 Good afternoon. I'm Carol Christ, I'm the Chancellor at UC Berkeley. It's my privilege and pleasure to welcome you to the second of this year's 110th annual Martin Meyerson Faculty Research Lectures. For more than a century, Berkeley's Academic Senate has singled out distinguished members of our faculty whose research has changed the trajectory of their disciplines and of our understanding. These lectures shine a light on an essential part of our mission, creating new knowledge. The curiosity and creativity that fuel the quest to learn and understand are at the heart of our commitment to making the world a better place through what we discover, what we teach, and the public service we provide. This year's lectures represent the continuation of a treasured tradition that has recurred annually with one exception. In the wake of World War One and the influenza pandemic, the faculty research lectures were suspended in 1919. In 2020, when virtual events were in vogue and Zoom kept us together, the lectures went on. Being selected to deliver a faculty research lecture is rightfully seen as a high honor. To stand out among peers who exemplify academic excellence is no small thing. For students, members of the campus community and the public, this is a wonderful way to experience scholarly research of the highest caliber. Join me in welcoming the past recipients who are with us today. Professors, please stand when I read your name and let's hold our applause till everyone is recognized. And if I don't have your name but you came in a little bit after my little list was made, please stand also. John Clark, Marvin Cohen, Bill Dietrich, Catherine Gallagher, Martin Jay, Victoria Khan, Thomas LeCurre, Anthony Long, David Rolay, Barbara Romanowicz, Nancy Sheper Hughes, and our first lecturer this year, Francesca Rockberg. The two individuals chosen by our academic senate to give this year's faculty research lectures are Francesca Rockberg, who spoke in February, and Jitendra Malik, who's presenting this afternoon. Today's lecture is also one of several taking place this spring that demonstrate the depth and impact of Berkeley's faculty work on the broad and evolving topic of artificial intelligence, a field of study and practice that's changing the world before our very eyes. For a complete schedule of these related talks and a few stories about our work in AI, you can visit Berkeley.edu. I should also note that today's lecture is being live streamed and will be available on YouTube from the faculty research lectures website and from the Berkeley AI website. Now I'm pleased to introduce today's speaker. Jitendra Malik is the Arthur J. Chick Professor in the Department of Electrical Engineering and Computer Sciences. He's also a faculty member in bioengineering and is a member of the Berkeley AI Research Cognitive Science and Vision Science Groups. Jitendra's research group has worked on many different topics in computer vision, human visual perception, robotics, machine learning, and artificial intelligence. Several well-known concepts and algorithms arose in this research such as anisotropic diffusion, normalized cuts, high dynamic range imaging, shape contexts, and RCNN. Over his 37 years at UC Berkeley, he's mentored more than 70 PhD students and postdoctoral fellows. Jitendra's honors include the 2013 Distinguished Researcher in Computer Vision Award, the 2014 KS Food Prize from the International Association of Pattern Recognition, the 2016 Allen Newell Award, the 2018 Award for Research Excellence in AI, and the 2019 Computer Society Computer Pioneer Award. He's a member of the National Academy of Engineering, the National Academy of Sciences, as well as a fellow of the American Academy of Arts and Sciences. Please join me in welcoming Professor Jitendra Malak, who will speak to us about the Sensory Motor Road to Artificial Intelligence. Thank you, Chancellor, for this very generous introduction and thank you to all my colleagues who have made time to come here. It's my pleasure to talk on this very, very hot topic today, but I'm going to talk about natural intelligence first, because we can't talk about artificial intelligence without knowing something about the natural variety. So we could talk about intelligence as having started about 550 million years ago in the Cambrian era when we had our first multicellular animals that could move about. So these were the first animals that could move and that gave them an advantage because they could find food in different places. But if you want to move and find food in different places, you need to perceive, you need to know where to go to, which means that you need to have some kind of a vision system or a perception system. And that's why we have this slogan, which is from Gibson. We see in order to move and we move in order to see. And that sets off an evolutionary arms race because once some animals are chasing you, you are a prey, then you try to move faster, you try to develop camouflage. So the predator has to become extra efficient in vision or move faster and so on and so forth. And this is through most of the history of evolution, those have been the most important components of the brain of an animal, this ability to move and this ability to perceive. If we come closer to the modern era, so let's say the hominid line, when we have separated off from our other primate, sort of hominid, fellow primates, in the last five