WEBVTT 00:00:00.017 --> 00:00:05.257 Ricardo, one of the speakers at our summer school, and Ricardo, 00:00:05.497 --> 00:00:13.537 you presented a view on machines, if you want, and robots, 00:00:13.777 --> 00:00:19.157 control systems, in which you made the argument that actually they should naturally 00:00:19.157 --> 00:00:24.097 include phenomena you could call consciousness. 00:00:25.397 --> 00:00:29.157 So, what are your considerations there to come to this rather, 00:00:29.277 --> 00:00:30.777 let's say, counterintuitive conclusion? 00:00:31.397 --> 00:00:35.017 Well, I think it is not a counterintuitive conclusion. 00:00:35.417 --> 00:00:40.777 I think it's just a need of how things are developing and all the needs of having 00:00:40.777 --> 00:00:46.997 more robust systems today where we can rely our lives upon in this changing 00:00:46.997 --> 00:00:48.857 environment where things are changing so fast. 00:00:48.857 --> 00:00:53.497 So, the question for the new machines and the machines of the future is that 00:00:53.497 --> 00:00:56.257 they must be much more adaptive in a sense. 00:00:56.537 --> 00:01:01.277 And not just adaptive to changes in the environment, but adaptive to their own changes. 00:01:01.617 --> 00:01:03.757 And that takes us to this question of consciousness. 00:01:04.657 --> 00:01:07.057 It's a question of name, more than a question of fashion. 00:01:08.321 --> 00:01:16.121 Thank you. Yeah, but in some sense, you make an argument based on the sense of complexity, right? 00:01:16.181 --> 00:01:20.201 Where you would say, well, if we want to start to deal with the world, 00:01:20.261 --> 00:01:26.441 the dynamical world, in a way that is acceptable, the controllers will get more complex. 00:01:26.641 --> 00:01:30.321 And at some point, the controller will find it necessary to control itself. 00:01:30.661 --> 00:01:34.041 And it's sort of somewhere at that point that you're saying consciousness comes 00:01:34.041 --> 00:01:37.721 in. So how should I understand that? Where's that transition point exactly? 00:01:38.321 --> 00:01:44.221 Well, the question is the moment where humans cannot really control or address 00:01:44.221 --> 00:01:45.821 the complexity of the systems. 00:01:45.941 --> 00:01:50.121 And the only possibility for that is to have a scalable approach. 00:01:50.421 --> 00:01:54.241 And having a scalable approach, humans don't scale well. That's a problem because 00:01:54.241 --> 00:01:56.061 humans typically think one by one. 00:01:56.141 --> 00:02:00.181 And thinking by teams is not yet solved for humans. 00:02:00.541 --> 00:02:05.621 But for machines, it could be. And the question is to transfer the competences 00:02:05.621 --> 00:02:11.081 for managing systems, managing complex systems into systems because they can scale up. 00:02:11.321 --> 00:02:16.781 And that's the question at the end. Because growing complexity can only be handled 00:02:16.781 --> 00:02:18.601 by systems that can grow. 00:02:18.861 --> 00:02:22.341 And humans cannot grow as they are today. And that's the problem. 00:02:22.881 --> 00:02:28.501 When we are in the case of technical systems, we are touching the limits of understandability. 00:02:28.501 --> 00:02:34.721 So in many cases failures cannot be understood by humans because of the complexity 00:02:34.721 --> 00:02:39.741 and problems that are appearing for example in countrywide electrical networks 00:02:39.741 --> 00:02:43.221 cannot be understood by humans and that's the problem, 00:02:43.321 --> 00:02:48.041 we need to transfer into systems that scale more than humans and that are able 00:02:48.041 --> 00:02:52.781 to handle these problems that seems counterintuitive, no? 00:02:52.881 --> 00:02:57.021 Because in some sense you're saying we want to have scalable systems to come 00:02:57.021 --> 00:03:02.981 to a scalable system they must be sentient they should have consciousness humans 00:03:02.981 --> 00:03:06.941 are not scalable, they don't scale very well however. 00:03:08.312 --> 00:03:11.632 You didn't say that, but that's what I was thinking. Humans are conscious. Yeah. 00:03:12.232 --> 00:03:16.692 So that would suggest that consciousness is not a necessary ingredient of these 00:03:16.692 --> 00:03:19.432 scalable systems. Or did I misunderstand the argument? 00:03:19.852 --> 00:03:24.812 No, the question, we are real. The question of consciousness and self-awareness, 00:03:24.932 --> 00:03:27.192 not trying to mimic humans at all. 00:03:27.332 --> 00:03:33.252 We reached a point finding some basic design principles or structure principles. 00:03:33.772 --> 00:03:38.352 And then we found this community dealing with self-awareness and consciousness 00:03:38.352 --> 00:03:40.112 in discussing these topics. 00:03:40.292 --> 00:03:45.212 But we came not from the side of mimicking humans, but from the side of finding 00:03:45.212 --> 00:03:47.992 some structural competences into 00:03:47.992 --> 00:03:51.852 the systems themselves, related to self-awareness and self-competence. 00:03:52.012 --> 00:03:55.532 And then we found that the people doing research on machine consciousness, 00:03:55.792 --> 00:03:59.412 in many of the cases, they are just trying to mimic human behavior. 00:03:59.592 --> 00:04:05.092 Without this understanding of the basic principles regarding self-awareness 00:04:05.092 --> 00:04:09.332 and self-management at the very end and deeply. 