Back

Jim Keller: Elon Musk and Tesla Autopilot | AI

Transcript

[00:00:01] Host Create clip all the cost is an equipment to do it. And the trend on equipment is, once you figure out a building equipment, the trend of cost zero, Ellen said. First you figure out what configure Asian you want the atoms in. And then how to put some there, right? Yeah, because, well, what do you hear that you know his His great insight is people are how constraint I have this thing. I know how it works. And then little tweaks to that will generate something as opposed to what do I actually want and then figure out how to build it. It's a very different mindset, and almost nobody has it obviously.

[00:00:40] Host Create clip Well, let me ask on that topic you were one of the key early people in the development of autopilot, at least in the hardware side. Elon Musk believes that autopilot and vehicle autonomy, if you just look at that problem, can follow this kind of exponential improvement in terms of the hot, the how question that we're talking about. There's no reason why I can't What are your thoughts on this particular space of vehicle autonomy and you're a part of it, and the law mosques and Tesla's vision for, well,

[00:01:13] Host Create clip the computer you need to build with straightforward and you could argue well doesn't need to be two times faster or five times or 10 times. But that's just a matter of time or price in the short run. So that's that's not a big deal. You don't have to be especially smart to drive a car, so it's not like a super hard problem. I mean, the big problem. Safety is attention, which computers air really good at, not skills.

[00:01:40] Host Create clip Well, let me push back on one. You see, everything you said is correct. But we, as humans tend to ah, tend to take for granted how how incredible our vision system is so

[00:01:55] Host Create clip you can drive a car or 2050 vision, and you can train a neural network to extract a distance of any object in the shape of any surface from a video and data.

[00:02:06] Host Create clip But that really simple. And that's simple. That's a simple data problem. It's not that simple. It's because you because it's not just detecting objects, it's understanding scene, and it's being able to do it in a way that doesn't make errors. So the beautiful thing about the human vision system and our entire brain around. The whole thing is we were able to fill in the gaps. It's not just about perfectly detecting cars. It's inferring the included cars. It's tryingto it's understanding. I think that's mostly

[00:02:40] Host Create clip a bigger problem.

[00:02:42] Host Create clip So you think What data? With compute with improvement of computation with improvement and collection.

[00:02:48] Host Create clip There's a you know when you're driving a car and somebody cuts you off, your brain has theories about why they did it. No, they're about person. They're distracted. They're dumb, you know, you could listen to yourself, right? So, you know, if you think that narrative is important to be on the success we drive a car, then current autopilot systems can't do it. But if cars air ballistic things with tracks and probabilistic changes of speed and direction and roads are fixed and given by the way, they don't change dynamically, you can map the world really thoroughly. You can place every object really thoroughly right. You can calculate trajectories of things really thoroughly right.

[00:03:34] Host Create clip But everything you said about really thoroughly has a different degree of difficulty. So

[00:03:40] Host Create clip you could say at some point, computer, autonomous systems were way better, things that humans are allows yet, like it'll be better abstention. They'll always remember there was a pothole in the road that humans keep forgetting about. Remember that this set of Rhodes has these weirdo lines on it that the computer's figured out once in, and especially if they get updates. So if somebody changes a given, like thick Ito robots and stuff, somebody's that is the maximized the givens, Right, Right. So we're having a robot pick up this bottle copies ways. You've put a red dot on the top because then you have to figure out, you know, if you want to do a certain thing with that, you know, maximize the givens is the thing and autonomous systems air happily maximizing the givens like like humans. When you drive someplace new, you remember it because you're processing at the whole time. And after the 50th time you drove to work, you get to work. You don't know how you got there, right? You're on autopilot, right? Autonomous cars, they're always on autopilot, but the cars have no theories about why they got cut off or while they're in traffic.