million years, you have this evolution of bipedalism, which frees the hand for toolmaking and tool use, the opposable thumb. And there's this interesting quote from Anoxagaris, a Greek philosopher, which is, it is because of his being armed with hand that man is the most intelligent animal. It was really the development of the brain followed the development of the hand. And I use that as a metaphor for manipulation, toolmaking and all the rest. And then when we come to the most recent era, the last 50,000 years or so, is when we have modern humans coming out of Africa and language, abstract thinking, symbolic behavior, all that we now call the common person in the street thinks of as intelligence. These are, if you want to think of the last 24 hours as the history of intelligence, then in the last three minutes, the last three minutes of that is essentially all this language, symbolic behavior, which we are so proud to call amongst us as a sign of intelligence. So now let's turn to artificial intelligence. And I cannot not mention the large language models, chat GPT, GPT 3, 4, 5 and 17 to come in the future. And these are remarkable. And we are going to have lectures in this series, I think on April 19, we have one of the key creators of chat GPT, who happens to be a Berkeley CSPHD, John Schulman, who's going to talk. So what I want to say is, this result is a table of results on various standard tests. So for example, the uniform bar exam, the current version of GPT does it the 90% and 11. Okay. There are results for SAT math, GREs and so on and so forth. I'm sure our chancellor will be pleased to note that English language and composition still is at level two. So there is hope for our colleagues in literature and composition yet. Just a few things. Often right now we have of course that kind of technology in computing that has enabled us to train these really large models, the kind of data on the web, like a trillion tokens which were used. But there are ideas which go back to the 1950s, the great Shannon, Claude Shannon, for people who are from electrical engineering and applied math backgrounds, and this linguist, Firth, who said, you shall know a word by the company it keeps. And that's the core idea of how we train this. We delete a word and we try to predict it from its model. Now, of course, Firth got forgotten and Chomsky became the ruler of the roost. And Chomsky was against, was a nativist. But in my view, Chomsky was wrong. And now we have proved that Chomsky is wrong. I don't want to pick up a fight with my linguist friends just yet, maybe at the end. We see in these systems emergent syntax and linguistic competence, they are like an associative memory of the web. Anyway, this is great success, okay? And contrast it with a great failure. Not yet a failure, but certainly not yet delivered success. We don't have self-driving cars. I mean, this is the thing that everybody in the street was expecting. And this is an old idea. I mean, this guy, Dick Mance, he had cars driving on the Autobahn in the 1980s. And then 90s and in fact, we had work at Berkeley in the 90s with self-driving cars. And of course, there's a lot of hype and the great Elon Musk has told us in, I think, 2019, he said, by the middle of next year, we will have over a million Tesla cars on the road with full self-driving hardware. Okay, whether that was 2019, then 2020, and we are 2022, and we still don't have a million cars with self-driving cars on the road. Okay, so now this is something that like a high school kid of age 16, we give them 20 hours of training and we think they can learn to drive. And there is the bar exam, which we think is the result of years of training. And we are at the 90% level in AI. So there seems to be something fundamentally wrong here or puzzling here. And I can make it worse. Here's a table I found of kitchen verbs, you know, various activities like stirring and slicing onions with a knife and mixing things and so on. These are things that a 12 year old can do. Okay, and no robot today can. Okay. And this is, I think, a very important thing for people, everybody to realize that in artificial intelligence, we suffer from what's known as Morovaix Paradox. Morovaix Paradox and Morovaix quoted it, said it in the 1980s, that it was sort of known earlier. It is comparatively easy to make computers exhibit adult level performance on intelligence tests or playing checkers and difficult or impossible to give them the skills of a one year old when it comes to perception and mobility. And Steve Pinker, slightly later, said it very pithily. The main lesson of 35 years of AI research is that the hard problems are easy, and that the easy problems are hard. Okay, what everybody in the person in the street thinks of as easy is actually hard. What a child has managed to accomplish by age two. What? However, we think of as hard. The result of years of education is actually not so hard. Okay, or we have made progress on it. I won't go on to Pinker's quote, where he sort of makes our sessions about which jobs are in danger, but the good news I'll state is he's gonna he claims that the gardeners receptionist and cooks are secure in their jobs for decades to come. Okay, so that's good. But anyway, my job is to try to challenge even their jobs, right? And this comes down to that's why the title of my talk is sensory motor intelligence, which is all that stuff which an evolutionary time