00:04:09.692 --> 00:04:15.912 So we came from the purely technical side into the topics where those people are also discussing. 00:04:16.192 --> 00:04:22.372 So in a sense, we are not really trying to build machines that mimic human consciousness at all. 00:04:22.452 --> 00:04:27.012 We are trying to build machines that have the competences that are associated 00:04:27.012 --> 00:04:31.272 to this self-awareness in humans, but perhaps not the same way. 00:04:32.192 --> 00:04:39.132 Okay, but explain something to me that I can, I mean, in some sense, 00:04:39.212 --> 00:04:41.472 you're coming from a control engineering perspective, right? 00:04:41.532 --> 00:04:44.052 And you say, okay, in control engineering, we're hitting a boundary now. 00:04:44.692 --> 00:04:50.772 But then what's this traditional control engineering view that we now have to change? 00:04:51.092 --> 00:04:53.712 What's this traditional view and why does it break down exactly? 00:04:54.172 --> 00:04:57.992 Well, the breakdown is that typically control engineers model, 00:04:58.092 --> 00:05:00.152 have a model of the plant they are controlling. 00:05:00.152 --> 00:05:03.472 But they don't have a model of the controller because 00:05:03.472 --> 00:05:06.292 the complexity of the controller is so high that they really cannot have 00:05:06.292 --> 00:05:09.292 a mathematical model the way they had in the past so you 00:05:09.292 --> 00:05:12.932 can have a differential equation model of a car and then you build a controller 00:05:12.932 --> 00:05:17.692 for it but then what happens when the controller is failing how can you control 00:05:17.692 --> 00:05:21.812 the controller and this is missing and that's the problem where traditional 00:05:21.812 --> 00:05:26.212 control engineering is not addressing at all because for example in other cars, 00:05:26.312 --> 00:05:27.712 you have plenty of networking inside, 00:05:27.992 --> 00:05:30.312 plenty of electronic control units and computing. 00:05:30.552 --> 00:05:33.552 If something fails there, there is no control over that. 00:05:33.972 --> 00:05:40.852 So if the failure is in the control system, current control technology is not 00:05:40.852 --> 00:05:42.812 addressing it. And that's the problem. 00:05:43.092 --> 00:05:46.592 Because they are only addressing faults at the level of the plant, 00:05:46.772 --> 00:05:48.632 but not faults at the level of the controller. 00:05:48.972 --> 00:05:54.172 So we need controllers that are able to control themselves to overcome these difficulties. 00:05:54.312 --> 00:05:59.312 Because of the growing complexity of the control systems. In that case. 00:06:00.660 --> 00:06:05.620 So that's clear to me. Following this argument, I could then say, 00:06:05.680 --> 00:06:11.140 well, but how do I now prevent falling into an infinite regress of these controllers 00:06:11.140 --> 00:06:14.340 and meta controllers and supervisory controllers and so on? 00:06:14.500 --> 00:06:17.580 Yeah, the question is something that was mentioned during this course. 00:06:17.980 --> 00:06:28.560 We bound the infinite regress if in every step the needs for representation are reduced. 00:06:28.560 --> 00:06:34.780 So, in a sense, the system is collapsing into a single compact and simplified 00:06:34.780 --> 00:06:36.000 representation of itself. 00:06:36.440 --> 00:06:39.860 So, that's the question. The question is not just to have a whole representation 00:06:39.860 --> 00:06:42.220 of the complete system, but a 00:06:42.220 --> 00:06:47.880 compact, more synthetic and more simple representation at each meta level. 00:06:48.180 --> 00:06:51.620 So at the end, the system will collapse into a single representation that is 00:06:51.620 --> 00:06:57.640 addressing, so to say, in a pyramidal way, the understanding of the lower layers. 00:06:58.700 --> 00:07:04.760 So is that a hope, a hypothesis, or is it a reality? No, it's just a hypothesis. Okay. 00:07:05.640 --> 00:07:11.640 It does sound a little bit like the old intentional systems stance of Dan Dennett, 00:07:11.720 --> 00:07:15.320 where he would say like, well, we can resolve the homunculus problem because 00:07:15.320 --> 00:07:18.900 we can just split it up in simpler, simpler, simpler homunculi. 00:07:18.900 --> 00:07:22.640 And in the end, it's just a binary decision that has to be made to its lowest level. 00:07:22.860 --> 00:07:30.680 Yeah. Yeah, but this is, in some sense, reducing away the problem and not really solving it, right? 00:07:30.740 --> 00:07:34.200 You explain it away because, actually, then you have to show that this is possible, 00:07:34.360 --> 00:07:38.720 that you can actually, if you use the word simplicity for this, 00:07:38.740 --> 00:07:42.860 that you can get to this kind of compression of complexity. Yeah. 00:07:44.420 --> 00:07:48.460 Could you give an example of this kind of compression? How do you think about 00:07:48.460 --> 00:07:49.960 this, even for a simple controller? 00:07:50.460 --> 00:07:54.640 For a simple controller, you can find this kind of approach in very complex 00:07:54.640 --> 00:08:03.360 plants where you can have thousands of control loops controlling basic properties 00:08:03.360 --> 00:08:05.660 in plants. And then you have loops over loops. 