazing things. 0:11:55.220,0:11:57.589 For instance, if you take the vector from[br]"King" to "Queen" 0:11:57.589,0:12:01.009 is pretty much the same vector[br]as to.. between "Man" and "Woman" 0:12:01.009,0:12:04.110 And because of this properties, because it's[br]really a high dimentional manifold 0:12:04.110,0:12:07.569 this concepts faces, we can do interesting[br]things, like machine translation 0:12:07.569,0:12:09.470 without understanding what it means. 0:12:09.470,0:12:13.929 That is without doing any proper mental representation,[br]that predicts the world. 0:12:13.929,0:12:16.989 So this is a type of meta representation,[br]that is somewhat incomplete, 0:12:16.989,0:12:21.199 but it captures the landscape that we share[br]in a culture. 0:12:21.199,0:12:25.089 And then there is another type of meta representation,[br]that is linguistic protocols. 0:12:25.089,0:12:27.699 Which is basically a formal grammar and vocabulary. 0:12:27.699,0:12:29.619 And we need this linguistic protocols 0:12:29.619,0:12:32.869 to transfer mental representations[br]between people. 0:12:32.869,0:12:36.019 And we do this by basically[br]scanning our mental representation, 0:12:36.019,0:12:38.660 disassembling them in some way[br]or disambiguating them. 0:12:38.660,0:12:43.040 And then we use it as discrete string of symbols[br]to get it to somebody else, 0:12:43.040,0:12:46.429 and he trains an assembler,[br]that reverses this process, 0:12:46.429,0:12:51.389 and build something that is pretty similar[br]to what we intended to convey. 0:12:51.389,0:12:53.569 And if you look at the progression of AI models, 0:12:53.569,0:12:55.600 it pretty much went the opposite direction. 0:12:55.600,0:13:00.279 So AI started with linguistic protocols, which[br]were expressed in formal grammars. 0:13:00.279,0:13:05.209 And then it got to concepts spaces, and now[br]it's about to address percepts. 0:13:05.209,0:13:09.689 And at some point in near future it's going[br]to get better at mental simulations. 0:13:09.689,0:13:11.730 And at some point after that we get to 0:13:11.730,0:13:14.769 attention directed and[br]motivationally connected systems, 0:13:14.769,0:13:16.600 that make sense of the world. 0:13:16.600,0:13:20.290 that are in some sense able to address meaning. 0:13:20.290,0:13:23.489 This is the hardware that we have can do. 0:13:23.489,0:13:25.629 What kind of hardware do we have? 0:13:25.629,0:13:28.480 That's a very interesting question. 0:13:28.480,0:13:32.230 It could start out with a question:[br]How difficult is it to define a brain? 0:13:32.230,0:13:35.439 We know that the brain must be[br]somewhere hidden in the genome. 0:13:35.439,0:13:38.290 The genome fits on a CD ROM.[br]It's not that complicated. 0:13:38.290,0:13:40.399 It's easier than Microsoft Windows. laughter 0:13:40.399,0:13:45.549 And we also know, that about 2%[br]of the genome is coding for proteins. 0:13:45.549,0:13:48.429 And maybe about 10% of the genome[br]has some kind of stuff 0:13:48.429,0:13:51.239 that tells you when to switch protein. 0:13:51.239,0:13:52.829 And the remainder is mostly garbage. 0:13:52.829,0:13:57.170 It's old viruses that are left over and has[br]never been properly deleted and so on. 0:13:57.170,0:14:01.420 Because there are no real[br]code revisions in the genome. 0:14:01.420,0:14:08.119 So how much of this 10%[br]that is 75 MB code for the brain. 0:14:08.119,0:14:09.469 We don't really know. 0:14:09.469,0:14:13.399 What we do know is we share[br]almost all of this with mice. 0:14:13.399,0:14:15.769 Genetically speaking human[br]is a pretty big mouse. 0:14:15.769,0:14:21.049 With a few bits changed, so.. to fix some[br]of the genetic expressions 0:14:21.049,0:14:25.879 And that is most of the stuff there is going[br]to code for cells and metabolism 0:14:25.879,0:14:27.999 and how your body looks like and so on. 0:14:27.999,0:14:33.679 But if you look at erghmm... how much is expressed[br]in the brain and only in the brain, 0:14:33.679,0:14:35.170 in terms of proteins and so on. 0:14:35.170,0:14:45.639 We find it's about... well of the 2% it's[br]about 5%. That is only the 5% of the 2% that 0:14:45.639,0:14:46.799 is only in the brain. 0:14:46.799,0:14:50.199 And another 5% of the 2% is predominantly[br]in the brain. 0:14:50.199,0:14:52.069 That is more in the brain than anywhere else. 0:14:52.069,0:14:54.249 Which gives you some kind of thing[br]like a lower bound. 0:14:54.249,0:14:59.379 Which means to encode a brain genetically[br]base on the hardware that we are using. 0:14:59.379,0:15:03.539 We need something like[br]at least 500 kB of code. 0:15:03.539,0:15:06.670 Actually ehmm.. this... we very conservative[br]lower bound. 0:15:06.670,0:15:08.720 It's going to be a little more I guess. 0:15:08.720,0:15:11.449 But it sounds surprisingly little, right? 0:15:11.449,0:15:13.709 But in terms of scientific theories[br]this is a lot. 0:15:13.709,0:15:16.519 I mean the universe,[br]according to the core theory 0:15:16.519,0:15:19.420 of the quantum mechanics and so on[br]is like so much of code. 0:15:19.420,0:15:20.569 It's like half a page of code. 0:15:20.569,0:15:23.100 That's it. That's all you need[br]to generate the universe. 0:15:23.100,0:15:25.489 And if you want to understand evolution[br]it's like a paragraph. 0:15:25.489,0:15:29.609 It's couple lines you need to understand[br]evolutionary process. 0:15:29.609,0:15:32.199 And there is a lots, lots of details, that's[br]you get afterwards. 0:15:32.199,0:15:34.220 Because this process itself doesn't define 0:15:34.220,0:15:37.259 how the animals are going to look like,[br]and in similar way is.. 0:15:37.259,0:15:41.269 the code of the universe doesn't tell you[br]what this planet is going to look like. 0:15:41.269,0:15:43.279 And what you guys are going to look like. 0:15:43.279,0:15:45.949 It's just defining the rulebook. 0:15:45.949,0:15:49.209 And in the same sense genome defines the rulebook, 0:15:49.209,0:15:51.569 by which our brain is build. 0:15:51.569,0:15:56.399 erghmmm,.. The brain boots itself[br]into developer process, 0:15:56.399,0:15:58.119 and this booting takes some time. 0:15:58.119,0:16:01.069 So subliminal learning in which[br]initial connections are forged 0:16:01.069,0:16:04.910 And basic models are build of the world,[br]so we can operate in it. 0:16:04.910,0:16:06.999 And how long does this booting take? 0:16:06.999,0:16:09.669 I thing it's about 80 mega seconds. 0:16:09.669,0:16:14.319 That's the time that a child is awake until[br]it's 2.5 years old. 0:16:14.319,0:16:16.449 By this age you understand Star Wars. 0:16:16.449,0:16:20.029 And I think that everything after[br]understanding Star Wars is cosmetics. 0:16:20.029,0:16:26.799 laughterapplause 0:16:26.799,0:16:32.820 You are going to be online, if you get to[br]arrive old age for about 1.5 giga seconds. 0:16:32.820,0:16:37.929 And in this time I think you are going to[br]get not to watch more than 5 milion concepts. 0:16:37.929,0:16:41.600 Why? I don't know real...[br]If you look at this child. 0:16:41.600,0:16:45.480 If a child would be able to form a concept[br]let say every 5 minutes, 0:16:45.480,0:16:48.529 then by the time it's about 4 years old,[br]it's going to have 0:16:48.529,0:16:51.549 something like 250 thousands concepts. 0:16:51.549,0:16:54.119 And... so... a quarter million. 0:16:54.119,0:16:56.809 And if we extrapolate this into our lifetime, 0:16:56.809,0:16:59.799 at some point it slows down,[br]because we have enough concepts, 0:16:59.799,0:17:01.230 to describe the world. 0:17:01.230,0:17:04.410 Maybe it's something... It's I think it's[br]less that 5 million. 0:17:04.410,0:17:07.140 How much storage capacity does the brain has? 0:17:07.140,0:17:12.319 I think that the... the estimates[br]are pretty divergent, 0:17:12.319,0:17:14.930 The lower bound is something like a 100 GB, 0:17:14.930,0:17:18.569 And the upper bound[br]is something like 2.5 PB. 0:17:18.569,0:17:21.890 There is even...[br]even some higher outliers this.. 0:17:21.890,0:17:25.630 If you for instance think that we need all[br]those synaptic vesicle to store information, 0:17:25.630,0:17:27.530 maybe even more fits into this. 0:17:27.530,0:17:31.740 But the 2.5 PB is usually based[br]on what you need 0:17:31.740,0:17:34.760 to code the information[br]that is in all the neurons. 0:17:34.760,0:17:36.770 But maybe the neurons[br]do not really matter so much, 0:17:36.770,0:17:39.930 because if the neuron dies it's not like the[br]word is changing dramatically. 0:17:39.930,0:17:44.270 The brain is very resilient[br]against individual neurons failing. 0:17:44.270,0:17:48.930 So the 100 GB capacity is much more[br]what you actually store in the neurons. 0:17:48.930,0:17:51.380 If you look at all the redundancy[br]that you need. 0:17:51.380,0:17:54.230 And I think this is much closer to the actual[br]Ballpark figure. 0:17:54.230,0:17:58.130 Also if you want to store 5 hundred...[br]5 million concepts, 0:17:58.130,0:18:02.330 and maybe 10 times or 100 times the number[br]of percepts, on top of this, 0:18:02.330,0:18:05.490 this is roughly the Ballpark figure[br]that you are going to need. 0:18:05.490,0:18:07.110 So our brain 0:18:07.110,0:18:08.320 is a prediction machine. 