[00:04:49] Host Create clip So they also never stopped paying attention, right? So I tend to believe you do have to have theories, mental models of other people, especially pedestrians, cyclists, but also with other cars. Everything you said is Ah, like, is actually essential to driving. Driving is a lot more complicated than people realize. I think so started to push back slightly. But Thio

[00:05:12] Host Create clip cut into traffic, right? You can't just wait for a gap. You have to be somewhat aggressive. You'll be surprised how simple calculation for that is.

[00:05:21] Host Create clip I may be on that particular point, but there's, um it. I maybe ask her to push back. I would be surprised. You know what? I'll just say where I stand. I would be very surprised, but I think it's you might be surprised how complicated it is that

[00:05:37] Host Create clip I say. I tell people like progress disappoints in the short run surprises in the long run.

[00:05:41] Host Create clip It's very possibly a

[00:05:43] Host Create clip suspect in 10 years. It'll be just like, taken for granted,

[00:05:47] Host Create clip but you're probably right now. Looks like

[00:05:49] Host Create clip it's gonna be a $50 solution that nobody cares about. Like GPS is like, Wow, GPS. We have satellites in space that tell you where your location is was a really big deal. Now everything is the GPS, and

[00:06:01] Host Create clip yeah, it's true. But I do think that systems that involve human behavior are more complicated than given credit for, so we can do incredible things with technology that don't involve humans. But when you think

[00:06:13] Host Create clip humans are less complicated than people you know frequently obscure I've

[00:06:18] Host Create clip maybe I felt

[00:06:18] Host Create clip his hand off right out of large numbers of patterns and just keep doing it over and over.

[00:06:23] Host Create clip But I can't trust you because you're human. That's something something a human would say. But I might be my hope was on. The point you've made is even if no matter who's right there. I'm hoping that there's a lot of things that humans aren't good at that machines are definitely good. I like you, said attention. Things like that will they'll be so much better that the overall picture safety in autonomy will be. Obviously, cars will be safe, even if they're not as good. And

[00:06:51] Host Create clip I'm a big believer in safety. I mean, they're already the current safety systems, like cruise control that doesn't let you run into people in lane keeping. There are so many features that you just look at the Prado of accidents and knocking off like 80% of them, you know, super doable.

[00:07:10] Host Create clip Just the linger on the autopilot team in the efforts there. The it seems to be that there is a very intense scrutiny by the media and the public in terms of safety, the pressure, the bar. But before autonomous vehicles, what do your sort of as a person? They're working on the hardware and trying to build a system that builds a safe vehicle and so on. What was your sense about that pressure? Is it unfair? Zit expected of new technology?

[00:07:39] Host Create clip Yeah, it seems reasonable. I was interested. I talked to both American and European regulators, and I was worried that the regulations would right into the rules. Technology solutions like modern brake systems imply hydraulic brakes. So if you'll read the regulations, t meet the letter of the law for breaks. It sort of has to be hydraulic, right? And the regulator said they're they're interested in the use cases like a head on crash and offset crash Don't hit. Pedestrians don't run into people. Don't leave the road. Don't run a red light or a stop light. They were very much into the scenarios and, you know, and they had they had all the data about which scenarios injured or killed the most people. And for the most part, those conversations were like, What's the right thing to do to take the next step? Now L. A. Is very interested also in the the benefits of autonomous driving or free people's time and attention as well. A safety, Um, and I think that's also an interesting thing. But, you know, building autonomous system so they're safe and safer than people seemed, since the goal is to be tannic, safer than people, having the bar to be safer than people and scrutinizing accident seems philosophically, you know, correct.

[00:09:06] Host Create clip So I think that's a good thing.

[00:09:08] Host Create clip What are is different than the things you work that the intel in de Apple with autopilot, chip design, hardware design, what are interesting and challenging aspects of building this specialized kind of computing system in the automotive space.