was early, right, perception movement, things like that. How do we make progress on those problems? Where is obviously these these later problems of advanced culture are not so hard. So there's a question as to why and this is debatable. I think Hans Morrowick gave a reason which I don't think is fully right. But let me state his intuition, which is that we'll have more difficulty in engineering to reverse engineer skills that are the result of hundreds of millions of years of evolution. So perception action was early in the process. And that's hard. We don't have anything which has the intelligence of a cat today. I think a better argument, at least from today, when we most of our advanced systems are based on machine learning, is that we lack the kind of digitized data on the web for training machine learning models. So this like the the the the text on the web, right? Every book has been digitized. Wikipedia exists. There are all these blogs. So as a result, there is there are a trillion words on the web. That's the kind of knowledge which the systems like chat, GPT and GPT four can exploit. This is not typically knowledge that a child acquires by the by age of five, you start reading books when you after you've gone to school. The challenge is the experiences that a child has had before age five. And those experiences are very personal and embodied. And they do not exist on the web, at least not yet. Okay. And and and I think that's part of the of the challenge. It's not that if they existed on the web, we would know how to exploit them. But certainly they don't exist on the web. So in in the rest of this talk, I'm going to talk a little bit about our attempts on that challenge, the challenge of trying to build a two year old, not not sort of somebody who can pass the law exam at 90% time. So and this is going to be a I'll show you work on our in robotics, mainly about legged locomotion and vision and things like that. And this is our robot. Can you hear? Can you maybe raise the volume a little bit? Anyway, this is on the Berkeley Marina, this robot is struggling, but it manages to walk. Here's another example. Okay, notice the variety in terrain, right? The rocky terrain versus this leafy terrain, down slopes, etc, etc. And more. So, so this is the challenge. So the challenge is being able to do stuff to walk around in this variety of terrain. Okay, and here's I think. This is the robustness of our motor control systems that they can manage deal with this variety. And I'm going to start by situating this work in some major intellectual traditions. So there's an intellectual tradition of pattern recognition, which is what is behind most of the successes of machine learning. We we have developed techniques by where we give lots and lots of examples and say, these are examples of dogs. We cannot define the picture of a dog sort of axiomatically like a mathematician might, but we can give lots of examples of dogs or lots of examples of cats. And then we train these, these statistical machine learning systems, which use those examples to induce what the pattern is. And generalization is the central problem. Not all dogs look the same, not all cats look the same, and we should be able to do that. In motor control, we have an additional challenge, which is that we need robustness to disturbances. This is sort of a central aspect of control. You you knock the system a little bit, it should still work correctly. And then there is adaptation. And I want to emphasize that I'm using two different terms here. Robustness is typically to deal with noise. Adaptation means that they are very significantly different terrains and your system must work well under all of these conditions. And this connects to, I will not have too many equations in case my colleagues get scared. The basic idea here, go back to the development of control theory from around the 1960s, both in the in the US and in the Soviet Union, there were parallel developments. And these were amazingly successful. These these intellectual results from that era, they led, you know, John F. Kennedy made this statement by the end of the decade, we'll have man on the moon. Well, the people who delivered were obviously the aerospace people, but also the control theory people, because we needed to have these aircraft, these spacecraft move in orbits, which were controlled and accurate. So, so that is an old tradition. And that's what we'll draw draw on the basic. Okay, if you want to know what that equation stands for, X is something like state like what physicists would call, you know, positions and velocities and so forth. And X dot is how the state changes. The additional thing here is you, which is about control, how we externally put inputs into the system. So for a rocket, it might be firing thrusters and so forth. And and we choose to apply the right input you so as to have a desired result. And I think again, going back to that sixties tradition is the concept of adaptive control. So this this matrix A here, this refers to characteristics of the system. And those characteristics change. So in my example, that robot has to walk on different kinds of terrain, right? And those create very different conditions for it to walk. And the old examples were about aircraft, which in the course of a flight, a lot of the fuel will get used up. So therefore, the mass will be lower and you need to do the right things here. And