Robo Taxis And Autonomous Vehicles with Steve Jurvetson

Within the next sort of five years, if current trajectories continue, we should be able to build computers that exceed all computation on Earth combined. And this would be a machine that fits in a single room, costs single digit million dollars to buy, and then give it like another five years, it could out compute the entire universe. What does that mean? I mean it's mind bending. And this was David Deutsch that first came up with this analogy that said if you took the entire matter of the universe and reconstituted it into the best computer one could ever imagine, designing it still cannot solve certain problems in these computer.

Welcome to the Capability Amplifier, the show for business owners and entrepreneurs who want high performance upgrades for their brains, bodies and bank accounts.

Steve Jurvetson, so nice to have you here. We're at the near future summit and I've been a big fan. I followed you for a long time and some of the work you've been doing is extraordinary. So I think I'd like to start this out by just asking what are you most excited about right now in terms of the patterns of behavior, investment opportunities, and where the industry as a whole is going from a technology perspective? Silicon Valley and everything that you're involved with, space and beyond? That's a big broad question. I did it on purpose.

Well, thank you. And I think that sets me up for maybe the best answer is I am most blown away by the diversity and breadth of opportunities, meaning the meta point that there are more industries ripe for venture capital and entrepreneurial revolution than ever before. Like if I Compare to the 90s, when I started venture capital, it was software, semiconductors and healthcare. That's it. If you invested in automotive, you'd lose your money. If you invested outside of this sort of narrow domain, it was a sort of fool's errand. And today almost every industry is going through a massive revolution. Some sooner than others, like automotive and aerospace, but eventually everything. Agriculture, financial services, you name it, they're all converting to software centric businesses, to intelligence centric businesses, ones where deep learning is going to have a major impact. And for a sense of how this is going to restructure the economy, just look at the automotive sector today and say, where's innovation coming from? Look at the aerospace sector even more dramatically with SpaceX fundamentally leading the charge with few followers. And I think you see a portent of what's to come in almost every industry. I have never been more excited. I love learning, I love technology, I'm a geek at heart. And so there's more learning and More frontiers than ever before. So, specifically, there's a lot going on, obviously, in autonomous vehicles and driving that'll touch all of our lives when we switch to these robo taxi fleets and urban equivalents, where you just imagine Uber or Lyft experience, but without the driver, without the smells, with a car that has no cockpit and, you know, you have all the room in the world and. And it's just a better experience in every way. Three times safer, three times cheaper, et cetera.

And Elon's got a bunch of cars ready to go. He's predicting. I mean, he's making some big claims right now. What do you think about the timing right now?

To be fair, let me caveat. I'm on the board of directors of Tesla.

That's true.

I know we got a SpaceX and a Quantum computer company called D Wave. So let me just caveat to say what he said publicly is that we'll have a million robotaxis on the road by 2020. What that means, of course, is the installed base, so that all of the vehicles shipping today ship with these incredible new chip that Tesla has developed to do neural networks more powerfully, like, you know, a couple of orders of magnitude more powerfully than predecessor technology. That he knows the hardware is there, the sensors are there, and they're working on the software stack. And so the. His belief is that by next year, that software stack will ship and it gets promoted over the wire. So all of a sudden, you know, a million vehicles are capable of being autonomous vehicles, capable of being recruited into a fleet that makes an income stream for their owner, such that it pays back the ownership of the car in a ridiculously short period of time and shifts the entire economics of what it means to own a car.

Right.

It's kind of like airbnb, your vehicle, you know, on, you know, a fractional share, like a few hours at a time. I mean, by the way, because consider like, your vehicle, on average is used about 3% of the time. It just sits in a parking lot the vast majority of the time. And that's just incredibly inefficient. This is what contributes to, you know, congestion, to traffic jams, to urban congestion. People spend between 30 and 40% of their urban driving time looking for parking. All that goes away if you just get out of the vehicle and you.

Just have a little schedule app, drives itself off.

Yeah.

So I just say, here's the times I want my car to be working for me.

Exactly. They get pulled out. And as a driver, your benefit is never looking for parking again, what an incredible blessing. By the way, every product SpaceX makes is an autonomous vehicle. The booster comes back autonomously, the, the Dragon capsule for crew. And much of their success can be attributed to the advances of computation, Moore's Law, both in navigation and control systems as well as the simulation rig. So both the autonomous car companies and the autonomous rocket companies basically benefit from billions of hours of simulation training before they ever fly or drive in the real world. And that's why a lot of these things are working the first time and more quickly than anyone would have guessed. So, by the way, to your earlier question, when will these robo taxis come? If you look back at everything SpaceX has done, every incumbent company, every expert in the field would say they're crazy, they'll never get it done, it'll never work. And in fact, it worked the first time.