0:18:08.320,0:18:11.490 It... What it does is it reduces the entropy[br]of the environment, 0:18:11.490,0:18:14.610 to solve whatever problems you are encountering, 0:18:14.610,0:18:17.790 if you don't have a... feedback loop, to fix[br]them. 0:18:17.790,0:18:20.240 So normally if something happens, we have[br]some kind of feedback loop, 0:18:20.240,0:18:23.440 that regulates our temperature or that makes[br]problems go away. 0:18:23.440,0:18:26.050 And only when this is not working[br]we employ recognition. 0:18:26.050,0:18:29.250 And then we start this arbitrary[br]computational processes, 0:18:29.250,0:18:31.830 that is facilitated by the neural cortex. 0:18:31.830,0:18:34.940 And this.. arhmm.. neural cortex has really[br]do arbitrary programs. 0:18:34.940,0:18:37.870 But it can do so[br]with only with very limited complexity, 0:18:37.870,0:18:42.070 because really you just saw,[br]it's not that complex. 0:18:42.070,0:18:43.900 The modeling of the world is very slow. 0:18:43.900,0:18:46.570 And it's something[br]that we see in our eye models. 0:18:46.570,0:18:48.150 To learn the basic structure of the world 0:18:48.150,0:18:49.330 takes a very long time. 0:18:49.330,0:18:52.650 To learn basically that we are moving in 3D[br]and objects are moving, 0:18:52.650,0:18:54.030 and what they look like. 0:18:54.030,0:18:55.130 Once we have this basic model, 0:18:55.130,0:18:59.300 we can get to very, very quick[br]understanding within this model. 0:18:59.300,0:19:02.110 Basically encoding based[br]on the structure of the world, 0:19:02.110,0:19:03.610 that we've learned. 0:19:03.610,0:19:07.100 And this is some kind of[br]data compression, that we are doing. 0:19:07.100,0:19:09.740 We use this model, this grammar of the world, 0:19:09.740,0:19:12.150 this simulation structures that we've learned, 0:19:12.150,0:19:15.190 to encode the world very, very efficently. 0:19:15.190,0:19:17.740 How much data compression do we get? 0:19:17.740,0:19:19.860 Well... if you look at the retina. 0:19:19.860,0:19:24.610 The retina get's data[br]in the order of about 10Gb/s. 0:19:24.610,0:19:27.500 And the retina already compresses these data, 0:19:27.500,0:19:31.120 and puts them into optic nerve[br]at the rate of about 1Mb/s 0:19:31.120,0:19:34.030 This is what you get fed into visual cortex. 0:19:34.030,0:19:36.370 And the visual cortex[br]does some additional compression, 0:19:36.370,0:19:42.110 and by the time it gets to layer four of the[br]first layer of vision, to V1. 0:19:42.110,0:19:46.880 We are down to something like 1Kb/s. 0:19:46.880,0:19:50.720 So if we extrapolate this, and you get live[br]to the age of 80 years, 0:19:50.720,0:19:54.140 and you are awake for 2/3 of your lifetime. 0:19:54.140,0:19:56.930 That is you have your eyes open for 2/3 of[br]your lifetime. 0:19:56.930,0:19:59.040 The stuff that you get into your brain, 0:19:59.040,0:20:03.700 via your visual perception[br]is going to be only 2TB. 0:20:03.700,0:20:05.370 Only 2TB of visual data. 0:20:05.370,0:20:06.680 Throughout all your lifetime. 0:20:06.680,0:20:09.430 That's all you are going to get ever to see. 0:20:09.430,0:20:11.160 Isn't this depressing? 0:20:11.160,0:20:12.790 laughter 0:20:12.790,0:20:16.540 So I would really like to eghmm..[br]to tell you, 0:20:16.540,0:20:22.750 choose wisely what you[br]are going to look at. laughter 0:20:22.750,0:20:26.940 Ok. Let's look at this problem of neural compositionality. 0:20:26.940,0:20:29.250 Our brains has this amazing thing[br]that they can put 0:20:29.250,0:20:31.510 meta representation together very, very quickly. 0:20:31.510,0:20:33.150 For instance you read a page of code, 0:20:33.150,0:20:35.190 you compile it in you mind[br]into some kind of program 0:20:35.190,0:20:37.700 it tells you what this page is going to do. 0:20:37.700,0:20:39.110 Isn't that amazing? 0:20:39.110,0:20:40.810 And then you can forget about this, 0:20:40.810,0:20:43.910 disassemble it all, and use the[br]building blocks for something else. 0:20:43.910,0:20:45.230 It's like legos. 0:20:45.230,0:20:48.000 How you can do this with neurons? 0:20:48.000,0:20:50.160 Legos can do this, because they have[br]a well defined interface. 0:20:50.160,0:20:52.180 They have all this slots, you know,[br]that fit together 0:20:52.180,0:20:53.600 in well defined ways. 0:20:53.600,0:20:54.530 How can neurons do this? 0:20:54.530,0:20:57.280 Well, neurons can maybe learn[br]the interface of other neurons. 0:20:57.280,0:20:59.780 But that's difficult, because every neuron[br]looks slightly different, 0:20:59.780,0:21:04.830 after all this... some kind of biologically[br]grown natural stuff. 0:21:04.830,0:21:06.610 laughter 0:21:06.610,0:21:10.620 So what you want to do is,[br]you want to encapsulate this erhmm... 0:21:10.620,0:21:13.020 diversity of the neurons to make the predictable. 0:21:13.020,0:21:14.820 To give them well defined interface. 0:21:14.820,0:21:16.410 And I think that nature solution to this 0:21:16.410,0:21:19.770 is cortical columns. 0:21:19.770,0:21:24.250 Cortical column is a circuit of[br]between 100 and 400 neurons. 0:21:24.250,0:21:26.860 And this circuit has some kind of neural network, 0:21:26.860,0:21:28.650 that can learn stuff. 0:21:28.650,0:21:31.070 And after it has learned particular function, 0:21:31.070,0:21:35.320 and in between, it's able to link up these[br]other cortical columns. 0:21:35.320,0:21:37.120 And we have about 100 million of those. 0:21:37.120,0:21:39.770 Depending on how many neurons[br]you assume is in there, 0:21:39.770,0:21:41.490 it's... erghmm we guess it's something, 0:21:41.490,0:21:46.500 at least 20 million and maybe[br]something like a 100 million. 0:21:46.500,0:21:48.330 And this cortical columns, what they can do, 0:21:48.330,0:21:50.280 is they can link up like lego bricks, 0:21:50.280,0:21:54.130 and then perform,[br]by transmitting information between them, 0:21:54.130,0:21:55.990 pretty much arbitrary computations. 0:21:55.990,0:21:57.540 What kind of computation? 0:21:57.540,0:22:00.130 Well... Solomonoff induction. 0:22:00.130,0:22:03.820 And... they have some short range links,[br]to their neighbors. 0:22:03.820,0:22:05.690 Which comes almost for free, because erghmm.. 0:22:05.690,0:22:08.490 well, they are connected to them,[br]they are direct neighborhood. 0:22:08.490,0:22:10.050 And they have some long range connectivity, 0:22:10.050,0:22:13.000 so you can combine everything[br]in your cortex with everything. 0:22:13.000,0:22:14.900 So you need some kind of global switchboard. 0:22:14.900,0:22:17.630 Some grid like architecture[br]of long range connections. 0:22:17.630,0:22:18.900 They are going to be more expensive, 0:22:18.900,0:22:20.640 they are going to be slower, 0:22:20.640,0:22:23.590 but they are going to be there. 0:22:23.590,0:22:26.070 So how can we optimize[br]what these guys are doing? 0:22:26.070,0:22:28.270 In some sense it's like an economy. 0:22:28.270,0:22:31.460 It's not enduring based system,[br]as we often use in machine learning. 0:22:31.460,0:22:32.780 It's really an economy. You have... 0:22:32.780,0:22:35.560 The question is, you have a fixed number of[br]elements, 0:22:35.560,0:22:37.970 how can you do the most valuable stuff with[br]them. 0:22:37.970,0:22:41.030 Fixed resources, most valuable stuff, the[br]problem is economy. 0:22:41.030,0:22:43.320 So you have an economy of information brokers. 0:22:43.320,0:22:45.830 Every one of these guys,[br]this little cortical columns, 0:22:45.830,0:22:48.150 is very simplistic information broker. 0:22:48.150,0:22:50.950 And they trade rewards against neg entropy, 0:22:50.950,0:22:54.140 Against reducing entropy in the...[br]in the world. 0:22:54.140,0:22:55.790 And to do this, as we just saw 0:22:55.790,0:22:58.890 that they need some kind of standardized interface. 0:22:58.890,0:23:02.090 And internally, to use this interface[br]they are going to 0:23:02.090,0:23:03.880 have some kind of state machine. 0:23:03.880,0:23:05.660 And then they are going to pass messages 0:23:05.660,0:23:07.400 between each other. 0:23:07.400,0:23:08.630 And what are these messages? 0:23:08.630,0:23:11.100 Well, it's going to be hard[br]to discover these messages, 0:23:11.100,0:23:12.800 by looking at brains. 0:23:12.800,0:23:14.800 Because it's very difficult to see in brains, 0:23:14.800,0:23:15.450 what the are actually doing. 0:23:15.450,0:23:17.250 you just see all these neurons. 0:23:17.250,0:23:18.790 And if you would be waiting for neuroscience, 0:23:18.790,0:23:20.970 to discover anything, we wouldn't even have 0:23:20.970,0:23:22.590 gradient descent or anything else. 0:23:22.590,0:23:23.720 We wouldn't have neuron learning. 0:23:23.720,0:23:25.420 We wouldn't have all this advances in AI. 0:23:25.420,0:23:28.230 Jürgen Schmidhuber said that the biggest, 0:23:28.230,0:23:30.010 the last contribution of neuroscience to 0:23:30.010,0:23:32.220 artificial intelligence[br]was about 50 years ago. 0:23:32.220,0:23:34.280 That's depressing, and it might be 0:23:34.280,0:23:37.870 overemphasizing the unimportance of neuroscience, 0:23:37.870,0:23:39.490 because neuroscience is very important, 0:23:39.490,0:23:41.090 once you know what are you looking for. 0:23:41.090,0:23:42.510 You can actually often find this, 0:23:42.510,0:23:44.320 and see whether you are on the right track. 