this is again a time honored problem. But I think it's a problem which today with the tools of deep learning neural networks, we can revisit and we can make substantial progress on. And what we have done, I think falls into that family. And we'll see the applications of that to robotics. And I'll show you a few in my talk. And the key work that I'll talk about in a bit more detail is what we call rapid motor adaptation for legged robots. Okay, how do we do this? Again, I want to give you a flavor of this line of work. It's basically it's reinforcement learning to train these control laws, which means that you need to tell the robot has to learn how to move each of its joints. And a traditional control theorist would write down the equations, write down the math, derive the equations and that's how they would do it. We will learn it. And the way you learn it is basically by trial and error. It's called reinforcement learning. But that's just a fancy name for trial and error. And we need zillions of trials. And by the way, this is also true for human babies. Karen Adolf, who is a researcher at NYU, she has done a lot of experiments showing that babies actually do a lot of trial and error and fall many times. It's not that they just succeed in walking in like once in one go. So our robot has to do a lot of trial and error. Real hardware gets damaged, so you do it in simulation. And what do we do? We basically don't build in anything. We must, we just have these objectives that we set. You must try to walk without falling. You try to have a desired velocity. You try to use minimum energy. Something reasonable like that. And so the jargon here is we have this physical simulator. Policy, policy refers to what's called the controller, the brain of the robot, if you will. And it gets as an input, the state. The state here means what are all my joint angles? And A here refers to actions, which means what action did it perform last time? So what was my previous action and where are, what are all my joint angles? And now the policy gets that as input. It's like keeping track of what's your current state and your history. And then the action you command is, okay, how should I change all my joint angles? Okay. And this is going to be learned by trial and error. Now it turns out that doing this is hard because of all the variability in these different environments that we might have. That, the policy that will work for walking on hard ground is not the same policy which will work for sand or slippery surfaces or upstairs or downstairs and, and so on. So what do we do? So what we can do is to say put in some, in the simulator we have prior knowledge of that. We know what the mass is. We know the friction. We can, we can choose lots of variability in the condition. So we, we must exercise the robot to walk in all these different conditions in simulation. And we can, okay, and then that gets encoded into this environmental factor encoder, which you can think of as another neural network. And it captures that knowledge in the form of this variable Z. Z sub t, which is maybe eight dimensional. So this Z sub t is going to encode things like am I on flat ground? Am I on sloping ground? Am I on slippery ground? Things like that. And it's all done implicitly. We are not choosing a meaning for those variables. It's all done as part of this giant learning problem. And you do this for lots and lots of trials and at some point your robot manages to learn how to walk. Okay, so these we call extrinsics. And reward is, what is important for walking? What's important for walking is that you move, but you don't fall and you use minimal energy. I mean for, if you think about a biological setting, animals use a lot of their energy in hunting for food or evading, being hunted and so on and so forth. So this is important. Okay, so we can train this and in fact we trained it and this is our robot and it's able to walk and it's in simulation. It's just that this robot, if we try to take it to the real world, okay, I have a little problem. And the little problem is that this environmental factors are not known. Okay, so these are not available. So now I have a problem because I cheated in my simulator, the simulator which you can think of as like a video game simulator, right? It's capturing the physics and in that I train in different conditions and I've got it, right? Well, in the real world, how do I know which condition that I am, am I part of? So I'm stuck. So here is where the aha moment comes in. There is learning of the policy. How does the robot move? And then we can add a layer of meta learning. Okay, what is meta learning here? Meta learning here is observe your own behavior and from that deduce something about the conditions you are in. So the intuition is something like this. If I'm walking on, if I was doing the same set of movements on a sandy beach, what would happen is I would put my foot down, lift it up with the same force, but it wouldn't come up. It will come up only partially. So the same actions that I apply in different conditions result in different outputs and I can become aware of that and that is the signal to me that I am in different conditions. So the same actions have different consequences and those can be like my readout of what conditions I am. Note, by the way, that the robot that I've shown so far is actually blind. So all it has are its tactile senses, its proprioceptive