00:08:05.840 --> 00:08:09.560 And when you go up through the pyramid of control, at the very end, 00:08:09.600 --> 00:08:10.820 what you have is just money. 00:08:11.320 --> 00:08:15.300 And you have a single loop controlling the money, the money that the plant is producing. 00:08:15.500 --> 00:08:23.240 And then beyond that, the safety and the maintainability and environmental impact. 00:08:23.480 --> 00:08:28.600 And then you have a unit control and then you have, well, this is a scale down 00:08:28.600 --> 00:08:30.560 and you can find this inside the plants. 00:08:30.640 --> 00:08:34.740 But the problem is that in that system, this kind of hierarchy is not self-aware. 00:08:34.980 --> 00:08:39.540 So it's not addressing the problems it has upon itself and humans are there 00:08:39.540 --> 00:08:42.920 to handle And it's the same problems that you can find in information technology, 00:08:42.980 --> 00:08:46.500 in networks and computing systems and infrastructures. 00:08:46.740 --> 00:08:51.240 It's the same problem that the computers controlling things are not controlling 00:08:51.240 --> 00:08:52.800 themselves. That's the big problem. 00:08:53.240 --> 00:08:55.880 But this is an interesting example, definitely. 00:08:57.240 --> 00:09:00.800 Also, as you say, right, it's always the humans enter the loop somewhere. 00:09:01.100 --> 00:09:06.660 And maybe it's on the basis of that luxury that we have these general purpose 00:09:06.660 --> 00:09:09.480 controllers running around, which are human beings. 00:09:10.000 --> 00:09:12.840 That then these engineered systems can actually operate. 00:09:13.320 --> 00:09:16.820 And this might give us then the false belief that the principles in which they 00:09:16.820 --> 00:09:18.760 operate are valid principles. 00:09:19.100 --> 00:09:22.040 Do you think that's a reasonable criticism? Yeah, it is. 00:09:23.860 --> 00:09:27.920 The question about humans is an old question in automation. 00:09:28.420 --> 00:09:32.380 Humans should be there or shouldn't be there. And the question is that today, 00:09:32.480 --> 00:09:35.960 in many of the systems, humans cannot cope with the problem. 00:09:36.620 --> 00:09:40.880 And that's the problem we are facing these days. Humans are not able to solve 00:09:40.880 --> 00:09:44.060 problems at the level of electrical networks, countrywide electrical networks. 00:09:44.280 --> 00:09:48.360 They are not able to solve problems at the level of flight control systems when 00:09:48.360 --> 00:09:53.260 faults appear. Because of the requirements of speed, of information integration, 00:09:53.720 --> 00:09:58.500 of focus of attention, humans cannot solve the problems. 00:09:58.860 --> 00:10:03.920 And then what happens? When those faults appear, systems do crash. 00:10:04.420 --> 00:10:08.800 And there is no solution today for this kind of problem. Then we have blackouts, 00:10:08.900 --> 00:10:11.860 we have very big airplane accidents. 00:10:12.200 --> 00:10:16.720 Because humans cannot cope, and we don't have solutions for handling those problems 00:10:16.720 --> 00:10:17.620 at the level that we need. 00:10:18.500 --> 00:10:21.780 Following your earlier argument, I could then say, well, humans cannot cope 00:10:21.780 --> 00:10:26.480 because the engineered system has created a situation in which it cannot cope 00:10:26.480 --> 00:10:28.660 and subsequently the human user cannot cope either. 00:10:29.060 --> 00:10:33.560 Yeah. Is that fair? It's fair. Okay. And that counts because we are all the 00:10:33.560 --> 00:10:36.680 time pushing forward the requirements for the technical systems. 00:10:36.800 --> 00:10:41.740 And now we're in the frontier of pushing forward the requirements of technical 00:10:41.740 --> 00:10:44.080 systems beyond human control competence. 00:10:44.820 --> 00:10:49.860 And we are in such a frontier today. There are phenomena that are appearing 00:10:49.860 --> 00:10:51.800 in technical systems that humans don't understand. 00:10:52.420 --> 00:10:56.580 Take an example of that? An example, for example, there are waves of electrical 00:10:56.580 --> 00:11:02.900 voltage crossing across Europe, north and south, and people don't understand what's going on. 00:11:04.460 --> 00:11:08.360 Because there is no way of understanding the whole picture at all. 00:11:08.580 --> 00:11:12.720 We need an information theory and a model of what's going on that goes beyond 00:11:12.720 --> 00:11:14.460 the capabilities of humans. 00:11:14.560 --> 00:11:21.240 That happens also with technical systems in information and infrastructures. 00:11:21.520 --> 00:11:27.880 The understanding of how a complex server, a computing server that is composed 00:11:27.880 --> 00:11:29.640 by thousands of computers at 00:11:29.640 --> 00:11:33.740 the same time, how it behaves when something fails, when there is a virus. 00:11:34.080 --> 00:11:38.340 Humans cannot cope with that. The only solution is to shut down the systems 00:11:38.340 --> 00:11:42.360 and start from scratch because nobody is understanding what's going on and there 00:11:42.360 --> 00:11:43.940 is no possibility of doing that. 00:11:44.140 --> 00:11:50.900 But the only strategy that we can think of is making systems keep themselves working. 