0:23:44.320,0:23:45.860 But it's very difficult to take neuroscience 0:23:45.860,0:23:47.940 to understand how the brain is working. 0:23:47.940,0:23:49.290 Because it's really like understanding 0:23:49.290,0:23:53.230 flight by looking at birds through a microscope. 0:23:53.230,0:23:55.150 So, what are these messages? 0:23:55.150,0:23:57.850 You are going to need messages,[br]that tell these cortical columns 0:23:57.850,0:24:00.160 to join themselves into a structure. 0:24:00.160,0:24:01.990 And to unlink again once they're done. 0:24:01.990,0:24:03.690 You need ways that they can request each other 0:24:03.690,0:24:06.040 to perform computations for them. 0:24:06.040,0:24:07.510 You need ways they can inhibit each other 0:24:07.510,0:24:08.320 when they are linked up. 0:24:08.320,0:24:10.990 So they don't do conflicting computations. 0:24:10.990,0:24:12.940 Then they need to tell you whether the computation, 0:24:12.940,0:24:14.110 the result of the computation 0:24:14.110,0:24:16.730 that the are asked to do is probably false. 0:24:16.730,0:24:19.340 Or whether it's probably true,[br]but you still need to wait for others, 0:24:19.340,0:24:21.990 to tell you whether the details worked out. 0:24:21.990,0:24:24.240 Or whether it's confirmed true that the concepts 0:24:24.240,0:24:26.730 that they stand for is actually the case. 0:24:26.730,0:24:28.150 And then you want to have learning, 0:24:28.150,0:24:29.630 to tell you how well this worked. 0:24:29.630,0:24:31.390 So you will have to announce a bounty, 0:24:31.390,0:24:34.380 that tells them to link up[br]and kind of reward signal 0:24:34.380,0:24:36.740 that makes do computation in the first place. 0:24:36.740,0:24:38.680 And then you want to have[br]some kind of reward signal 0:24:38.680,0:24:40.550 once you got the result as an organism. 0:24:40.550,0:24:42.280 But you reach your goal if you made 0:24:42.280,0:24:45.810 the disturbance go away[br]or what ever you consume the cake. 0:24:45.810,0:24:47.710 And then you will have[br]some kind of reward signal 0:24:47.710,0:24:49.250 that's you give everybody. 0:24:49.250,0:24:50.650 That was involved in this. 0:24:50.650,0:24:52.720 And this reward signal facilitates learning, 0:24:52.720,0:24:55.230 so the.. difference between the announce reward 0:24:55.230,0:24:57.530 and consumption reward is the learning signal 0:24:57.530,0:24:58.740 for these guys. 0:24:58.740,0:25:00.210 So they can learn how to play together, 0:25:00.210,0:25:02.700 and how to do the Solomonoff induction. 0:25:02.700,0:25:04.660 Now, I've told you that Solomonoff induction 0:25:04.660,0:25:05.280 is not computable. 0:25:05.280,0:25:07.630 And it's mostly because of two things, 0:25:07.630,0:25:09.280 First of all it's needs infinite resources 0:25:09.280,0:25:11.200 to compare all the possible models. 0:25:11.200,0:25:13.530 And the other one is that we do not know 0:25:13.530,0:25:15.440 the priori probability for our Bayesian model. 0:25:15.440,0:25:19.280 If we do not know[br]how likely unknown stuff is in the world. 0:25:19.280,0:25:22.520 So what we do instead is,[br]we set some kind of hyperparameter, 0:25:22.520,0:25:25.050 Some kind of default[br]priori probability for concepts, 0:25:25.050,0:25:28.110 that are encoded by cortical columns. 0:25:28.110,0:25:30.580 And if we set these parameters very low, 0:25:30.580,0:25:32.140 then we are going to end up with inferences 0:25:32.140,0:25:35.250 that are quite probable. 0:25:35.250,0:25:36.480 For unknown things. 0:25:36.480,0:25:37.690 And then we can test for those. 0:25:37.690,0:25:41.350 If we set this parameter higher, we are going[br]to be very, very creative. 0:25:41.350,0:25:43.670 But we end up with many many theories, 0:25:43.670,0:25:45.140 that are difficult to test. 0:25:45.140,0:25:48.470 Because maybe there are[br]too many theories to test. 0:25:48.470,0:25:50.650 Basically every of these cortical columns[br]will now tell you, 0:25:50.650,0:25:52.240 when you ask them if they are true: 0:25:52.240,0:25:54.960 "Yes I'm probably true,[br]but i still need to ask others, 0:25:54.960,0:25:56.980 to work on the details" 0:25:56.980,0:25:58.670 So these others are going to be get active, 0:25:58.670,0:26:00.640 and they are being asked by the asking element: 0:26:00.640,0:26:01.730 "Are you going to be true?", 0:26:01.730,0:26:04.380 and they say "Yeah, probably yes,[br]I just have to work on the details" 0:26:04.380,0:26:05.930 and they are going to ask even more. 0:26:05.930,0:26:07.980 So your brain is going to light up like a[br]christmas tree, 0:26:07.980,0:26:10.240 and do all these amazing computations, 0:26:10.240,0:26:12.450 and you see connections everywhere,[br]most of them are wrong. 0:26:12.450,0:26:16.310 You are basically in psychotic state[br]if your hyperparameter is too high. 0:26:16.310,0:26:20.790 You're brain invents more theories[br]that it can disproof. 0:26:20.790,0:26:24.550 Would it actually sometimes be good[br]to be in this state? 0:26:24.550,0:26:27.850 You bet. So i think every night our brain[br]goes in this state. 0:26:27.850,0:26:31.720 We turn up this hyperparameter.[br]We dream. We get all kinds 0:26:31.720,0:26:34.100 weird connections, and we get to see connections, 0:26:34.100,0:26:36.140 that otherwise we couldn't be seeing. 0:26:36.140,0:26:38.080 Even though... because they are highly improbable. 0:26:38.080,0:26:42.750 But sometimes they hold, and we see... "Oh[br]my God, DNA is organized in double helix". 0:26:42.750,0:26:44.640 And this is what we remember in the morning. 0:26:44.640,0:26:46.870 All the other stuff is deleted. 0:26:46.870,0:26:48.440 So we usually don't form long term memories 0:26:48.440,0:26:51.480 in dreams, if everything goes well. 0:26:51.480,0:26:56.670 If you accidentally trip this up.. your modulators, 0:26:56.670,0:26:59.100 for instance by consuming illegal substances, 0:26:59.100,0:27:01.690 or because you just gone randomly psychotic 0:27:01.690,0:27:04.600 you was basically entering[br]a dreaming state I guess. 0:27:04.600,0:27:06.990 You get to a state[br]when the brain starts inventing more 0:27:06.990,0:27:10.860 concepts that it can disproof. 0:27:10.860,0:27:13.600 So you want to have a state[br]where this is well balanced. 0:27:13.600,0:27:16.180 And the difference between[br]highly creative people, 0:27:16.180,0:27:20.070 and very religious people is probably[br]a different setting of this hyperparameter. 0:27:20.070,0:27:21.890 So I suspect that people that people[br]that are genius, 0:27:21.890,0:27:23.880 like people like Einstein and so on, 0:27:23.880,0:27:26.600 do not simply have better neurons than others. 0:27:26.600,0:27:29.130 What they mostly have is a slightly hyperparameter, 0:27:29.130,0:27:33.860 that is very finely tuned, so they can get[br]better balance than other people 0:27:33.860,0:27:43.850 in finding theories that might be true,[br]but can still be disprooven. 0:27:43.850,0:27:49.480 So inventiveness could be[br]a hyperparameter in the brain. 0:27:49.480,0:27:54.169 If you want to measure[br]the quality of belief that we have 0:27:54.169,0:27:56.370 we are going to have to have[br]some kind of some cost function 0:27:56.370,0:27:58.710 which is based on motivational system. 0:27:58.710,0:28:02.400 And to identify if belief[br]is good or not we can abstract criteria, 0:28:02.400,0:28:06.440 for instance how well does it predict the[br]wourld, or how about does it reduce uncertainty 0:28:06.440,0:28:07.590 in the world, 0:28:07.590,0:28:10.020 or is it consistency and sparse. 0:28:10.020,0:28:14.080 And then of course utility, how about does[br]it help me to satisfy my needs. 0:28:14.080,0:28:18.920 And the motivational system is going[br]to evaluate all this things by giving a signal. 0:28:18.920,0:28:24.200 And the first signal.. kind of signal[br]is the possible rewards if we are able to compute 0:28:24.200,0:28:25.020 the task. 0:28:25.020,0:28:27.430 And this is probably done by dopamine. 0:28:27.430,0:28:30.350 So we have a very small area in the brain,[br]substantia nigra, 0:28:30.350,0:28:33.610 and the ventral tegmental area,[br]and they produce dopamine. 0:28:33.610,0:28:38.180 And this get fed into lateral frontal cortext[br]and the frontal lobe, 0:28:38.180,0:28:41.920 which control attention,[br]and tell you what things to do. 0:28:41.920,0:28:46.020 And if we have successfully done[br]what you wanted to do, 0:28:46.020,0:28:49.300 we consume the rewards. 0:28:49.300,0:28:51.940 And we do this with another signal[br]which is serotonine. 0:28:51.940,0:28:53.480 It's also announce to motivational system, 0:28:53.480,0:28:55.870 to this very small are the Raphe nuclei. 0:28:55.870,0:28:58.690 And it feeds into all the areas of the brain[br]where learning is necessary. 0:28:58.690,0:29:02.160 A connection is strengthen[br]once you get to result. 0:29:02.160,0:29:07.559 These two substances are emitted[br]by the motivational system. 0:29:07.559,0:29:09.710 The motivational system is a bunch of needs, 0:29:09.710,0:29:11.510 essentially you regulate it below the cortext. 