sense and so on. So that's the key idea. This so-called adaptation module and I'm skipping details for obvious reasons. What it does is that it is like it's like that meta reasoning thing. It's operating at one level above. It has access to the past history. So maybe the last 0.2 seconds, maybe the last one second and in that one second I commanded the actions. So I know what I did. Biologists call this the efference copy. I have the states which I know because I have sensors in my body which keep track of all the joint angles and from that history I can estimate this z which is this extrinsic which you can think of as a proxy to the environment that I am in and that's it and it turns out so the discrepancy between the expected movement and actual measured movement and we continuously estimate these online and okay and then here is a little the next observation which is that in fact this process itself can be trained. So we have two levels of learning. Learning the basic policy and this meta policy and the meta policy also gets trained in simulation because in simulation I can vary the conditions and then I can train this estimator which estimates which condition I am in and yeah that's this part and then okay so now hopefully not use too much of jargon here and now we are ready to go and we have basically the same policy which runs in all of these conditions. So our slogan is one policy to walk them all okay and why it works is because it is not the same policy to walk them all that is one style of work that's the there's a what's called robust a robust policy might be one which where you use exactly the same everywhere no what we want is something which is adaptive it should do different things I should walk differently on concrete versus walking on sand okay and that's and that I can do if I can estimate what conditions I am in and basically we have the machinery for that so these are these are just some kinds of results okay so let me give you another example to illustrate how this is happening by analysis of adaptation module so soon you are going to see Ashish Kumar who is the student was the lead author in this work and this was work done in the COVID era so I told him you can take the robot home and then basically he was sleeping with the robot for two years and great research resulted as a okay and now let's see what he does so he's going to pour some olive oil on this mattress okay okay a waste of good olive oil no doubt and then you'll see the robot soon okay okay and if you notice he's got little plastic shoes on the socks on the robot and now and now the robot is going to walk okay and let's see this in slow mo okay so the robot has a default policy for walking on hard ground and it's getting into trouble on the slippery surface where the friction is much too low so what should happen what but it recovers so what's actually happening underneath the hood is that this meta thing this adaptation module is taking over the adaptation module is figuring out oh something must be wrong so now we see this and what you see in the charts the top rows correspond to the foot placement and then the the middle is the the force applied the torque applied at the knee but these these two curves correspond to two of the dimensions of this eight dimensional vector of state so notice so when it recovers it has what it has what's happened in the first phase is that it's it's it's got the wrong estimates of the physical conditions so it's not doing the right thing as soon as it's got the right estimate of which environmental condition it is in it which is happening sort of towards the right of that plot so think of the two bottom curves then after that it's got the right z and then the policy is different and then it does it works so so here's a different example of a modification which is that we throw a weight on this robot and this poor guy has to struggle okay and and now you'll see it struggling but adaptation and then it recovers and it does it fairly quickly like point two seconds so when we fall we do not fall instantaneously there's a physical process and it takes like half a second one second and in that time if you have estimated the conditions you're good and here are some examples where the top is our policy the the bottom shows what happened without that adaptation okay so the society for prevention of cruelty to robots has prevented me from doing more experiments like this okay now i'll show you some more stuff i think i've got through if you the anyway the meat the core idea i have now done and now i'm going to give you many corollaries of this so one is that this is the advantage of a learning framework is that many things emerge i mean the emergence is the the big thing in learning so here what we're going to do is we'll just give the robot the assignment go at point five meters per second go at one meter per second go at one point five meters per second everything else is the same it's exactly the same machinery and here's what we discover so what you see in these plots are the footfalls so the solid parts so it's right foot right front left front right rear left rear and the solid part means that the foot is on the ground so when it's asked to walk at point three seven five meters per second you see a gate which is like this slow walk if you ask it to walk at point nine meters per second okay then what you get is a trot so in for horses you have all these