00:11:51.880 --> 00:11:56.220 However, the way you phrase the problem now, which definitely is a serious problem, 00:11:56.320 --> 00:12:01.700 seems to create almost a paradox because certainly where I'm coming from, 00:12:01.820 --> 00:12:04.440 there's this belief like, okay, if we take a biomimetic approach, 00:12:04.640 --> 00:12:09.500 an approach more based on understanding of the brain, we can solve these kinds of problems. 00:12:09.760 --> 00:12:12.600 But in some sense, what you're saying for all practical purposes, 00:12:12.740 --> 00:12:15.300 also the human brain is not able to solve these problems today. 00:12:15.740 --> 00:12:18.860 So my belief is actually based on a false assumption. 00:12:19.400 --> 00:12:23.240 Is that correct? Yeah, in a sense, it is correct. But the question is that we 00:12:23.240 --> 00:12:27.300 are proposing is not bio-inspired things. What we are saying is that the things 00:12:27.300 --> 00:12:29.920 that humans do are an instance of this kind of approach. 00:12:30.740 --> 00:12:35.920 That is not enough. I mean, there is a kind of solutions that are, 00:12:36.100 --> 00:12:37.920 so to say, the self-awareness solutions. 00:12:38.200 --> 00:12:41.360 And humans are just an instance of that with their own limitations. 00:12:41.860 --> 00:12:46.280 And making a copy of humans for solving those problems is not the solution. 00:12:46.500 --> 00:12:52.020 The solution is going beyond particular human implementations and particular 00:12:52.020 --> 00:12:55.140 human brains and understanding the basic principles that are there. 00:12:55.320 --> 00:12:59.440 And that's the problem. We need to understand the very principles of self-awareness 00:12:59.440 --> 00:13:05.940 to be able to create a new class of system that has much more competence regarding speed and. 00:13:07.458 --> 00:13:10.378 Capability of integrating information that humans do have 00:13:10.378 --> 00:13:13.398 but to do that we can get by inspiration 00:13:13.398 --> 00:13:16.378 but in the sense of understanding how humans do and extracting 00:13:16.378 --> 00:13:19.818 the basic principles of that not copying the brain but understanding how it 00:13:19.818 --> 00:13:24.138 works and use that kind of architecture to scale it up that's very exciting 00:13:24.138 --> 00:13:29.458 so you're saying actually the human brain is is like an imperfect approximation 00:13:29.458 --> 00:13:34.018 of some ultimate brain i would say that we might that we might be able to figure 00:13:34.018 --> 00:13:36.298 out i would say say that. That's a good word for it. 00:13:36.618 --> 00:13:42.418 Okay, so how do I know that I found that perfect brain? 00:13:43.058 --> 00:13:47.078 No, the question is, can we find those patterns in the brain? 00:13:47.198 --> 00:13:51.998 Can we understand how humans think about themselves and get some basic principles about that? 00:13:52.158 --> 00:13:55.338 Can we have a good theory of how humans think about humans? 00:13:55.818 --> 00:13:59.898 And that's a big problem. Not just for extracting the basic principles, 00:14:00.018 --> 00:14:03.238 that's a big problem also for interacting systems with humans. 00:14:03.498 --> 00:14:07.098 The main problem, to my understanding about humans interacting with computers 00:14:07.098 --> 00:14:09.658 is that computers don't have a good theory of the human. 00:14:09.998 --> 00:14:12.998 It's not a problem of having good theories about computers. 00:14:13.278 --> 00:14:18.658 It's the question of how humans are working and how is the perceptual systems of humans. 00:14:18.758 --> 00:14:24.478 The question for me is, can we understand humans in a sense of understanding 00:14:24.478 --> 00:14:29.358 how they work to make systems that are not human at all, but are exploiting 00:14:29.358 --> 00:14:33.378 some basic competences that humans are able to do? 00:14:34.278 --> 00:14:38.378 For example, a basic competence is the cognitive flexibility that humans do have. 00:14:39.198 --> 00:14:44.078 Humans can learn how things are working. Can we translate that into a machine? 00:14:44.578 --> 00:14:49.478 Not by building a human, but by building into the machine the capability of 00:14:49.478 --> 00:14:51.318 understanding how things are working. 00:14:51.778 --> 00:14:54.358 This is a basic competence that we need for technical systems. 00:14:54.598 --> 00:14:56.158 And that does not mean making a human. 00:14:56.398 --> 00:15:00.798 That's a big mistake. People try to think that to make intelligent machines, 00:15:00.998 --> 00:15:04.518 we should copy humans. I don't think that's the problem. 00:15:04.978 --> 00:15:10.738 Would you say then that the human brain is compromised because of, 00:15:10.818 --> 00:15:13.638 let's say, the way it's embodied? 00:15:14.058 --> 00:15:19.498 Of the evolutionary constraints imposed? Or why is our brain imperfect in that sense? 00:15:19.818 --> 00:15:25.438 Well, the brains are imperfect because evolutionary change does not create optimal 00:15:25.438 --> 00:15:27.158 solutions, just good enough solutions. 00:15:27.638 --> 00:15:30.278 Solutions that are good enough for the ecological needs. They are working. 00:15:30.538 --> 00:15:37.818 So humans' brains are good for controlling mammals in the context of 60,000 00:15:37.818 --> 00:15:40.238 years ago or so. They are not optimal solutions. 00:15:40.538 --> 00:15:44.198 They are scaled up to the level of solving the problems at that time. 00:15:44.658 --> 00:15:49.718 And that means that if we want to translate that into a different body with 00:15:49.718 --> 00:15:54.498 a different set of requirements, we cannot really copy it unless we are doing a humanoid robot. 