0:29:11.510,0:29:14.490 They are not part of your mental representations. 0:29:14.490,0:29:16.930 They are part of something[br]that is more primary than this. 0:29:16.930,0:29:19.360 This is what makes us go,[br]this is what makes us human. 0:29:19.360,0:29:22.290 This is not our rationality, this is what we want. 0:29:22.290,0:29:27.000 And the needs are physiological,[br]they are social, they are cognitive. 0:29:27.000,0:29:28.960 And you pretty much born with them. 0:29:28.960,0:29:30.470 They can not be totally adaptive, 0:29:30.470,0:29:33.340 because if we were adaptive,[br]we wouldn't be doing anything. 0:29:33.340,0:29:35.390 The needs are resistive. 0:29:35.390,0:29:38.290 They are pushing us against the world. 0:29:38.290,0:29:40.170 If you wouldn't have all this needs, 0:29:40.170,0:29:41.740 If you wouldn't have this motivational system, 0:29:41.740,0:29:43.630 you would just be doing what best for you. 0:29:43.630,0:29:45.150 Which means collapse on the ground, 0:29:45.150,0:29:49.010 be a vegetable, rod, give into gravity. 0:29:49.010,0:29:50.270 Instead you do all this unpleasant things, 0:29:50.270,0:29:52.690 to get up in the morning,[br]you eat, you have sex, 0:29:52.690,0:29:54.120 you do all this crazy things. 0:29:54.120,0:29:58.809 And it's only because the[br]motivational system forces you to. 0:29:58.809,0:30:00.850 The motivational system[br]takes this bunch of matter, 0:30:00.850,0:30:02.890 and makes us to do all these strange things, 0:30:02.890,0:30:05.940 just so genomes get replicated and so on. 0:30:05.940,0:30:10.470 And... so to do this, we are going to build[br]resistance against the world. 0:30:10.470,0:30:13.360 And the motivational system[br]is in a sense forcing us, 0:30:13.360,0:30:15.470 to do all this things by giving us needs, 0:30:15.470,0:30:18.330 and the need have some kind[br]of target value and current value. 0:30:18.330,0:30:21.850 If we have a differential[br]between the target value and current value, 0:30:21.850,0:30:24.590 we perceive some urgency[br]to do something about the need. 0:30:24.590,0:30:26.680 And when the target value[br]approaches the current value 0:30:26.680,0:30:28.660 we get the pleasure, which is a learning signal. 0:30:28.660,0:30:30.540 If it gets away from it[br]we get a displeasure signal, 0:30:30.540,0:30:31.870 which is also a learning signal. 0:30:31.870,0:30:35.370 And we can use this to structure[br]our understanding of the world. 0:30:35.370,0:30:36.870 To understand what goals are and so on. 0:30:36.870,0:30:40.020 Goals are learned. Needs are not. 0:30:40.020,0:30:42.780 To learn we need success[br]and failure in the world. 0:30:42.780,0:30:45.940 But to do things we need anticipated reward. 0:30:45.940,0:30:48.120 So it's dopamine that's makes brain go round. 0:30:48.120,0:30:50.560 Dopamine makes you do things. 0:30:50.560,0:30:52.750 But in order to do this in the right way, 0:30:52.750,0:30:54.610 you have to make sure,[br]that the cells can not 0:30:54.610,0:30:55.880 produce dopamine themselves. 0:30:55.880,0:30:59.100 If they do this they can start[br]to drive others to work for them. 0:30:59.100,0:31:01.870 You are going to get something like[br]bureaucracy in your neural cortext, 0:31:01.870,0:31:05.650 where different bosses try[br]to set up others to they own bidding 0:31:05.650,0:31:07.910 and pitch against other groups in nerual cortext. 0:31:07.910,0:31:09.730 It's going to be horrible. 0:31:09.730,0:31:12.210 So you want to have some kind of central authority, 0:31:12.210,0:31:16.290 that make sure that the cells[br]do not produce dopamine themselves. 0:31:16.290,0:31:19.679 It's only been produce in[br]very small area and then given out, 0:31:19.679,0:31:21.059 and pass through the system. 0:31:21.059,0:31:23.350 And after you're done with it's going to be gone, 0:31:23.350,0:31:26.070 so there is no hoarding of the dopamine. 0:31:26.070,0:31:29.770 And in our society the role of dopamine[br]is played by money. 0:31:29.770,0:31:32.150 Money is not reward in itself. 0:31:32.150,0:31:35.570 It's in some sense way[br]that you can trade against the reward. 0:31:35.570,0:31:36.850 You can not eat money. 0:31:36.850,0:31:40.500 You can take it later and take[br]a arbitrary reward for it. 0:31:40.500,0:31:45.400 And in some sense money is the dopamine[br]that makes organizations 0:31:45.400,0:31:48.410 and society, companies[br]and many individuals do things. 0:31:48.410,0:31:50.500 They do stuff because of money. 0:31:50.500,0:31:53.309 But money if you compare to dopamine[br]is pretty broken, 0:31:53.309,0:31:54.850 because you can hoard it. 0:31:54.850,0:31:57.400 So you are going to have this[br]cortical columns in the real world, 0:31:57.400,0:31:59.670 which are individual people[br]or individual corporations. 0:31:59.670,0:32:03.250 They are hoarding the dopamine,[br]they sit on this very big pile of dopamine. 0:32:03.250,0:32:07.890 They are starving the rest[br]of the society of the dopamine. 0:32:07.890,0:32:10.630 They don't give it away,[br]and they can make it do it's bidding. 0:32:10.630,0:32:13.970 So for instance they can pitch[br]substantial part of society 0:32:13.970,0:32:16.130 against understanding of global warming. 0:32:16.130,0:32:20.110 because they profit of global warming[br]or of technology that leads to global warming, 0:32:20.110,0:32:22.850 which is very bad for all of us. applause 0:32:22.850,0:32:28.850 So our society is a nervous system[br]that lies to itself. 0:32:28.850,0:32:30.429 How can we overcome this? 0:32:30.429,0:32:32.480 Actually, we don't know. 0:32:32.480,0:32:34.639 To do this we would need[br]to have some kind of centrialized, 0:32:34.639,0:32:36.660 top-down reward motivational system. 0:32:36.660,0:32:39.010 We have this for instance in the military, 0:32:39.010,0:32:42.520 you have this system of[br]military rewards that you get. 0:32:42.520,0:32:44.950 And this are completely[br]controlled from the top. 0:32:44.950,0:32:47.260 Also within working organizations[br]you have this. 0:32:47.260,0:32:49.600 In corporations you have centralized rewards, 0:32:49.600,0:32:51.850 it's not like rewards flow bottom-up, 0:32:51.850,0:32:55.120 they always flown top-down. 0:32:55.120,0:32:57.850 And there was an attempt[br]to model society in such a way. 0:32:57.850,0:33:03.380 That was in Chile in the early 1970,[br]the Allende government had the idea 0:33:03.380,0:33:07.320 to redesign society or economy[br]in society using cybernetics. 0:33:07.320,0:33:12.590 So Allende invited a bunch of cyberneticians[br]to redesign the Chilean economy. 0:33:12.590,0:33:14.550 And this was meant to be the control room, 0:33:14.550,0:33:17.460 where Allende and his chief economists[br]would be sitting, 0:33:17.460,0:33:19.709 to look at what the economy is doing. 0:33:19.709,0:33:23.880 We don't know how this would work out,[br]because we know how it ended. 0:33:23.880,0:33:27.260 In 1973 there was this big putsch in Chile, 0:33:27.260,0:33:30.290 and this experiment ended among other things. 0:33:30.290,0:33:34.170 Maybe it would have worked, who knows?[br]Nobody tried it. 0:33:34.170,0:33:38.370 So, there is something else[br]what is going on in people, 0:33:38.370,0:33:40.030 beyond the motivational system. 0:33:40.030,0:33:43.610 That is: we have social criteria, for learning. 0:33:43.610,0:33:47.670 We also check if our ideas[br]are normativly acceptable. 0:33:47.670,0:33:50.510 And this is actually a good thing,[br]because individual may shortcut 0:33:50.510,0:33:52.590 the learning through communication. 0:33:52.590,0:33:55.260 Other people have learned stuff[br]that we don't need to learn ourselves. 0:33:55.260,0:33:59.800 We can build on this, so we can accelerate[br]learning by many order of magnitutde, 0:33:59.800,0:34:00.970 which makes culture possible. 0:34:00.970,0:34:04.190 And which makes many anything possible,[br]because if you were on your own 0:34:04.190,0:34:06.860 you would not be going to find out[br]very much in your lifetime. 0:34:08.520,0:34:11.270 You know how they say?[br]Everything that you do, 0:34:11.270,0:34:14.250 you do by standing on the shoulders of giants. 0:34:14.250,0:34:17.779 Or on a big pile of dwarfs[br]it works either way. 0:34:17.779,0:34:27.089 laughterapplause 0:34:27.089,0:34:30.379 Social learning usually outperforms[br]individual learning. You can test this. 0:34:30.379,0:34:33.949 But in the case of conflict[br]between different social truths, 0:34:33.949,0:34:36.659 you need some way to decide who to believe. 0:34:36.659,0:34:39.498 So you have some kind of reputation[br]estimate for different authority, 0:34:39.498,0:34:42.399 and you use this to check whom you believe. 0:34:42.399,0:34:45.748 And the problem of course is this[br]in existing society, in real society, 0:34:45.748,0:34:48.389 this reputation system is going[br]to reflect power structure, 0:34:48.389,0:34:51.699 which may distort your belief systematically. 0:34:51.699,0:34:54.759 Social learning therefore leads groups[br]to synchronize their opinions. 0:34:54.759,0:34:57.220 And the opinions become ...get another role. 0:34:57.220,0:35:02.180 They become important part[br]of signalling which group you belong to. 0:35:02.180,0:35:06.630 So opinions start to signal[br]group loyalty in societies. 