gates trot gallop canter etc and these were not programmed in these just emerge and then if you have set a really high speed and now of course you notice that most of the time the all the full there are times when all four feet are above the ground so what's an explanation for this okay so this is what happens from learning but as scientists we want to know more and it turns out that people had already thought of it in biomechanics and the explanation is energy energy matters and you want to be efficient energetically so for example for humans at low speeds energetically walking is efficient and at high speeds running is efficient for horses you have multiple gates and these were studies done by you know taking horses and putting a bag around their nose and measuring oxygen and so on and there are particular phases in which this efficiency occurs and exactly the same is true for our quadruped and then here are some fun examples of basically we now are just remotely telling it to go at different speeds and its gate changes smoothly and I think this is just meant to you okay and now let me say by the way everything here was done blind okay and I'm a I've spent 35 years of my life studying vision but we wanted to be but I our theory was blind people can walk so certainly blind robots should be able to walk then why do you need vision well you need vision in these kinds of conditions right you you need to walk on stepping stones on stairs and so on and this is some work that I'll show you and okay I'll not go too much technical the that that there's a traditional technique for this which is to build maps by combining information from multiple views it turns out that view is is suffers from the effect of noise because every time you estimate the camera it's noisy so the resulting map is way too noisy and what we did was we we developed direct policies so direct here means that you have the visual data and you try to develop control so it's like you're trying to make it into a reflex rather than a very conscious process of build a map plan your footsteps and so on and and I'm going fast but the philosophy is very similar to what we did earlier in simulation you can train it with privileged information so you can give the robot access to what the terrain is and it ignore all the symbols please it learns a policy for moving in this terrain and now what you do is to deploy it what we do is we put a camera we put a camera on the head of the robot and this camera gets you this kind of an image and with that image what you try to do is to try to guess what the environmental parameters are so it's now going to be more about terrain geometry whereas previously it was more about terrain friction hardness and the like but roughly speaking think of it as being the same philosophy and then we can implement this and uh okay blah blah okay uh okay let's let's see some demo okay so this is the robot it has a camera on its head it has no advanced knowledge of anything okay it does not know the terrain at once okay so different examples so we have to improvise some obstacles okay so these are stairs okay these are fairly tall stairs okay this you might recognize this is the rose garden in berkley and we were convinced that this would not work will it okay anyway you get the idea so so this is a remarkable success because this is not something which typically any other system could do effectively so this paper won the best prize at the robotics conference in last December okay i i will see if this video works so this is work with a colleague in mechanical engineering let me see if this works okay so this is a biped so bipeds are harder quadrupeds are more stable and uh and uh okay i think i'll i'll uh i think i will do let's see if i'll move it i'll speed the video a bit okay okay so someday this robot will carry loads for you okay uh okay uh this general idea is applicable not just for locomotion but also for manipulation so i'll show you an example where we have a hand and again this hand is trying to twirl different objects and the interesting point here is that there are all these different objects of different sizes shapes weights friction and so on and it succeeds and it turns out it's the same basic philosophy what we call rapid motor adaptation which is you you perform some action you see the consequences of the action and you use this to at the meta level figure out what conditions you are in okay and uh okay i'll go i will skip this okay let me i wanted to make a philosophical point here okay uh how to think about intelligence so i have the last five minutes so i want to get to that this is a beautiful quote from alan turing who's considered like the founding figure of computer science and he said this in this 1950 article which proposes the turing test and everybody reads that article but never goes to section six which is where this occurs instead of trying to produce a program to simulate the adult mind why not rather try to produce one which simulates the child's if this were then subjected to an appropriate course of education one would obtain the adult brain and during was 1950 but since then we know a lot more our colleagues in psychology and neuroscience and child development have told us a lot about how children learn so children's learning uh my colleague Alison Gopnik has this phrase the scientist in the crib the child is doing these experiments putting things into her mouth