Well, part of the reason for that, I assume, is all the sensors, all the data that's being acquired and aggregated right now is allowing so much predictive analysis to take place, et cetera, et cetera. It's phenomenal. Just, again, the value of the data.

That's right. And closing that feedback loop, you're exactly right that both autonomous cars and rockets are festooned with sensors. You can see it with the video feeds, they're just walking sensor arrays. And by the way, every Tesla vehicle driving, even if it's not using autopilot, is gathering that data. So if, if they want to go back and say, hey, give us examples of random trash in the road so that we can train a neural network on what to do about it, well, they can use their fleets of cars to get hundreds of thousands of images like that at the snap of a finger that are real world examples of trash in the road. And like, what do we do in each case?

And presumably rain over here, sleet over here, time of day, combine that data and extrapolate.

Exactly.

So let's talk a little bit about quantum computing, because you've committed a lot of resources to that. There's a lot going on and I know it's a big interest of yours, so we better start. Just in case any of our listeners don't know what quantum computing is, let's do the 30 second commercial on what it is and why it's so important.

Sure. Quantum computing is as weird as the name may imply, it uses quantum mechanics as a fundamental basis of computation in a way that is unlike any computer we've ever heard of before. So the important thing to keep in mind is it's not just, oh, another supercomputer, oh, a marginal improvement. The way you hear about from an intel or a Cray superconductor in the past. This could be most analogously, or maybe the way to best rattle the listener to understand how different this is, is that the only explanation for why and how it works so powerfully is that it engages the computational resources of parallel universes. Now, if you're only competing in this universe, how could you compete with parallel universes of trillions of refractive echoes across these computational resources that are replicated across parallel universe? I mean, like your mind explodes. And by the way, the physicists who are the domain experts in this field, their mind explodes too. It's like they cannot ever make rational sense in a way that we would intuitively understand how this works. So let me try to make it more pragmatic. Another way to think about it is imagine you had a register of binary bits. The ones and zeros that we know make up all of computation. But instead of being a one or a zero, they can simultaneously, and this is a little bit of hand waving, but this gives you a sense of it be a simultaneously a one and a zero. So if you were trying to say what is the code that will unlock a key, a digital key, let's say you've built an encryption system and you want to know what series of bits of ones and zeros will unlock this secret lock. What if you could present all possible combinations of ones and zeros simultaneously and just it pops open. It's that radically different than a brute force search, like a normal digital computer would say, let's start with 1, 2, 3, 4, and count up until we eventually break the lock, because we have no reason to know what the code will be. So if you can simultaneously explore all of the variables, all the quantum variables, let's say, of a problem space, you can solve problems in times that normal computing can never catch up. Or phrase the different way, within the next sort of five years, if current trajectories continue, we should be able to build computers that exceed all computation on Earth combined. And this would be a machine that fits in a single room, costs single digit million dollars to buy, and then give it like another five years, it could out compute the entire universe. What does that mean? I mean, it's mind bending. And this was David Deutsch that first came up with this analogy that said if you took the entire matter of the universe and reconstituted it into the best computer one could ever imagine, designing it still cannot solve certain problems than these computers could. So where does this get applied? Scientific computing optimization, especially route optimization, protein folding, and scientific computing in general in a whole class of problems that map to what could simply be thought of as optimization. The traveling salesman problem. Finding the ideal solution to a multi variable constrained problem is difficult. You can do random walks, you can do heuristics and try to find the best path through a particular maze. But to be able to just jump to the answer near instantaneously is unprecedented. And so once these machines show a capability to exceed a normal computer, which some are calling quantum supremacy, to solve a problem that normal computers can't, it's not as if you just wait for Moore's law and there'll be a back and forth. It's like there's no looking back. Like every year these machines add resources that are exponentially more powerful than a traditional computer as opposed to linearly. And so we're just at that cusp of a crossover point where quantum computing moves from being not interesting to ridiculously interesting for a wide variety of problems.

All right, so I just have a practical question, which is what's the probability of running traditional software on quantum computers versus do you have to write the code from scratch through a completely different filter?

That's a great question. And that is the biggest hindrance. So that's an important caveat in oh, gotcha is none of the software we now know and love is going to just map beautifully onto one of these systems. In fact, almost everything you're currently using a laptop for. You would not think to use a quantum computer for the place where you'd see it, presumably in some cloud service thing, you know, like Amazon Web Services and Google. Google, by the way, later this year is going to avail a whole series of quantum computers that they're working on. The company I'm involved with, D Wave Systems, also has a bunch available online people can play with today. But your point is well taken. You have to program them and historically you had to be a quantum physicist to know how to program them, which is not. It's a non starter. It'd be like buying an intel chip and having to understand how to set, you know, gate oxide levels and, you know, voltage thresholds across gates, know how.