0:35:06.630,0:35:11.170 And people in this, and that's the actual world,[br]they should optimize not for getting the best possible 0:35:11.170,0:35:12.619 opinions in terms of truth. 0:35:12.619,0:35:17.289 They should guess... they should optimize[br]for doing... having the best possible opinion, 0:35:17.289,0:35:19.799 with respect to agreement with their peers. 0:35:19.799,0:35:22.029 If you have the same opinion[br]as your peers, you can signal them 0:35:22.029,0:35:24.299 that you are the part of their ingroup,[br]they are going to like you. 0:35:24.299,0:35:28.160 If you don't do this, chances are[br]they are not going to like you. 0:35:28.160,0:35:34.049 There is rarely any benefit in life to be[br]in disagreement with your boss. Right? 0:35:34.049,0:35:39.230 So, if you evolve an opinion forming system[br]in these curcumstances, 0:35:39.230,0:35:41.220 you should be ending up[br]with an opinion forming system, 0:35:41.220,0:35:42.980 that leaves you with the most usefull opinion, 0:35:42.980,0:35:45.400 which is the opinion in your environment. 0:35:45.400,0:35:48.400 And it turns out, most people are able[br]to do this effortlessly. 0:35:48.400,0:35:50.969 laughter 0:35:50.969,0:35:55.529 They have an instinct, that makes them adapt[br]the dominant opinion in their social environment. 0:35:55.529,0:35:56.599 It's amazing, right? 0:35:56.599,0:36:01.040 And if you are nerd like me,[br]you don't get this. 0:36:01.040,0:36:08.999 laugingapplause 0:36:08.999,0:36:12.999 So in the world out there,[br]explanations piggyback on you group allegiance. 0:36:12.999,0:36:15.900 For instance you will find that there is a[br]substantial group of people that believes 0:36:15.900,0:36:18.380 the minimum wage is good[br]for the economy and for you 0:36:18.380,0:36:20.549 and another one believes that its bad. 0:36:20.549,0:36:23.470 And its pretty much aligned[br]with political parties. 0:36:23.470,0:36:25.970 Its not aligned with different[br]understandings of economy, 0:36:25.970,0:36:30.740 because nobody understands[br]how the economy works. 0:36:30.740,0:36:36.330 And if you are a nerd you try to understand[br]the world in terms of what is true and false. 0:36:36.330,0:36:40.680 You try to prove everything by putting it[br]in some kind of true and false level 0:36:40.680,0:36:43.589 and if you are not a nerd[br]you try to get to right and wrong 0:36:43.589,0:36:45.609 you try to understand[br]whether you are in alignment 0:36:45.609,0:36:49.559 with what's objectively right[br]in your society, right? 0:36:49.559,0:36:55.680 So I guess that nerds are people that have[br]a defect in there opinion forming system. 0:36:55.680,0:36:57.069 laughing 0:36:57.069,0:37:00.609 And usually that's maladaptive[br]and under normal circumstances 0:37:00.609,0:37:03.099 nerds would mostly be filtered[br]from the world, 0:37:03.099,0:37:06.529 because they don't reproduce so well,[br]because people don't like them so much. 0:37:06.529,0:37:07.960 laughing 0:37:07.960,0:37:11.119 And then something very strange happened.[br]The computer revolution came along and 0:37:11.119,0:37:14.170 suddenly if you argue with the computer[br]it doesn't help you if you have the 0:37:14.170,0:37:17.849 normatively correct opinion you need to[br]be able to understand things in terms of 0:37:17.849,0:37:26.029 true and false, right? applause 0:37:26.029,0:37:29.779 So now we have this strange situation that[br]the weird people that have this offensive, 0:37:29.779,0:37:33.410 strange opinions and that really don't[br]mix well with the real normal people 0:37:33.410,0:37:38.119 get all this high paying jobs[br]and we don't understand how is that happening. 0:37:38.119,0:37:42.599 And it's because suddenly[br]our maladapting is a benefit. 0:37:42.599,0:37:47.300 But out there there is this world of the[br]social norms and it's made of paperwalls. 0:37:47.300,0:37:50.349 There are all this things that are true[br]and false in a society that make 0:37:50.349,0:37:51.549 people behave. 0:37:51.549,0:37:57.390 It's like this japanese wall, there.[br]They made palaces out of paper basically. 0:37:57.390,0:38:00.339 And these are walls by convention. 0:38:00.339,0:38:04.009 They exist because people agree[br]that this is a wall. 0:38:04.009,0:38:06.630 And if you are a hypnotist[br]like Donald Trump 0:38:06.630,0:38:11.109 you can see that these are paper walls[br]and you can shift them. 0:38:11.109,0:38:14.079 And if you are a nerd like me[br]you can not see these paperwalls. 0:38:14.079,0:38:20.230 If you pay closely attention you see that[br]people move and then suddenly middair 0:38:20.230,0:38:22.869 they make a turn. Why would they do this? 0:38:22.869,0:38:24.360 There must be something[br]that they see there 0:38:24.360,0:38:26.549 and this is basically a normative agreement. 0:38:26.549,0:38:29.690 And you can infer what this is[br]and then you can manipulate it and understand it. 0:38:29.690,0:38:32.640 Of course you can't fix this, you can[br]debug yourself in this regard, 0:38:32.640,0:38:34.690 but it's something that is hard[br]to see for nerds. 0:38:34.690,0:38:38.109 So in some sense they have a superpower:[br]they can think straight in the presence 0:38:38.109,0:38:39.079 of others. 0:38:39.079,0:38:42.590 But often they end up in their living room[br]and people are upset. 0:38:42.590,0:38:45.810 laughter 0:38:45.810,0:38:49.789 Learning in a complex domain can not[br]guarantee that you find the global maximum. 0:38:49.789,0:38:53.970 We know that we can not find truth[br]because we can not recognize whether we live 0:38:53.970,0:38:57.059 on a plain field or on a[br]simulated plain field. 0:38:57.059,0:39:00.579 But what we can do is, we can try to[br]approach a global maximum. 0:39:00.579,0:39:02.339 But we don't know if that[br]is the global maximum. 0:39:02.339,0:39:05.509 We will always move along[br]some kind of belief gradient. 0:39:05.509,0:39:09.110 We will take certain elements of[br]our belief and then give them up 0:39:09.110,0:39:12.650 for new elements of a belief based on[br]thinking, that this new element 0:39:12.650,0:39:15.049 of belief is better than the one[br]we give up. 0:39:15.049,0:39:17.079 So we always move along[br]some kind of gradient. 0:39:17.079,0:39:19.789 and the truth does not matter,[br]the gradient matters. 0:39:19.789,0:39:23.650 If you think about teaching for a moment,[br]when I started teaching I often thought: 0:39:23.650,0:39:27.489 Okay, I understand the truth of the[br]subject, the students don't, so I have to 0:39:27.489,0:39:30.069 give this to them[br]and at some point I realized: 0:39:30.069,0:39:33.450 Oh, I changed my mind so many times[br]in the past and I'm probably not going to 0:39:33.450,0:39:35.769 stop changing it in the future. 0:39:35.769,0:39:38.710 I'm always moving along a gradient[br]and I keep moving along a gradient. 0:39:38.710,0:39:43.099 So I'm not moving to truth,[br]I'm moving forward. 0:39:43.099,0:39:45.230 And when we teach our kids[br]we should probably not think about 0:39:45.230,0:39:46.390 how to give them truth. 0:39:46.390,0:39:51.039 We should think about how to put them onto[br]an interesting gradient, that makes them 0:39:51.039,0:39:55.079 explore the world,[br]world of possible beliefs. 0:39:55.079,0:40:03.150 applause 0:40:03.150,0:40:05.359 And this possible beliefs[br]lead us into local minima. 0:40:05.359,0:40:08.150 This is inevitable. This are like valleys[br]and sometimes this valleys are 0:40:08.150,0:40:11.210 neighbouring and we don't understand[br]what the people in the neighbouring 0:40:11.210,0:40:15.700 valley are doing unless we are willing to[br]retrace the steps they have been taken. 0:40:15.700,0:40:19.569 And if you want to get from one valley[br]into the next, we will have to have some kind 0:40:19.569,0:40:21.789 of energy that moves us over the hill. 0:40:21.789,0:40:27.739 We have to have a trajectory were every[br]step works by finding reason to give up 0:40:27.739,0:40:30.380 bit of our current belief and adopt a[br]new belief, because it's somehow 0:40:30.380,0:40:34.739 more useful, more relevant,[br]more consistent and so on. 0:40:34.739,0:40:38.349 Now the problem is that this is not[br]monotonous we can not guarantee that 0:40:38.349,0:40:40.499 we're always climbing,[br]because the problem is, that 0:40:40.499,0:40:44.599 the beliefs themselfs can change[br]our evaluation of the belief. 0:40:44.599,0:40:50.390 It could be for instance that you start[br]believing in a religion and this religion 0:40:50.390,0:40:54.299 could tell you: If you give up the belief[br]in the religion, you're going to face 0:40:54.299,0:40:56.500 eternal damnation in hell. 0:40:56.500,0:40:59.489 As long as you believe in the religion,[br]it's going to be very expensive for you 0:40:59.489,0:41:02.430 to give up the religion, right?[br]If you truly belief in it. 0:41:02.430,0:41:05.109 You're now caught[br]in some kind of attractor. 0:41:05.109,0:41:08.680 Before you believe the religion it is not[br]very dangerous but once you've gotten 0:41:08.680,0:41:13.019 into the attractor it's very,[br]very hard to get out. 0:41:13.019,0:41:16.309 So these belief attractors[br]are actually quite dangerous. 