00:15:55.485 --> 00:15:58.325 Would have the same kind of problems and the same kind of interaction of humans. 00:15:58.625 --> 00:16:03.445 If the context, if the ecological niche for the machine is very different, 00:16:04.145 --> 00:16:05.665 copying the brain will not work. 00:16:05.805 --> 00:16:08.285 We need something very different for that. 00:16:08.445 --> 00:16:11.525 So there are plenty of constraints in the human brain that are related with 00:16:11.525 --> 00:16:17.005 environments of the humans, with the body of the humans, with the evolutionary history of humans. 00:16:17.525 --> 00:16:21.965 And we need to identify them to get rid of the constraints and the limitations 00:16:21.965 --> 00:16:24.625 and go to the basic principles indeed. Right. 00:16:25.125 --> 00:16:30.225 Now, there's some fear mongering going on certainly across the Atlantic Ocean 00:16:30.225 --> 00:16:34.905 about what's called the singularity. In case we're going to build this perfect 00:16:34.905 --> 00:16:37.365 machine, it might decide it doesn't need humans anymore. 00:16:37.425 --> 00:16:40.525 We get in some sort of Skynet Terminator scenario. 00:16:41.685 --> 00:16:46.525 Do you share these concerns? Not really. Not really for a single reason. 00:16:46.645 --> 00:16:54.905 I don't think that we are able to create the level of self-awareness and selfiness, 00:16:54.945 --> 00:16:57.365 I would say, into machines that humans do have. 00:16:57.865 --> 00:17:02.725 We can build some basic properties of humans into the machines, but not create…. 00:17:04.569 --> 00:17:08.909 Today, these complex entities. I'm not afraid of that. 00:17:09.209 --> 00:17:15.329 I'm pretty sure that when we reach the level of being able to create this kind of entity, 00:17:15.829 --> 00:17:21.969 the technology will be so sophisticated that we can put any limitation that 00:17:21.969 --> 00:17:25.629 we want in the sense of Asimov's laws of robotics. 00:17:26.109 --> 00:17:31.429 We will be able to bound the behavior. That's a big problem because today the 00:17:31.429 --> 00:17:37.949 problem against building complex systems is how to bound the wrong behaviors at the end. 00:17:38.069 --> 00:17:42.929 And that's a very big problem, but I think that technologies today are very aware of that. 00:17:43.689 --> 00:17:49.389 That before you press the start button of a machine, you need to really bound 00:17:49.389 --> 00:17:53.109 it, bound the behavior, and know how they are bound. Right, absolutely. 00:17:53.629 --> 00:17:58.229 So now, coming from this perspective of control engineering, 00:17:59.229 --> 00:18:03.309 and also as we discussed in the summer school, there is often this idea like, 00:18:03.309 --> 00:18:04.889 Like, well, the brain is a control system. 00:18:05.049 --> 00:18:12.289 But now, as we discussed, the control systems that we know as engineers are incomplete. 00:18:13.309 --> 00:18:19.889 So how far does it get us to just make that claim the brain is a control system? 00:18:20.049 --> 00:18:23.669 I mean, is it a bit too easy to say? 00:18:23.789 --> 00:18:28.589 At what point of definition does it actually give us insight? 00:18:28.589 --> 00:18:35.289 The understanding of why I say that the brain is a control system is very simple. 00:18:35.909 --> 00:18:41.949 Most of the inputs to the brain and the outputs of the brain are just feeding 00:18:41.949 --> 00:18:46.409 the brain and getting some products, waste products on it. 00:18:47.069 --> 00:18:50.669 And all the meaningful inputs are information. 00:18:51.629 --> 00:18:55.649 And that means that what is happening in the brain is an informational process. 00:18:56.509 --> 00:19:02.269 It's not a chemical process producing any kind of molecule that the body is using. 00:19:03.069 --> 00:19:06.609 The molecules that appear there are molecules related to information. 00:19:08.251 --> 00:19:11.051 So, what is happening in the brain is information coming in, 00:19:11.051 --> 00:19:12.511 information coming out. That is 00:19:12.511 --> 00:19:15.211 the function of the brain, and that is the function of the control system. 00:19:15.951 --> 00:19:21.151 So when systems that do that are information processors, and when information 00:19:21.151 --> 00:19:25.291 processors link sensors and actuators, they are controllers, 00:19:25.591 --> 00:19:28.151 they are no other thing. So it's not a question of... 00:19:28.811 --> 00:19:34.531 We can say that it's a question of observing the thing. So we get cables coming 00:19:34.531 --> 00:19:39.471 from sensors into informational processor and then going out into actuators. 00:19:39.671 --> 00:19:42.671 That's a control system as we build it today. 00:19:43.251 --> 00:19:48.151 But that's in some sense a very broad framework in which we define control. 00:19:48.631 --> 00:19:53.651 So can we be more specific? I mean, do you say like, well, we have engineered 00:19:53.651 --> 00:19:59.391 specific kinds of control systems that we think are particularly relevant when 00:19:59.391 --> 00:20:01.991 we look at the brain or when we look at a certain area in the brain? 