To think completely different.

Historically we're at the cusp of some software layers that are trying to bring abstractions to that. So if you're just a machine learning programmer and you think in terms of training a black box in an iterative algorithm, there are some overlays that are just now being developed to try to provide a software layer so that folks that are just normal programmers could start to use the assistance productively. And that is starting to come out and will.

So I might be able to offload certain specific functions for simulations, et cetera.

Co processor. Kind of like we initially thought of an Nvidia GPU for. We'd be like, oh, let's do the graphics off chip. But then all the core computations on chip. This would be a similar. Let's just farm out this massively hairy computational chunk like protein folding calculation or route optimization or you name it, Monte Carlo simulations. It just goes on and on. Just take something that the scientific community knows well and farm it off, get the answer back and then integrate it into an overall compute interface that you're used to.

So there's a huge industry that can be created just in doing that, for example.

I think so.

The opportunities are incredible. So for.

And unprecedented. In other words, Moore's Law, you want to count on that keeping going like for the indefinite future. But this is almost like Moore's Law compounded would be a way of thinking of it.

It's so, so exciting. I was just talking to someone about the how challenging it is just to do biology simulation, for example, because there's so much chaos and it seems like quantum computing would be perfect for that.

That's exactly right. In fact, that's why I first invested in 2003. So when you said I've invested a lot. It's been on the board of directors of this company for 16 years. D Wave. And when I first invested, it was chemical simulations that basically you can see a one to one mapping between a quantum variable in a quantum computer, many of these qubits as they're called, and one of the state variables in. They call it a wave function. But basically when you're modeling, let's say, how does a small molecule act, how does a particular material bond to another material? What's going on literally in the quantum chemistry of it. There's a 1, 1 mapping between the simulation you can run in a quantum computer and the real world because it is a quantum mechanical world. And so the poetic beauty of a quantum computer is that it simulates a quantum mechanical system perfectly. And it is very difficult for a normal computer to do that. I mean, we can't even computationally predict how water will behave with a normal computer.

All right, so I want to tackle one more big question here before we have to finish up. And that is through a lens, you have an unfair advantage. Access to Incredible opportunities, resources, money, technology. And you've been playing the long game. You've been really involved in investment, technology, software, everything revolving. And then now space and of course automate or vehicles for the average listener, the business owner, looking at opportunities and how certain technologies, AI are going to affect ordinary business. What are a few ideas that you can share right now? And you say, pay attention to this, this and this. It's going to affect all of us. And if you are prepared for it, if you're playing the long game, you're going to have an unfair advant.

Sure.

How would you answer that question?

A couple jumped to mind. The first would be if you were going back to the graduate and you had just one word, you know, like pay attention to this. It would be machine intelligence or machine learning or deep learning, basically all synonymous terms for a new way of building complex systems, a new way of doing engineering. And for many computer scientists, and this is kind of ironic, the domain practitioners, the ones you'd expect to be on top of this, it's a bit foreign, it's almost like a different way of thinking. I draw the analogy when I was on stage here earlier that it's more like parenting than programming, where it's a little out of control. You don't understand the thing you just created, but you can understand the process of creation, you can make a better one next time, but you don't know the thing you've made because you've basically evolved it, you've trained it, you've iterated it into existence. The thing I would share though is I actually started a PhD in this field back in the 80s and then abandoned it. And I can just tell you it is actually remarkably easy to learn. So if I was going to tell a small business owner or technologist anywhere out there, just spend a little bit of time, watch some YouTube videos, maybe take a Coursera course, Udacity course on this field of machine learning. It's remarkably simple to learn and once you understand some of the precepts, you at least understand how it's likely to interject your business, given the data that you have, in a way that you want to make sure you're not blindsided by it. And the more maybe provocative statement I would make is that if you were starting any new business today, to not have a machine learning element to your business model would be as bizarre as to ignore mobility or the Internet. Be like this possible that you'd start a new tech related business, or frankly any business, and pretend the Internet doesn't exist. Mobile Phones don't exist and have a really good idea. Really unlikely though, right? I can't think of one. Off top of my head, it will soon be obvious that to do such a thing without having taken advantage of the benefit of the learning loop of deep learning would be absurd. So that's one. Pay attention to that. I think if I was to make an even more provocative end statement here. I believe deep learning, machine learning is the most important advance in how we do engineering since the scientific method itself. It's essentially a process event. How do we learn? How do we build complex systems? How do we transcend our own intelligence eventually to build things that are smarter than ourselves? And we are already creating superhuman technology artifacts in let's say, vision systems and auditory recognition systems in medical information systems that exceed any human today routinely and done by non practitioners of the art. So that's amazing. That'd be one. The second broad thing that most people haven't factored into their business planning is, is the near term ubiquitous availability of broadband data globally, specifically by satellites. So SpaceX, OneWeb and a bunch of other companies, actually scores of other companies, are working on launching thousands of satellites over the next two years to blanket the globe with Internet broadband connectivity. So every remote village, airplane, boat, oil platform, Internet of things thing out in the middle of nowhere will have gigabit ethernet access, gigabit access to, you know, frankly, low latency Internet connectivity, unlike any satellite data you've heard before. This is going to be just as good or in some cases better than fiber, optical fiber. And it's going to be everywhere. And so if you're a business that's trying to reach people, realize the doors are going to be flung wide open. And you saw the first wave of Internet fiber and connectivity, enabling cloud services and all this for the developed world. The rest of the globe is going to catch up. So if you've got a product or service that makes the most sense for a poor performer in Africa, and it's been hard to reach them, realize that's all going to change. Starting those businesses now with an eye to the connectivity that's coming down the pipe would be potentially forward looking great.