0:41:16.309,0:41:19.920 You can get not only to chaotic behaviour,[br]where you can not guarantee that your 0:41:19.920,0:41:23.470 current belief is better than the last one[br]but you can also get into beliefs that are 0:41:23.470,0:41:26.849 almost impossible to change. 0:41:26.849,0:41:33.739 And that makes it possible to program[br]people to work in societies. 0:41:33.739,0:41:37.529 Social domains are structured by values.[br]Basically a preference is what makes you 0:41:37.529,0:41:40.769 do things, because you anticipate[br]pleasure or displeasure, 0:41:40.769,0:41:45.339 and values make you do things[br]even if you don't anticipate any pleasure. 0:41:45.339,0:41:49.809 These are virtual rewards.[br]They make us do things, because we believe 0:41:49.809,0:41:51.799 that is stuff[br]that is more important then us. 0:41:51.799,0:41:55.109 This is what values are about. 0:41:55.109,0:42:00.690 And these values are the source[br]of what we would call true meaning, deeper meaning. 0:42:00.690,0:42:05.220 There is something that is more important[br]than us, something that we can serve. 0:42:05.220,0:42:08.769 This is what we usually perceive as[br]meaningful life, it is one which 0:42:08.769,0:42:12.759 is in the serves of values that are more[br]important than I myself, 0:42:12.759,0:42:15.749 because after all I'm not that important.[br]I'm just this machine that runs around 0:42:15.749,0:42:20.789 and tries to optimize its pleasure and[br]pain, which is kinda boring. 0:42:20.789,0:42:26.329 So my PI has puzzled me, my principle[br]investigator in the Havard department, 0:42:26.329,0:42:29.349 where I have my desk, Martin Nowak. 0:42:29.349,0:42:33.970 He said, that meaning can not exist without[br]god; you are either religious, 0:42:33.970,0:42:36.950 or you are a nihilist. 0:42:36.950,0:42:42.789 And this guy is the head of the[br]department for evolutionary dynamics. 0:42:42.789,0:42:45.769 Also he is a catholic.. chuckling 0:42:45.769,0:42:49.729 So this really puzzled me and I tried[br]to understand what he meant by this. 0:42:49.729,0:42:53.200 Typically if you are a good atheist[br]like me, 0:42:53.200,0:42:57.920 you tend to attack gods that are[br]structured like this, religious gods, 0:42:57.920,0:43:02.940 that are institutional, they are personal,[br]they are some kind of person. 0:43:02.940,0:43:08.239 They do care about you, they prescribe[br]norms, for instance don't mastrubate 0:43:08.239,0:43:10.060 it's bad for you. 0:43:10.060,0:43:14.759 Many of this norms are very much aligned[br]with societal institutions, for instance 0:43:14.759,0:43:20.799 don't questions the authorities,[br]god wants them to be ruling above you 0:43:20.799,0:43:23.839 and be monogamous and so on and so on. 0:43:23.839,0:43:28.979 So they prescribe norms that do not make[br]a lot of sense in terms of beings that 0:43:28.979,0:43:31.200 creates world every now and then, 0:43:31.200,0:43:34.619 but they make sense in terms of[br]what you should be doing to be a 0:43:34.619,0:43:36.730 functioning member of society. 0:43:36.730,0:43:40.799 And this god also does things like it[br]creates world, they like to manifest as 0:43:40.799,0:43:43.660 burning shrubbery and so on. There are[br]many books that describe stories that 0:43:43.660,0:43:45.700 these gods have allegedly done. 0:43:45.700,0:43:48.819 And it's very hard to test for all these[br]features which makes this gods very 0:43:48.819,0:43:54.280 improbable for us. And makes Atheist[br]very dissatisfied with these gods. 0:43:54.280,0:43:56.569 But then there is a different kind of god. 0:43:56.569,0:43:58.599 This is what we call the spiritual god. 0:43:58.599,0:44:02.410 This spiritual god is independent of[br]institutions, it still does care about you. 0:44:02.410,0:44:06.489 It's probably conscious. It might not be a[br]person. There are not that many stories, 0:44:06.489,0:44:10.579 that you can consistently tell about it,[br]but you might be able to connect to it 0:44:10.579,0:44:15.259 spiritually. 0:44:15.259,0:44:19.470 Then there is a god that is even less[br]expensive. That is god as a transcendental 0:44:19.470,0:44:23.489 principle and this god is simply the reason[br]why there is something rather then 0:44:23.489,0:44:28.150 nothing. This god is the question the[br]universe is the answer to, this is the 0:44:28.150,0:44:29.600 thing that gives meaning. 0:44:29.600,0:44:31.489 Everything else about it is unknowable. 0:44:31.489,0:44:34.190 This is the god of Thomas of Aquinus. 0:44:34.190,0:44:38.089 The God that Thomas of Aquinus discovered[br]is not the god of Abraham this is not the 0:44:38.089,0:44:39.180 religious god. 0:44:39.180,0:44:43.559 It's a god that is basically a principle[br]that us ... the universe into existence. 0:44:43.559,0:44:47.140 It's the one that gives[br]the universe it's purpose. 0:44:47.140,0:44:50.200 And because every other property[br]is unknowable about this, 0:44:50.200,0:44:52.010 this god is not that expensive. 0:44:52.010,0:44:55.960 Unfortunately it doesn't really work.[br]I mean Thomas of Aquinus tried to prove 0:44:55.960,0:45:00.049 god. He tried to prove an necessary god,[br]a god that has to be existing and 0:45:00.049,0:45:02.779 I think we can only prove a possible god. 0:45:02.779,0:45:05.339 So if you try to prove a necessary god,[br]this god can not exist. 0:45:05.339,0:45:11.650 Which means your god prove is going to[br]fail. You can only prove possible gods. 0:45:11.650,0:45:13.259 And then there is an even more improper god. 0:45:13.259,0:45:15.890 And that's the god of Aristotle and he said: 0:45:15.890,0:45:20.069 "If there is change in the universe,[br]something in going to have to change it." 0:45:20.069,0:45:23.640 There must be something that moves it[br]along from one state to the next. 0:45:23.640,0:45:26.289 So I would say that is the primary[br]computational transition function 0:45:26.289,0:45:35.079 of the universe.[br]laughingapplause 0:45:35.079,0:45:38.439 And Aristotle discovered it.[br]It's amazing isn't it? 0:45:38.439,0:45:41.509 We have to have this because we[br]can not be conscious in a single state. 0:45:41.509,0:45:43.279 We need to move between states[br]to be conscious. 0:45:43.279,0:45:45.979 We need to be processes. 0:45:45.979,0:45:50.859 So we can take our gods and sort them by[br]their metaphysical cost. 0:45:50.859,0:45:53.290 The 1st degree god would be the first mover. 0:45:53.290,0:45:56.069 The 2nd degree god is the god of purpose and meaning. 0:45:56.069,0:45:59.089 3rd degree god is the spiritual god.[br]And the 4th degree god is this bound to 0:45:59.089,0:46:01.229 religious institutions, right? 0:46:01.229,0:46:03.720 So if you take this statement[br]from Martin Nowak, 0:46:03.720,0:46:07.759 "You can not have meaning without god!"[br]I would say: yes! You need at least 0:46:07.759,0:46:14.990 a 2nd degree god to have meaning.[br]So objective meaning can only exist 0:46:14.990,0:46:19.119 with a 2nd degree god. chuckling 0:46:19.119,0:46:22.269 And subjective meaning can exist as a[br]function in a cognitive system of course. 0:46:22.269,0:46:24.180 We don't need objective meaning. 0:46:24.180,0:46:27.410 So we can subjectively feel that there is[br]something more important to us 0:46:27.410,0:46:30.509 and this makes us work in society and[br]makes us perceive that we have values 0:46:30.509,0:46:34.329 and so on, but we don't need to believe[br]that there is something outside of the 0:46:34.329,0:46:36.869 universe to have this. 0:46:36.869,0:46:40.650 So the 4th degree god is the one[br]that is bound to religious institutions, 0:46:40.650,0:46:45.400 it requires a belief attractor and it[br]enables complex norm prescriptions. 0:46:45.400,0:46:48.430 It my theory is right then it should be[br]much harder for nerds to believe in 0:46:48.430,0:46:52.039 a 4th degree god then for normal people. 0:46:52.039,0:46:56.489 And what this god does it allows you to[br]have state building mind viruses. 0:46:56.489,0:47:00.269 Basically religion is a mind virus. And[br]the amazing thing about these mind viruses 0:47:00.269,0:47:02.489 is that they structure behaviour[br]in large groups. 0:47:02.489,0:47:06.130 We have evolved to live in small groups[br]of a few 100 individuals, maybe somthing 0:47:06.130,0:47:07.249 like a 150. 0:47:07.249,0:47:10.059 This is roughly the level[br]to which reputation works. 0:47:10.059,0:47:15.369 We can keep track of about 150 people and[br]after this it gets much much worse. 0:47:15.369,0:47:18.290 So in this system where you have[br]reputation people feel responsible 0:47:18.290,0:47:21.349 for each other and they can[br]keep track of their doings 0:47:21.349,0:47:23.049 and society kind of sort of works. 0:47:23.049,0:47:27.789 If you want to go beyond this, you have[br]to right a software that controls people. 0:47:27.789,0:47:32.420 And religions were the first software,[br]that did this on a very large scale. 0:47:32.420,0:47:35.319 And in order to keep stable they had to be[br]designed like operating systems 0:47:35.319,0:47:36.039 in some sense. 