00:20:02.371 --> 00:20:07.231 Is that... No, in fact, this is the whole thing. This is all materials from the 50s and 60s. 00:20:07.271 --> 00:20:11.511 This is the whole cybernetics movement in the... The problem that happened with 00:20:11.511 --> 00:20:15.111 cybernetics is that it was split in different fields. 00:20:15.671 --> 00:20:19.631 One was information technology. The other one was control technology that is 00:20:19.631 --> 00:20:22.451 related to building technical systems at the end. 00:20:22.791 --> 00:20:27.551 Some people kept it into the world of networks and brain understanding and biology. 00:20:27.911 --> 00:20:32.991 But the field is the field of the 50s and 60s. So this is an old picture. 00:20:33.411 --> 00:20:39.571 This is not something new. And in fact, if we read what people like McCulloch 00:20:39.571 --> 00:20:44.211 and Wiener said, they were saying that controllers in technical systems and 00:20:44.211 --> 00:20:45.751 brains are doing the same thing. 00:20:45.751 --> 00:20:50.491 And the analysis they did in the past are the analysis that we need to foster today. 00:20:50.851 --> 00:20:55.771 And now we have something that they didn't have, that these huge amounts of competing. 00:20:56.971 --> 00:21:00.351 Fast enough as to control bodies of enormous complexity. 00:21:00.691 --> 00:21:05.951 So we are now in a situation that we really can go back into this cybernetic 00:21:05.951 --> 00:21:12.111 vision and really understand how brains work as controllers. I mean, this picture. 00:21:12.891 --> 00:21:17.071 So in the domain of, let's say, the neuroscience of motor control, 00:21:17.751 --> 00:21:22.751 there's quite some excitement nowadays on particular kinds of concepts coming 00:21:22.751 --> 00:21:26.931 from control engineering and people believe, okay, this is now explaining what 00:21:26.931 --> 00:21:32.731 the brain is doing, what structure X is doing in terms of forward modeling or inverse modeling, etc. 00:21:33.011 --> 00:21:34.511 What do you think of these approaches? 00:21:34.711 --> 00:21:37.571 Is this actually helpful or do you think it's sort of confusing? 00:21:37.851 --> 00:21:42.071 No, it is very helpful. The problem that we have is that when we look at control 00:21:42.071 --> 00:21:47.231 technology and control theory, the problem is that control theory, as we, 00:21:47.751 --> 00:21:54.151 I would say, control engineers do, do is based on solid mathematical models 00:21:54.151 --> 00:21:56.571 and sets of equations that we can solve. 00:21:57.111 --> 00:22:00.051 And we can solve analytically, and that's the question. 00:22:00.251 --> 00:22:04.191 When we have models that are so complex that we really cannot solve analytically, 00:22:04.431 --> 00:22:08.451 then that's the field for other people trying to do simulations and trying to 00:22:08.451 --> 00:22:12.931 do computer implementations and trying to do, well, environments for testing, 00:22:13.131 --> 00:22:14.971 forward testing of models, in a sense. 00:22:15.151 --> 00:22:19.031 And this is the field of people that are modeling brains, because the mathematical 00:22:19.031 --> 00:22:20.711 description of a brain is so 00:22:20.711 --> 00:22:24.491 complex that we really cannot solve the equations by analytical methods. 00:22:25.131 --> 00:22:29.711 So the communities, in a sense, are split into two ways of doing things. 00:22:29.951 --> 00:22:34.331 One way is the mathematical-analytical method, and the other way is the forward 00:22:34.331 --> 00:22:40.011 model-based synthetic method of using computers for looking at what's going on. 00:22:40.191 --> 00:22:44.231 Because really, we cannot invert the model, and that's the point. 00:22:44.231 --> 00:22:49.411 When complexity grows, when non-linearity appears, we really cannot solve the 00:22:49.411 --> 00:22:51.691 equations and invert the models to make the controllers. 00:22:51.891 --> 00:22:56.211 Then we revert into this alternative way of exploring. 00:22:57.467 --> 00:23:00.167 And by simulation what's going on and trying to understand that. 00:23:00.347 --> 00:23:04.247 Because really we cannot analytically solve the problems. 00:23:04.447 --> 00:23:08.847 But the community should get together. Again, there are problems and theoretical 00:23:08.847 --> 00:23:11.427 concepts that are critical to be shared. 00:23:11.727 --> 00:23:16.127 The concept of observability, for example, the concept of controllability. 00:23:16.727 --> 00:23:23.007 Those are concepts that should be a common trait across all these technologies. 00:23:23.547 --> 00:23:27.887 So how would you see a concept like controllability giving us leverage when we're look at the brain? 00:23:28.547 --> 00:23:33.087 Well, the question of controllability is a question of how can we do the body 00:23:33.087 --> 00:23:37.387 do whatever we want and what things we can do with this. 00:23:37.687 --> 00:23:40.607 And the question is, can we solve analytically that problem? 00:23:40.727 --> 00:23:45.587 Can we decide if a particular body, for example, of a humanoid robot is controllable? 00:23:45.787 --> 00:23:47.287 And that is a theoretical problem 00:23:47.287 --> 00:23:50.887 that can be solved. But many people in robotics is not trying to do that. 00:23:51.007 --> 00:23:55.267 It's trying just to, for example, people that are trying to do flexible They 00:23:55.267 --> 00:23:59.287 are trying to solve a problem by just experimentation and hacking, 00:23:59.527 --> 00:24:05.007 I would say, instead of having the deep models that they need to address the problem. 