And if you were going to make a recommendation on some platforms that are useful, I'm just thinking about from a marketing perspective, sales, internal automation, some platforms that employ machine learning right now that you can subscribe to or use. Are there some companies that pop up in your head that are worth investigating for our listener?

Well, gosh, I wasn't expecting that question. So off top of my head fetcher AI jumps to mind just because I know it quite well. It's a company that lets you recruit folks, you know, basically hiring and it does the simple AI matching between a job description and all of the profiles that exist on LinkedIn and GitHub and Stack Overflow and you name it. Right. So it's like matching for recruiting and does it at like, you know, 100x cheaper than a recruiter both in house or outhouse. Sounds wrong. You know, external recruiter. There's others for loan processing company called Capital is working on this. It's not yet soon to be announced but basically how about just in time loan approvals where they look at your cloud based data sets. Right. Look at your, you know, basically what do you have online in terms of your systems of record your sales system, your ERP system equivalent with Netsuite or whatever it might be and say immediately, oh yeah, we'll loan you money and here's the rate. And in any given day you can see if your sort of fiscal health is going up or down.

So you don't have to wait for weeks.

Exactly. And it's no longer opaque. It's like, oh, here's why I'm being declined or approved. It's also being used through mica lending and what have you basically everywhere within any decision making stack. To not use it would be bizarre. Right. So I guess I could go on and on. There's like it's hard to think of a sector that's not being touched by it. So if you can find a sector that's not being touched by it, that's probably good entrepreneurial activity.

Right, Right. Well, I think part of my goal is to start the rabbit hole. You know, let's, let's throw them down a rabbit hole. Check it out. Look at this. And we've got a couple resources that I'll make sure are available in the, in the show notes.

That's right. We can add some more over time as well.

That sounds awesome. So I know we've got to go because it's time to meet our governor.

That's right. And Kevin Newsom has something he wants to announce. We don't know what it is yet.

Right on. It ought to be exciting. So maybe as we wrap up where what's the best place to follow you through social online, Visit your site, etc.

Absolutely. Our website is Future Ventures. I love that new domain name Ventures. I'm on Twitter, it's Future Jurvetson. And then my last name, Jurvetson Also Facebook username Jurvetson and on going way back we had the Flickr founder on stage just a moment ago. I've been a Flickr user since 2004. That's sort of my long form blogging format. So I'm a username Jurvetson on Flickr as well.

Easy, easy. Well, I'll make sure all those are in the show notes. Definitely want to follow up with you for a follow up conversation. Go deeper on whatever is most popular with our audience. I know there's plenty to talk about, but thank you very much Steve.

Thank you so much. That's fantastic.

All right, that's it for this episode. But don't go anywhere because my co host Dan Sullivan and I have a really easy ask for you. Will you open up your podcast app and give us a five star review and leave a comment about what you love about it most? Dan and I read every review and it'll take less than a minute. And while you're at it, share this episode or tell someone about it because the best way to go grow an audience is by word of mouth. Now if you want detailed show notes, photos, links to all the cool stuff we talked about, or want to ask a question, have a show idea, or want to leave a voice message for Dan or me, just head over to capabilityamplifier.com for all this and lots of free goodies including copies of our best selling books. This is Mike Koenigs and so on behalf of Dan and me, thanks for subscribing and listening and we'll see you on the next episode.