[00:09:27] Host Create clip I mean, there's two trucks to building like an automotive computer. One is the software. Our team, the machine Learning team, is developing algorithms that are changing fast. So has your building the accelerator. You have this. You no worry our intuition that the algorithms will change enough that the accelerator will be the wrong one. Right? And there's the generic thing, which is, if you build a really good general purpose computer say its performance is one and then deep. You guys will deliver about five extra performance for the same amount of silicon because instead of discovering parallelism, you're given parallelism and then special accelerators get another 2 to 5 X on top of a GPU. Because you say, I know the math is always a bit in titters into 32 bit accumulators and operations air the subset of mathematical possibilities. So auto, you know, aye, aye. Accelerators have a claim performance benefit over G pews because in the narrow maths space, you're nailing the algorithm. No, you still try to make it programmable, but ah, I I feel this changing really fast. So there is a you know, there's a little creative tension there of. I want the acceleration afforded by specialization without being over specialized so that the new algorithm is so much more effective that you'd have been better off on a GPS. So there is a tension there.

[00:10:57] Host Create clip Toe build a good computer for an application. Like a lot of motive. There's all kinds of center inputs and safety processors and a bunch of stuff. So one of the loans goals to make it super affordable. So every car gets an autopilot computer. So some of the recent starts you look at it and they have a server in the trunk because they're saying, I'm gonna build this autopilot computer replaces the driver, so their costs budgets 10 or $20,000. And Dylan's constraint was, I'm gonna put one every in every car, whether people by auto Tana striping or not. So the cost constraint he had in mind is great, right? And to hit that you had to think about the system design. It's complicated. It's it's fun, you know. It's like it's like it's craftsman's work like a violin maker, right?

[00:11:41] Host Create clip You

[00:11:42] Host Create clip could say Stradivarius. This is incredible thing. The musicians are incredible, but the guy making the violin, you know, picked wood and sanded it, and then he cut it, you know, they glued it, you know, when he waited for the right day so that when you put the finish on it. Didn't you know, do something dumb? That's craftsman's work, right? You may be a genius craftsman because you have the best techniques and you discover a new one. But most engineering's craftsman's work and humans really like to do that. You know,

[00:12:12] Host Create clip stress, Mark. Humans know everybody. All human. I

[00:12:15] Host Create clip I used. I dug ditches when I was in college. I got really good at it. Satisfied? Yeah. So

[00:12:21] Host Create clip digging ditches also cross middleware.

[00:12:23] Host Create clip Yeah, of course. So So there's an expression called complex mastery behavior. So when you're learning something, that's fun, because you're learning something when you do something and throat and simple, it's not that satisfying. But if the steps that you have to do are complicated and you're good Adam, it's satisfying to do them. And then, if you're intrigued by it all, as you're doing them, you sometimes learn new things that you can raise your game. But Christmas work is good, and engineers like engineering is complicated enough that you have to go learn a lot of skills. And then a lot of what you do is an craftsman's work, which is fun.

[00:13:01] Host Create clip Tom was driving, building a very resource constrained computer, so computer has to be cheap enough that put in every single car that's essentially boils down to craftsman's work, its engineering and

[00:13:13] Host Create clip other thoughtful decisions and problems to solve and tradeoffs. To make you need 10 camera and ports or eight. You know you've been building for the current car, the next one. You know, How do you do the safety stuff? You know, there's there's a whole bunch of details, but it's fun. But it's not like I'm building a new type of neural network, which has a new mathematics and the new computer at work. Do you know that that's like there's there's more invention than that, but the rejection of practice. Once you pick the architecture, you look inside, what do you see? ADDers and multipliers memories, and you know the basics. So computers was always this. This weird sort of abstraction layers off ideas and thinking that reduction to practice is transistors and wires and, you know, pretty basic stuff. That's an interesting phenomenon, by the way that, like factory work like lots of people think factory workers, road assembly stuff. I've been on the assembly line like the people work that really like it. It's a really great job. It's really complicated putting cars together, heart right and in. The car's moving and the parts are moving, and sometimes the parts are damaged and you have to coordinate putting all this stuff together and people are good at it. They're good at it. And I remember one day I went to work in the line was shut down for some reason and some of the guys sitting around we're really bummed because they had reorganized a bunch of stuff and they were gonna hit a new record for number cars built that day. And they were all gonna go to do it on the news or big, tough buggers, you know. But what they did was complicated and you couldn't do it.