has multiple senses touch uh touch vision hearing etc and then they are cross coupled uh and then that child eventually uh in that first year has become essentially a sensory motor genius and then in the second and then at some point starts to acquire language and then of course goes to school and then of course then the the the the stage the kind of learning which is embodied in gpt 3 and gpt 4 takes off but it takes off on a base of this more elementary sensory motor learning and my belief is and this is for now an ideological belief that this is the pathway for in artificial intelligence as well that we need to copy that and uh this is uh article from linda smith and michael gaser or two psychologists that these are aspects of how children learn multimodal sensing touch audition vision incremental we build on our past knowledge physical it's not the brain in a vat it's not the mind in a vat it is interacting with the world embodied uh explore social we learn from others and then of course we do like use language so i wish to fight against the tyranny of linguistic imperialism okay and uh in the last three two minutes i'll talk about a project uh with antonio locations here and she shared myself and this one we tried to take this idea of sort of copy the idea of childhood learning which is cross model and we said here's a robot which has got a camera but it's going to learn in the real world how to use its vision system so it starts out blind so it's struggling okay and then the vision system with a vision system of course you can climb stairs and you'll see that but how do we train the vision system and we wanted it to learn in the wild so here was our intuition if you think of a robot in the stairs it's in its proprioception its senses its joint angles can let it compute what's the depth of the its left leg and right leg and so on it has that geometry from its joint angles from its internal state so can we use it for training without uh okay so the idea was the proprioception predicts the depth of every leg and the vision system gets an image and what we ask the vision system to do is to predict what the depth will be one point five seconds later so so and that was the idea that you just shift what signal it will know one point five seconds later and use that to do this advanced prediction and so we have this robot which is learning day by day so the first day it's clumsy the second day it goes up further and then on the right you see the success rate and then finally on the third day you will see that it actually will it make it all the way it makes it all the way and we can do experiments like we can mess up its vision system there's a ability that humans have that we are calibrating all the time so Antonio is going to rotate the camera and that's going to make it uh it's so it's messed it up it's like taking your eyes and you know switching them moving them 30 degrees but because it is always learning it's always adapting it's going to initially it will struggle I hope not I don't hope but it will okay and then it recovers because it's learning from its own experiences and and then this is a little let me play this one so this is before adaptation and then this is after adaptation and adaptation is just a few minutes so this is called the prism test in in vision science and humans can do it in like 10 minutes well so can a robot so I hope I've taken you on this little journey of what some of the some work into some what I think is potentially interesting work in sensory motor learning and I want to conclude with acknowledging some people acknowledgement first I want to thank all my past and present PhD student this work is really due to them I merely a spokesperson on the stage and of course all my past and present postdoctoral fellows thank you thank you very much and then I want to say something about my colleagues particularly in the ECS department with whom I've spent 37 wonderful years starting with a lab I started as a 25 year old assistant professor and I think Berkeley was the best place for starting assistant professor it probably still is and I'm very thankful for the opportunity that was given to me I'm thankful to my colleagues in the rest of the university from whom I learned so much at vision science cognitive science neuroscience I learned all about the brain perception and action from them I want to thank my currently my colleagues in bear with whom I have such fun time and the wonderful Angie who keeps the place running and finally of course my wife Isha and my son Kaby for so much patience and support thank you very much we we have a robot which you can which we which we're we're gonna skip or outside he'll have he'll show it outside so after the talk people can check it out yeah but I'm happy to answer questions yes can you please use the mic yes I'm wondering whether there are any limitations thank you for the great talk thank you are there any limitations between the time that you need to learn a task and that the time that you actually want or or need to perform it to to be efficient you see what I mean because there's two times yes right so so I distinguish between two kinds of learning so in our we do this learning in simulation which takes a very very long time but since it's done in simulation it's totally safe and we can take a very huge complexity it can have a huge complexity then the learning which you do and and so there are no