00:24:05.287 --> 00:24:09.307 But the problem here is that we have isolated communities with different ontologies, 00:24:09.547 --> 00:24:13.427 some more rigorous and more mathematical, more physical, the other one more 00:24:13.427 --> 00:24:17.327 related on empirical aspects and, so to say, philosophical even, 00:24:17.407 --> 00:24:19.367 even metaphysical assumptions. 00:24:19.367 --> 00:24:24.467 Assumptions, and we need to get rid of the separation and have a single community 00:24:24.467 --> 00:24:29.067 across this, from the very simple controllers that we can find anywhere, 00:24:29.327 --> 00:24:32.387 to the brain, and to the psychological, 00:24:32.787 --> 00:24:36.887 more complex and more sophisticated psychological aspects, and have a single. 00:24:37.867 --> 00:24:40.107 Ontology that can be shared across all this stuff. 00:24:40.387 --> 00:24:42.727 Just take, for example, the concept of information. 00:24:43.487 --> 00:24:46.447 Nobody agrees on that. And when we talk to different people, 00:24:46.547 --> 00:24:49.647 they don't reach an agreement on that. And that's a problem. 00:24:49.787 --> 00:24:56.687 The problem, I think, the big problem that we have is the community being split on those areas. 00:24:57.687 --> 00:25:02.427 Now, to bring them together, if we, for instance, go back to this notion of 00:25:02.427 --> 00:25:06.167 the brain as a controller, or more specifically, as was discussed in the school, 00:25:06.347 --> 00:25:08.007 the brain is an adaptive filter. 00:25:08.567 --> 00:25:11.727 And if you have this very broad definition of a controller, like, 00:25:11.747 --> 00:25:14.627 okay, whenever I'm mapping some input states to output states, whatever 00:25:14.627 --> 00:25:17.687 happens in between i consider a controller in some 00:25:17.687 --> 00:25:21.787 sense then it's not very a very specific formulation right because in that sense 00:25:21.787 --> 00:25:25.227 and indeed anything that is dealing with an input output mapping would fall 00:25:25.227 --> 00:25:29.987 in that category and automatically also the brain so for adaptive filter it 00:25:29.987 --> 00:25:32.747 would be sort of roughly the same thing because i have an input output mapping 00:25:32.747 --> 00:25:37.667 that i'm just sort of adjusting to some to some criteria so is it. 00:25:38.571 --> 00:25:42.731 So at what point does it actually give us leverage when we look at this? 00:25:42.831 --> 00:25:47.411 When it really adds understanding and at what point is it sort of covering a 00:25:47.411 --> 00:25:51.531 little bit actually the important details of the questions we should ask? 00:25:51.711 --> 00:25:55.331 Yeah, this is related with what I was saying before about the ontology. 00:25:55.491 --> 00:25:58.831 When people say an adaptive filter, when people say a filter, 00:25:58.991 --> 00:26:03.391 when people say a controller, some people will say that they are different things. 00:26:03.511 --> 00:26:07.071 But if you go at the basics of the thing, they are the same kind of thing. 00:26:07.071 --> 00:26:10.231 You were saying input in, output out, and that's all. 00:26:10.551 --> 00:26:13.691 So you have a system that has inputs and has outputs, and that's all. 00:26:13.931 --> 00:26:16.851 If you connect that system a particular way, it will be a controller. 00:26:17.151 --> 00:26:20.091 If you connect that system in another way, it will be a filter. 00:26:20.391 --> 00:26:24.691 That will depend on how you use the inputs and outputs of the system itself. 00:26:25.271 --> 00:26:28.271 And that's a very big problem that is related to something that has appeared 00:26:28.271 --> 00:26:30.151 in the course, that is the idea of function. 00:26:30.711 --> 00:26:36.011 What is the function of such a module? It is a controller or is a or is a filter. 00:26:36.591 --> 00:26:39.791 And there are two interpretations of this function thing. 00:26:40.011 --> 00:26:43.931 One is the relation input-output that should be described by the mathematics, 00:26:44.511 --> 00:26:47.271 describing the thing. And the other one is what it is for. 00:26:48.011 --> 00:26:52.051 What is the function? It is playing in the system and it can be a controller 00:26:52.051 --> 00:26:53.471 or can be a filter, can be whatever. 00:26:54.211 --> 00:26:59.551 The same kind of mathematical description input-output can be applied for different 00:26:59.551 --> 00:27:01.411 functions in different systems. 00:27:01.671 --> 00:27:08.391 And the clarification of the The terminology is one of the major tasks that we need to solve here. 00:27:08.991 --> 00:27:14.771 Because we can have an endless discussion about is this a filter or is it a controller? 00:27:15.111 --> 00:27:19.591 Or from a theoretical point of view, this discussion is no sense. 00:27:19.871 --> 00:27:24.791 Because those two kinds of systems are the same kind of thing at the end. 00:27:25.031 --> 00:27:27.011 The only difference is how you use them. 00:27:27.411 --> 00:27:29.371 That will depend on the use, not on the thing. 00:27:30.151 --> 00:27:35.131 The thing is not a filter, it's not an intruder, is a dynamical system. That's all. 00:27:36.519 --> 00:27:41.039 And many of the discussions we can find in current designs are related to not 00:27:41.039 --> 00:27:46.199 understanding those basic, so to say, theoretical underpinnings of the system. 