[00:14:47] Host Create clip Yeah, and I mean,

[00:14:49] Host Create clip well, after a while, you could. But you have to work your way up because, you know, like putting a bright what's called the brights. The trim on a car on a moving assembly line where it has to be attached 25 places in a minute and 1/2 is unbelievably complicated, and Andi human beings could do It's really good. I think that's harder than driving a car, by the way,

[00:15:12] Host Create clip Putting together working,

[00:15:14] Host Create clip working on the factory

[00:15:16] Host Create clip to smart people can disagree. Yeah, I think Dr. Driving a car.

[00:15:22] Host Create clip We'll get you the factory something. We'll see them.

[00:15:24] Host Create clip Not for us. Humans. Driving a car is easy. I'm saying building a machine that drives a car is not easy. Okay, Okay. Driving a car is easy for humans because we've been evolving for billions of years. Drive

[00:15:38] Host Create clip cars, you know, to do wth the trail. If the car's air supercool

[00:15:44] Host Create clip Now you join the rest of the mocking me. Okay? Just yeah.

[00:15:51] Host Create clip Intrigued by your You know, your anthropology.

[00:15:54] Host Create clip Yeah, I

[00:15:55] Host Create clip have to go dig into that.

[00:15:56] Host Create clip There's some inaccuracies there. Yes, Okay. But in general, what have you learned in terms of thinking about passion, craftsmanship, tension, chaos, You know, the whole mess of it. What have you learned? Have taken away from your time working with L. A mosque, working at Tesla, which is known to be a place of chaos, innovation, craftsmanship. And I really like the way

[00:16:32] Host Create clip he thought, like you think you have an understanding about what first principles of something is. And then you talk to you alone about it. and you you didn't scratch the surface. You know, he has a deep belief. No matter what you do is a local maximum Right now, I had a friend. He invented a better electric motor. And, uh, it was like a lot better than what we were using. And when they came by, he said, You know, I'm a little disappointed because, you know, this is really great. You didn't seem that impressed. And I said, You know, in the super intelligent aliens come, Are they gonna be looking for you? Where is he? The guy built a motor. Yeah, probably not. You know, like like the butt, doing interesting work that's both innovative. And let's say Craftsman's work on the car and saying it's really satisfying is good. And that's cool. And then Ellen was gonna take everything apart, like, what's the deep First principle? Oh, no. What's really know what's really know? You know, you know that that you know, ability to look at it without assumptions and and how constraints This super wild, you know, he built rockets happened

[00:17:44] Host Create clip usually looking car, you know

[00:17:45] Host Create clip everything and that's super fund. And he's into it, too, Like when they first landed to Space X rockets to Tessa. We had a video projector in the big room and, like 500 people came down, and when they landed, everybody cheered and some people cried. It was so cool. All right. But how did you do that? Well, no super hard. And then people say, Well, it's chaotic, really? To get out of your your assumptions. You think that's not gonna be unbelievably painful. And New Zealand tough? Yeah, probably the people look back on it and say, Boy, I'm really happy I had that experience to go. Take apart that many layers of assumptions sometimes Superfund, sometimes painful,

[00:18:33] Host Create clip that could be emotionally and intellectually painful, that whole process just stripping away assumptions.

[00:18:38] Host Create clip Yeah, imagine 99% of your thought process is protecting yourself conception, and 98% of that's wrong. Now you got that math, right? How do you think you're feeling when you get back into that one bet that's useful. And now you're open and you have the ability to do something different. You know, if I got the mass right, it might be 99 point knowing, but in 8 50

[00:19:06] Host Create clip imagining it, the 50% is hard enough.