0:47:36.039,0:47:39.930 They give people different roles[br]like insects in a hive. 0:47:39.930,0:47:44.529 And they have even as part of this roles is[br]to update this religion but it has to be 0:47:44.529,0:47:48.380 done very carefully and centrally[br]because otherwise the religion will split apart 0:47:48.380,0:47:51.719 and fall together into new religions[br]or be overcome by new ones. 0:47:51.719,0:47:54.259 So there is some kind of[br]evolutionary dynamics that goes on 0:47:54.259,0:47:55.930 with respect to religion. 0:47:55.930,0:47:58.519 And if you look the religions,[br]there is actually a veritable evolution 0:47:58.519,0:47:59.739 of religions. 0:47:59.739,0:48:04.789 So we have this Israelic tradition and[br]the Mesoputanic mythology that gave rise 0:48:04.789,0:48:13.019 to Judaism. applause 0:48:13.019,0:48:16.299 It's kind of cool, right? laughing 0:48:16.299,0:48:36.289 Also history totally repeats itself.[br]roaring laughterapplause 0:48:36.289,0:48:41.889 Yeah, it totally blew my mind when[br]I discovered this. laughter 0:48:41.889,0:48:45.039 Of course the real tree of programming[br]languages is slightly more complicated, 0:48:45.039,0:48:48.599 And the real tree of religion is slightly[br]more complicated. 0:48:48.599,0:48:51.229 But still its neat. 0:48:51.229,0:48:54.289 So if you want to immunize yourself[br]against mind viruses, 0:48:54.289,0:48:58.570 first of all you want to check yourself[br]whether you are infected. 0:48:58.570,0:49:02.809 You should check: Can I let go of my[br]current beliefs without feeling that 0:49:02.809,0:49:07.670 meaning departures me and I feel very[br]terrible, when I let go of my beliefs. 0:49:07.670,0:49:11.279 Also you should check: All the other[br]people around there that don't 0:49:11.279,0:49:17.019 share my belief, are they either stupid,[br]or crazy, or evil? 0:49:17.019,0:49:19.890 If you think this chances are you are[br]infected by some kind of mind virus, 0:49:19.890,0:49:23.710 because they are just part[br]of the out group. 0:49:23.710,0:49:28.059 And does your god have properties that[br]you know but you did not observe. 0:49:28.059,0:49:32.490 So basically you have a god[br]of 2nd or 3rd degree or higher. 0:49:32.490,0:49:34.589 In this case you also probably got a mind virus. 0:49:34.589,0:49:37.259 There is nothing wrong[br]with having a mind virus, 0:49:37.259,0:49:39.920 but if you want to immunize yourself[br]against this people have invented 0:49:39.920,0:49:44.059 rationalism and enlightenment,[br]basically to act as immunization against 0:49:44.059,0:49:50.660 mind viruses.[br]loud applause 0:49:50.660,0:49:53.869 And in some sense its what the mind does[br]by itself because, if you want to 0:49:53.869,0:49:56.949 understand how you go wrong,[br]you need to have a mechanism 0:49:56.949,0:49:58.839 that discovers who you are. 0:49:58.839,0:50:03.109 Some kind of auto debugging mechanism,[br]that makes the mind aware of itself. 0:50:03.109,0:50:04.779 And this is actually the self. 0:50:04.779,0:50:08.339 So according to Robert Kegan:[br]"The development of ourself is a process, 0:50:08.339,0:50:13.400 in which we learn who we are by making[br]thing explicit", by making processes that 0:50:13.400,0:50:17.249 are automatic visible to us and by[br]conceptualize them so we no longer 0:50:17.249,0:50:18.859 identify with them. 0:50:18.859,0:50:22.019 And it starts out with understanding[br]that there is only pleasure and pain. 0:50:22.019,0:50:25.180 If you are a baby, you have only[br]pleasure and pain you identify with this. 0:50:25.180,0:50:27.869 And then you turn into a toddler and the[br]toddler understands that they are not 0:50:27.869,0:50:31.059 their pleasure and pain[br]but they are their impulses. 0:50:31.059,0:50:34.259 And in the next level if you grow beyond[br]the toddler age you actually know that 0:50:34.259,0:50:38.880 you have goals and that your needs and[br]impulses are there to serve goals, but its 0:50:38.880,0:50:40.210 very difficult to let go of the goals, 0:50:40.210,0:50:42.789 if you are a very young child. 0:50:42.789,0:50:46.329 And at some point you realize: Oh, the[br]goals don't really matter, because 0:50:46.329,0:50:49.509 sometimes you can not reach them, but[br]we have preferences, we have thing that we 0:50:49.509,0:50:52.950 want to happen and thing that we do not[br]want to happen. And then at some point 0:50:52.950,0:50:55.869 we realize that other people have[br]preferences, too. 0:50:55.869,0:50:58.979 And then we start to model the world[br]as a system where different people have 0:50:58.979,0:51:01.940 different preferences and we have[br]to navigate this landscape. 0:51:01.940,0:51:06.420 And then we realize that this preferences[br]also relate to values and we start 0:51:06.420,0:51:09.700 to identify with this values as members of[br]society. 0:51:09.700,0:51:13.469 And this is basically the stage if you[br]are an adult being, that you get into. 0:51:13.469,0:51:16.910 And you can get to a stage beyond that,[br]especially if you have people this, which 0:51:16.910,0:51:20.059 have already done this. And this means[br]that you understand that people have 0:51:20.059,0:51:23.660 different values and what they do[br]naturally flows out of them. 0:51:23.660,0:51:26.849 And this values are not necessarily worse[br]than yours they are just different. 0:51:26.849,0:51:29.450 And you learn that you can hold different[br]sets of values in your mind at 0:51:29.450,0:51:33.019 the same time, isn't that amazing?[br]and understand other people, even if 0:51:33.019,0:51:36.660 they are not part of your group.[br]If you get that, this is really good. 0:51:36.660,0:51:39.269 But I don't think it stops there. 0:51:39.269,0:51:43.019 You can also learn that the stuff that[br]you perceive is kind of incidental, 0:51:43.019,0:51:45.339 that you can turn it of and you can[br]manipulate it. 0:51:45.339,0:51:49.940 And at some point you also can realize[br]that yourself is only incidental that you 0:51:49.940,0:51:52.559 can manipulate it or turn it of.[br]And that your basically some kind of 0:51:52.559,0:51:57.420 consciousness that happens to run a brain[br]of some kind of person, that navigates 0:51:57.420,0:52:04.279 the world in terms to get rewards or avoid[br]displeasure and serve values and so on, 0:52:04.279,0:52:05.130 but it doesn't really matter. 0:52:05.130,0:52:08.119 There is just this consciousness which[br]understands the world. 0:52:08.119,0:52:11.009 And this is the stage that we typically[br]call enlightenment. 0:52:11.009,0:52:14.549 In this stage you realize that you are not[br]your brain, but you are a story that 0:52:14.549,0:52:25.640 your brain tells itself.[br]applause 0:52:25.640,0:52:29.630 So becoming self aware is a process of[br]reverse engineering your mind. 0:52:29.630,0:52:32.890 Its a different set of stages in which[br]to realize what goes on. 0:52:32.890,0:52:33.799 So isn't that amazing. 0:52:33.799,0:52:38.930 AI is a way to get to more self awareness? 0:52:38.930,0:52:41.319 I think that is a good point to stop here. 0:52:41.319,0:52:44.499 The first talk that I gave in this series[br]was 2 years ago. It was about 0:52:44.499,0:52:45.979 how to build a mind. 0:52:45.979,0:52:49.670 Last year I talked about how to get from[br]basic computation to consciousness. 0:52:49.670,0:52:53.709 And this year we have talked about[br]finding meaning using AI. 0:52:53.709,0:52:57.470 I wonder where it goes next.[br]laughter 0:52:57.470,0:53:22.769 applause 0:53:22.769,0:53:26.489 Herald: Thank you for this amazing talk![br]We now have some minutes for Q&A. 0:53:26.489,0:53:31.190 So please line up at the microphones as[br]always. If you are unable to stand up 0:53:31.190,0:53:36.430 for some reason please very very visibly[br]rise your hand, we should be able to dispatch 0:53:36.430,0:53:40.099 an audio angle to your location[br]so you can have a question too. 0:53:40.099,0:53:44.030 And also if you are locationally[br]disabled, you are not actually in the room 0:53:44.030,0:53:49.069 if you are on the stream, you can use IRC[br]or twitter to also ask questions. 0:53:49.069,0:53:50.989 We also have a person for that. 0:53:50.989,0:53:53.779 We will start at microphone number 2. 0:53:53.779,0:53:59.940 Q: Wow that's me. Just a guess! What[br]would you guess, when can you discuss 0:53:59.940,0:54:04.559 your talk with a machine,[br]in how many years? 0:54:04.559,0:54:07.400 Joscha: I don't know! As a software[br]engineer I know if I don't have the 0:54:07.400,0:54:12.619 specification all bets are off, until I[br]have the implementation. laughter 0:54:12.619,0:54:14.509 So it can be of any order of magnitude. 0:54:14.509,0:54:18.249 I have a gut feeling but I also know as a[br]software engineer that my gut feeling is 0:54:18.249,0:54:23.450 usually wrong, laughter[br]until I have the specification. 0:54:23.450,0:54:28.200 So the question is if there are silver[br]bullets? Right now there are some things 0:54:28.200,0:54:30.569 that are not solved yet and it could be[br]that they are easier to solve 0:54:30.569,0:54:33.469 than we think, but it could be that[br]they're harder to solve than we think. 