limits there we can train forever then there are limits then what there is the learning which you do in the real world that has to be very rapid so the initial adaptation that I showed that was on the order of a 0.5 second or 0.3 second that kind of thing if it is longer than that you'll fall right but we are adapting all the time if I sprain my ankle I can walk and that's adaptation then there was the second so that's the time constant that I need it needs to be fast but not instantaneous and then I talked about this recalibration that we have which is of our entire sensory motor system it happens when we wear glasses it happens when we grow and things like that that you have a bit more leeway but it's in our system we showed it could be done in a few minutes which is roughly comparable to what humans can do for this effect it's called the prism adaptation so I don't have a generic answer for all things but I'm just showing you for these systems what it is it's important the timing is important adaptation has to be quick enough otherwise it won't work question yeah yeah thank you for your talk the title of your talk kind of implies that there are multiple roads to artificial intelligence what would be the second road well the the obvious road that people are taking right now is the is the road from linguistic data from all the words on the web that's how you train GPT-3 and chart GPT and I am saying that captures a certain kind of knowledge but it is an incomplete story when we want to talk about intelligence as natural intelligence goes as an engineer you can totally use that incomplete system in some way but I mean I'm still striving after human level intelligence in all its facets and for someone like me that is not enough I want that intelligence which is embodied and grounded in the real world which must start from that that experience yeah I don't know who's yeah okay I think then there's a question here in the front yeah yeah yeah there's someone there and then maybe you'll get yeah yeah thank you thank you very much for a great talk you started your talk with with autonomous vehicles and I think recently in use was a situation where autonomous vehicles in San Francisco was in an intersection and there was another car like doing donuts on the intersection and the problem was that the autonomous car hadn't seen the situation before so there is the one old data in the simulation and the car wasn't able to deal with the situation right and my question is how to deal with safety critical situations because in such situations for example the robot has no rights to you know to engage in this situation and the human can reason about the situation based on other knowledge yeah so the question was really about situations which are not seen in training so this comes down to the emergence issue what abilities emerge so humans certainly have I mean I'm driving on a on the road and there's this truck and some junk falls from the back if it's newspapers I will drive over if it's rocks I'll switch to a different lane right because we know something about the properties of rocks versus newspaper now that is connected to our vast wider knowledge so the the article of ideological belief among AI researchers is that with enough knowledge and that then these abilities will emerge it remains to be proved certainly the more narrowly trained systems that we have today don't have it but that's one answer the second answer is that we may want to explicitly have what's called system one and system two so this is terminology from Daniel Kahneman system one is more reflexive system two is is this deliberative logical conscious system okay by the way there's our dog walking okay it can do fancier things than then walk on the ground but at least it is it can walk on the ground but a colleague here had a question I think he had a question over there yeah you were using the infant infant learning for this adaptability of the of your AI but it takes say a very say you know a very simple animal perhaps it's not a simple animal like an ant the ant can walk and do all that ants need to do immediately from a hatching so I'm wondering yeah yeah sorry his question is a is a fundamental and deep one which is that there are certainly species which don't need all this learning period okay so this is Alison Gopnik has an article on this so this is the thing about precociality so and what you get is a very hardwired system which is not so adaptive but you but you can work on day zero and human babies don't have that which gives us much greater adaptability so what the price you pay for that is that there's a period of vulnerability for a period of 10 years the human child cannot survive by himself or herself right that's the price you pay but as a result you get a system which is much more adapted to the to the world and somehow evolution their different species have made these different trade-offs on this dimension what training we do in simulation I like to think of it as being somewhat akin to the process of evolutionary search I mean as a species we have we have certain knowledge which is encoded in our genome and that has come from some 550 million years search process so I want to distinguish between the the learning that occurs over that time versus the learning that occurs in the life of an individual I mean the world keeps changing right and that that is what we get from this thank you thank you