[00:19:11] Host Create clip Now For a long time, I've suspected you could get better. Look, you can think better. You can think more clearly. You can take things apart. There's lots of examples of that. People who do that.

[00:19:25] Host Create clip So any Lana's example of that pair? You are an example. So I don't

[00:19:30] Host Create clip know if I am. I'm fun to talk to. Certainly I've learned a lot of stuff,

[00:19:36] Host Create clip right? Well, here's

[00:19:37] Host Create clip the other thing is like I don't like like I read books and people think, Oh, you read books? Well, no. I brought a couple of books awake for 55 years. Well, maybe 50 cause I didn't read, learned retailers age or something, and, uh, and it turns out when people write books, they often take 20 years of their life where they passionately did something, reduce it to 2 200 pages. That's kind of fun. And then they go, you go online and you can find out who wrote the best books and who like you know, that's kind of my eldest daughter is this wild selection process, and then you can read it for the most part, understand it, and then you can go apply it like I went to one company and I thought I haven't managed much before. I read 20 management books and I started talking to him. And basically, compared to all the VP is running around. I'd run night, read 19 more management books in anybody else, wasn't even that heart and half the stuff worked. Like first time. It wasn't even rocket science.

[00:20:41] Host Create clip But at the core of that is questioning. The assumptions are sort of entering thinking first principles, thinking sort of looking at the reality of the situation and using him using that knowledge, applying that knowledge. So

[00:20:56] Host Create clip yeah, so I would say my brain has this idea that you can question first assumptions, and but I can go days at a time and forget that you have to kind of like circle back that observation

[00:21:10] Host Create clip because it is because you are challenging.

[00:21:12] Host Create clip Well, it's hard to keep it front center because, you know, you're you operate on so many levels all the time and, you know, getting this done takes priority. Or, you know, being happy takes priority. Or, you know, screwing around takes priority. Like like like how you go through life is complicated. And then you remember Oh, yeah. I could really think First principle. So much shit that's tiring. You know what you do for what, Helen, that's kind of cool.

[00:21:40] Host Create clip So just as a last question, your sense from the big picture from the first principles, Do you think you kind of answered already? But do you think autonomous driving something weaken solve on a timeline of years? So 1235 10 years, as opposed to a century just to linger in a little longer. Where's the confidence coming from? Is it the fundamentals of the problem, The fundamentals of building the hardware and the software

[00:22:14] Host Create clip as a computational problem understanding ballistics rolls, topography It seems pretty solvable. I mean, and you can see this, you know, like like speech recognition for a long time. People are doing, you know, frequency and domain analysis and and all kinds of stuff. And it didn't work for at all, right? And then they did deep learning about it, and it worked great. And it took multiple iterations and, you know, autonomous driving his way past the frequency analysis point. You know, use radar, don't run into things. And the data gathered is going up in the computations going up and the algorithm understand is going up, and there's a whole bunch of problems getting solved like that.

[00:22:59] Host Create clip The data side is really powerful, but I disagree with both you and your honor. I'll tell you on once again, as I did before that, that when you add human beings into the picture, the it's no longer ballistics problem. It's something more complicated, but I could be very well proven wrong.

[00:23:18] Host Create clip Cards are hardly damped in terms of ready to change, like the steering and the steering systems really slow compared to a computer. The acceleration of the accelerations really slow,

[00:23:28] Host Create clip Yeah, on a certain time scale on a ballistics timescale. But human behavior? I don't know it. E Shouldn't seeing

[00:23:35] Host Create clip beings really flow to? Weirdly, we operate, you know, half a second behind reality. I'll be really understands that one either. It's pretty funny.

[00:23:44] Host Create clip Yeah, yeah,

[00:23:46] Host Create clip so it

[00:23:48] Host Create clip will be very well could be surprised. And I think with the rate of improvement in all aspects on both the computer in the the software in the hardware, there's going to be pleasant surprises all over the place.