0:54:33.469,0:54:36.710 Before I stumbled on this cortical[br]self organization thing, 0:54:36.710,0:54:40.719 I thought it's going to be something like[br]maybe 60, 80 years and now I think it's 0:54:40.719,0:54:47.289 way less, but again this is a very[br]subjective perspective. I don't know. 0:54:47.289,0:54:49.240 Herald: Number 1, please! 0:54:49.240,0:54:55.589 Q: Yes, I wanted to ask a little bit about[br]metacognition. It seems that you kind of 0:54:55.589,0:55:01.329 end your story saying that it's still[br]reflecting on input that you get and 0:55:01.329,0:55:04.900 kind of working with your social norms[br]and this and that, but Colberg 0:55:04.900,0:55:11.839 for instance talks about what he calls a[br]postconventional universal morality 0:55:11.839,0:55:17.420 for instance, which is thinking about[br]moral laws without context, basically 0:55:17.420,0:55:23.069 stating that there is something beyond the[br]relative norm that we have to each other, 0:55:23.069,0:55:29.579 which would only be possible if you can do[br]kind of, you know, meta cognition, 0:55:29.579,0:55:32.599 thinking about your own thinking[br]and then modifying that thinking. 0:55:32.599,0:55:37.229 So kind of feeding back your own ideas[br]into your own mind and coming up with 0:55:37.229,0:55:43.779 stuff that actually can't get ...[br]well processing external inputs. 0:55:43.779,0:55:48.469 Joscha: Mhm! I think it's very tricky.[br]This project of defining morality without 0:55:48.469,0:55:53.119 societies exists longer than Kant of[br]course. And Kant tried to give this 0:55:53.119,0:55:56.869 internal rules and others tried to.[br]I find this very difficult. 0:55:56.869,0:56:01.069 From my perspective we are just moving[br]bits of rocks. And this bits of rocks they 0:56:01.069,0:56:07.589 are on some kind of dust mode in a galaxy[br]out of trillions of galaxies and how can 0:56:07.589,0:56:08.609 there be meaning? 0:56:08.609,0:56:11.180 It's very hard for me to say: 0:56:11.180,0:56:13.969 One chimpanzee species is better than[br]another chimpanzee species or 0:56:13.969,0:56:16.559 a particular monkey[br]is better than another monkey. 0:56:16.559,0:56:18.539 This only happens[br]within a certain framework 0:56:18.539,0:56:20.160 and we have to set this framework. 0:56:20.160,0:56:23.700 And I don't think that we can define this[br]framework outside of a context of 0:56:23.700,0:56:26.420 social norms, that we have to agree on. 0:56:26.420,0:56:29.650 So objectively I'm not sure[br]if we can get to ethics. 0:56:29.650,0:56:33.769 I only think that is possible based on[br]some kind of framework that people 0:56:33.769,0:56:38.339 have to agree on implicitly or explicitly. 0:56:38.339,0:56:40.630 Herald: Microphone number 4, please. 0:56:40.630,0:56:46.559 Q: Hi, thank you, it was a fascinating talk.[br]I have 2 thought that went through my mind. 0:56:46.559,0:56:51.589 And the first one is that it's so[br]convincing the models that you present, 0:56:51.589,0:56:56.709 but it's kind of like you present[br]another metaphor of understanding the 0:56:56.709,0:57:01.670 brain which is still something that we try[br]to grasp on different levels of science 0:57:01.670,0:57:07.469 basically. And the 2nd one is that your[br]definition of the nerd who walks 0:57:07.469,0:57:10.950 and doesn't see the walls is kind of[br]definition... or reminds me 0:57:10.950,0:57:15.229 Richard Rortys definition of the ironist[br]which is a person who knows that their 0:57:15.229,0:57:20.799 vocabulary is finite and that other people[br]have also a finite vocabulary and 0:57:20.799,0:57:24.599 then that obviously opens up the whole question[br]of meaning making which has been 0:57:24.599,0:57:28.979 discussed in so many[br]other disciplines and fields. 0:57:28.979,0:57:32.930 And I thought about Darridas[br]deconstruction of ideas and thoughts and 0:57:32.930,0:57:36.300 Butler and then down the rabbit hole to[br]Nietzsche and I was just wondering, 0:57:36.300,0:57:39.009 if you could maybe[br]map out other connections 0:57:39.009,0:57:44.430 where basically not AI helping us to[br]understand the mind, but where 0:57:44.430,0:57:49.819 already existing huge, huge fields of[br]science, like cognitive process 0:57:49.819,0:57:53.359 coming from the other end could help us[br]to understand AI. 0:57:53.359,0:57:59.680 Joscha: Thank you, the tradition that you[br]mentioned Rorty and Butler and so on 0:57:59.680,0:58:02.989 are part of a completely different belief[br]attractor in my current perspective. 0:58:02.989,0:58:06.209 That is they are mostly[br]social constructionists. 0:58:06.209,0:58:10.880 They believe that reality at least in the[br]domains of the mind and sociality 0:58:10.880,0:58:15.359 are social constructs they are part[br]of social agreement. 0:58:15.359,0:58:17.190 Personally I don't think that[br]this is the case. 0:58:17.190,0:58:19.630 I think that patterns that we refer to 0:58:19.630,0:58:23.890 are mostly independent of your mind.[br]The norms are part of social constructs, 0:58:23.890,0:58:28.099 but for instance our motivational[br]preferences that make us adapt or 0:58:28.099,0:58:32.719 reject norms, are something that builds up[br]resistance to the environment. 0:58:32.719,0:58:35.660 So they are probably not part[br]of social agreement. 0:58:35.660,0:58:41.569 And the only thing I can invite you to is[br]try to retrace both of the different 0:58:41.569,0:58:45.640 belief attractors, try to retrace the[br]different paths on the landscape. 0:58:45.640,0:58:48.529 All this thing that I tell you, all of[br]this is of course very speculative. 0:58:48.529,0:58:52.390 These are that seem to be logical[br]to me at this point in my life. 0:58:52.390,0:58:55.400 And I try to give you the arguments[br]why I think that is plausible, but don't 0:58:55.400,0:58:59.109 believe in them, question them, challenge[br]them, see if they work for you! 0:58:59.109,0:59:00.559 I'm not giving you any truth. 0:59:00.559,0:59:05.720 I'm just going to give you suitable encodings[br]according to my current perspective. 0:59:05.720,0:59:11.739 Q:Thank you![br]applause 0:59:11.739,0:59:15.099 Herald: The internet, please! 0:59:19.179,0:59:26.029 Signal angel: So, someone is asking[br]if in this belief space you're talking about 0:59:26.029,0:59:30.109 how is it possible[br]to get out of local minima?[br] 0:59:30.109,0:59:33.959 And very related question as well: 0:59:33.959,0:59:38.530 Should we teach some momentum method[br]to our children, 0:59:38.530,0:59:41.599 so we don't get stuck in a local minima. 0:59:41.599,0:59:44.829 Joscha: I believe at some level it's not[br]possible to get out of a local minima. 0:59:44.829,0:59:50.329 In an absolute sense, because you only get[br]to get into some kind of meta minimum, 0:59:50.329,0:59:56.769 but what you can do is to retrace the[br]path that you took whenever you discover 0:59:56.769,0:59:59.989 that somebody else has a fundamentally[br]different set of beliefs. 0:59:59.989,1:00:02.769 And if you realize that this person is[br]basically a smart person that is not 1:00:02.769,1:00:07.359 completely insane but has reasons to[br]believe in their beliefs and they seem to 1:00:07.359,1:00:10.579 be internally consistent it's usually[br]worth to retrace what they 1:00:10.579,1:00:12.180 have been thinking and why. 1:00:12.180,1:00:15.930 And this means you have to understand[br]where their starting point was and 1:00:15.930,1:00:18.279 how they moved from their current point[br]to their starting point. 1:00:18.279,1:00:22.219 You might not be able to do this[br]accurately and the important thing is 1:00:22.219,1:00:25.369 also afterwards you discover a second[br]valley, you haven't discovered 1:00:25.369,1:00:27.059 the landscape inbetween. 1:00:27.059,1:00:30.839 But the only way that we can get an idea[br]of the lay of the land is that we try to 1:00:30.839,1:00:33.200 retrace as many paths as possible. 1:00:33.200,1:00:36.339 And if we try to teach our children, what[br]I think what we should be doing is: 1:00:36.339,1:00:38.650 To tell them how to explore[br]this world on there own. 1:00:38.650,1:00:43.900 It's not that we tell them this is the[br]valley, basically it's given, it's 1:00:43.900,1:00:47.599 the truth, but instead we have to tell[br]them: This is the path that we took. 1:00:47.599,1:00:51.239 And these are the things that we saw[br]inbetween and it is important to be not 1:00:51.239,1:00:54.390 completely naive when we go into this[br]landscape, but we also have to understand 1:00:54.390,1:00:58.170 that it's always an exploration that[br]never stops and that might change 1:00:58.170,1:01:01.140 everything that you believe now[br]at a later point. 1:01:01.140,1:01:05.700 So for me it's about teaching my own[br]children how to be explorers, 1:01:05.700,1:01:10.950 how to understand that knowledge is always[br]changing and it's always a moving frontier. 1:01:10.950,1:01:17.230 applause 1:01:17.230,1:01:22.259 Herald: We are unfortunately out of time.[br]So, please once again thank Joscha! 1:01:22.259,1:01:24.069 applause[br]Joscha: Thank you! 1:01:24.069,1:01:28.239 applause 1:01:28.239,1:01:32.719 postroll music 1:01:32.719,1:01:40.000 subtitles created by c3subtitles.de[br]Join, and help us!