00:27:46.959 --> 00:27:50.719 But now you also have been in projects where you have been really trying to 00:27:50.719 --> 00:27:54.539 link your understanding of control engineering to, let's say, the biology. 00:27:54.899 --> 00:28:00.199 Yeah. So what are the lessons from that experience? How successful have these attempts been? 00:28:00.659 --> 00:28:05.539 Well, I'm not very happy and I'm very happy. So how can that be? be. 00:28:05.639 --> 00:28:11.239 I'm not very happy because it's difficult to break the frontiers between communities. 00:28:12.139 --> 00:28:19.519 So having biologists talk with roboticists by providing a common language to them is not working. 00:28:19.979 --> 00:28:24.959 It's not working because the languages of the communities are being kept as they were in the past. 00:28:25.219 --> 00:28:28.799 It's not easy to move or to share concepts at all. 00:28:29.219 --> 00:28:32.279 So it's a happy situation. 00:28:32.659 --> 00:28:38.339 On the other side, it is good to to see that people is hearing and then trying 00:28:38.339 --> 00:28:40.679 to collaborate and hearing the others. 00:28:41.019 --> 00:28:45.779 So I think that the first thing to do is to get this common language at the 00:28:45.779 --> 00:28:46.839 end, and then common ontology. 00:28:46.919 --> 00:28:49.879 That is, to my understanding, the basic ontology of physics, 00:28:50.079 --> 00:28:52.179 and then on that, going up. 00:28:52.699 --> 00:28:56.399 And I think something will help is that, for example, in neuroscience, 00:28:56.979 --> 00:29:00.479 most of people doing neuroscience today are physicists by training. 00:29:01.119 --> 00:29:10.519 So they start to talk about the mathematical basic concepts or theoretical concepts that are there. 00:29:10.799 --> 00:29:14.219 So I hope that things will change in the future. But I think that the main problem 00:29:14.219 --> 00:29:15.219 is that of the communities. 00:29:15.499 --> 00:29:19.799 In my experience, the problem is the breaking of the communities because the 00:29:19.799 --> 00:29:24.759 problem is so big that nobody has the possibility of addressing the solution. 00:29:25.539 --> 00:29:29.279 Because someone has the tools, the other one has the global understanding, 00:29:29.599 --> 00:29:33.999 another one has the data, the ground truth, the experimental conditions, 00:29:34.319 --> 00:29:39.179 and we need to put all there into a single picture, otherwise it will solve the problem. 00:29:39.599 --> 00:29:42.599 To work our way towards the finish line. 00:29:44.339 --> 00:29:49.919 You have been going around in these fields for quite a bit and you also have 00:29:49.919 --> 00:29:54.119 this very unique experience of interfacing two of these other disciplines and 00:29:54.119 --> 00:29:56.239 really to try to solve problems very concretely. 00:29:56.279 --> 00:30:01.579 So if there is this one law, the one Ricardo Sanz law that we should all adhere 00:30:01.579 --> 00:30:06.419 to and try to understand the brain and how to then build this ideal brain that 00:30:06.419 --> 00:30:10.199 can be the ultimate controller, what's Ricardo Sanz law that we should adhere to? 00:30:13.356 --> 00:30:17.116 The most important thing that we should do is to try to be rigorous and formal 00:30:17.116 --> 00:30:18.976 about the terms that we are using. 00:30:19.716 --> 00:30:24.376 If we are able to do that, then we will see that the concepts that we are using 00:30:24.376 --> 00:30:25.896 in different fields are the same. 00:30:26.336 --> 00:30:33.536 And that will help transition from a pre-scientific domain, that is what we 00:30:33.536 --> 00:30:38.076 are now, into a newborn scientific community. 00:30:38.076 --> 00:30:43.096 So I would say that the first step to do is to be rigorous about the terms we 00:30:43.096 --> 00:30:51.356 are using that is the very first step very good five years from now I'm gonna go up to Madrid and. 00:30:53.156 --> 00:30:57.956 I'm gonna ask you look Ricardo there was this prediction you gave me and now 00:30:57.956 --> 00:31:00.976 I want to know whether it came out or not what's this one prediction you're 00:31:00.976 --> 00:31:07.616 willing to stick your neck out for today well I hope that in some years we will have a, 00:31:08.076 --> 00:31:10.416 real implementation of the fundamental concept. 00:31:11.416 --> 00:31:17.416 Working in a wide scale of different systems, from simple machines and simple 00:31:17.416 --> 00:31:21.996 robots into all networks and real large industrial plants. 00:31:22.636 --> 00:31:28.456 Having a single set of concepts implemented in such a variety of systems will 00:31:28.456 --> 00:31:31.816 give credit to the vision, in a sense. 00:31:32.776 --> 00:31:35.956 So what's the most advanced concept I can hold you to five years from now? 00:31:36.456 --> 00:31:40.536 The most advanced concept, I think, is a system that is able to control itself 00:31:40.536 --> 00:31:44.096 based on the model that engineers used to build it. 00:31:44.616 --> 00:31:51.636 Okay, cool. And that's our key cornerstone for the work. 00:31:51.796 --> 00:31:54.276 Great. I'll be back and I'll ask you about it. Ricardo Sanz, 00:31:54.356 --> 00:31:56.376 thank you very much for this conversation. Thank you.