Inside Cerebral Valley: Autonomous Vehicles & AI Investment

0:00This is brought to you by Forethought, building AI agents for every customer moment. Hey, it's Madeline from the Newcomer Podcast. We're fresh off the Cerebral Valley AI Summit in London, and we're going to bring you two of our liveliest onstage discussions into the podcast feed. Both touch on one of the biggest themes of the event, which was where AI will enter our lives beyond the chatbot, whether it's through voice, on a device, or even in our cars. First up, we'll hear from Uber CEO Dara Khosrowshahi and Alex Kendall from Wave, who are teaming up to bring self-driving cars to the UK. They actually rode in a wave self-driving car to the summit, which was really cool to see.

0:37The next panel is a conversation between some of the top VCs in Europe, Philippe Bateri from Excel, Jan Hammer from Index Ventures, and Tom Holm from Google Ventures about where they're seeing the next big opportunities for AI founders globally. Give it a listen.

0:56Hey, well, we're down to the last two, Due to the discussions I'm most excited about, so excited to be here with Dara and Alex. I, you know, covered all the Uber periods. I remember riding in an Uber self-driving car in 2016, pre-Dara, when it felt like there was a lot of hype, a lot of interventions, and then Uber had to pull them off the road because I think they maybe went a little too early. And now, you know, almost a decade later, I think I've gone from being a self-driving bear to being really excited about the opportunity. So I just want to start off. I mean, your two companies are working together to bring self-driving cars to London.

1:44Dara, you want to start off? Why work with Wave and what is the partnership going to look like here? Well, we're working with Wave because they're an absolute leader in the field. Our strategy at Uber is we are no longer developing our own self-driving technology. We want to really work with the ecosystem and look for the best and brightest who are developing self-driving technology. We've built a platform that brings incredible demand on a global basis. We can help these highly technical companies with operations, local operations, etc. and we can be a vehicle to bring this self-driving technology and introduce it to consumers.

2:25And so we have a rare view into all of the development that is happening in the industry, and Wave is unquestionably one of the leaders. I think they're taking, and Alice can talk to it, kind of a bit of a differentiated approach in terms of an end-to-end large model. And once we saw what they were doing, once we experienced the driving efficacy of their driver and how it can generalize, actually, Alex and I were driven here by a Wave driver from Wave headquarters. With a safety driver, right? With a safety driver. Any interventions? Zero. How was the drive? Perfect. Perfect. And there were some really, really complex maneuvers that we had to go through construction, trucks parked on the left.

3:20So it had to go around trucks, pedestrians in London being pedestrians in London. It was fairly extraordinary. I love all the look left, look right on the ground. You really need the instructions. But these guys have been a real leader. And so we very much wanted to work with them. We've invested in the company. and we're very much looking forward to bringing their technology to London and then beyond. Maybe you can talk a little bit about that. I mean, regulators have said, oh, you could do it next year. I'm sure you guys were supportive of that, but are you prepared to do it next year? How do you feel, how ready are you to deliver self-driving cars with Uber in London?

3:58Well, look, I think the thing that's very exciting for us is the confidence and clarity we have from regulators here. So we're working as hard as we can to get this launched in a way that scales, of course, in London, but really worldwide. And I can't wait to do that on the Uber network. But the clarity and conviction that the UK government has said, look, we want to make this possible, we want to accelerate regulation by a bit over a year and make this possible from next year, it's great for us because we can say, OK, we can commit to building this out here and it won't be a full service all of a sudden from day zero.

4:33There will be, of course, steps that we take in a controlled and responsible way, but ultimately... You know what neighborhood will be first? Oh, we've all got our favorites, but that debate's still ongoing. But I think being able to... What we enable with our end-to-end AI is, of course, to move away from the concept of high-definition maps. You don't need to map each region. Actually, the AI has the kind of intelligence to understand scenarios it's never seen before. My favorite experience from our drive, we went through this intersection where there was a car that stopped in front of us and we had to navigate, wait for an oncoming car on our left, safely change lanes to get around them and keep the traffic flowing.

5:10And all of this kind of complexity our AI can handle naturally. And I think it's not just about having an AI driver that gets you from A to B, but keeps the traffic flowing in a way that doesn't cause any road angst around you. I think that's what's going to be really loved by London is a new experience that fits in with the London driving culture as well as being, of course, safe. Are you using LIDAR or what's your philosophy in terms of sensors? We built an AI foundation model that can address all ranges of autonomy. So the car we drove in today was camera only, although the vehicles... What kind of car was it today?

5:42It was a Ford Mach-E. But the vehicles we target for the Level 4 RoboTaxi applications, they're going to have redundant sensors. So camera, radar, LiDAR. The key thing, though, is that they're not going to have a retrofitted spinning LiDAR on the roof, but integrated sensors that the manufacturers can produce at mass volume. And I think that's going to make this economically viable and scalable. I think just one comment on what's really cool about the UK and the regulators actually accelerating the timeline here is there are very few times when you have a true platform revolution the way that you're seeing with AI.

6:16And I do think that from a European standpoint, there could be the view that the U.S. has been the country that has benefited from, you know, the big platform changes, internet, broadband, mobile, cloud, etc. It has been the U.S. companies that have grabbed a significant share of the enormous opportunity that has been created as a result of these platform changes. AI, I mean, you've heard it. This is something that you're deeply in, represents such a platform change. And the talent level that we see in the UK coming out of the educational institutions is incredible. And so the UK has an opportunity to be a leader here in terms of this change and kind of the application of AI, generally physical AI.

7:06Are we at a point yet where cities, they should almost be bidding for the opportunity? In San Francisco, it feels like tourists are trying to go see Waymos. Do you get a sense that the mood in cities has pivoted at all to, oh, man, being first is a compelling pitch? We're seeing cities leaned in. Now, you should be leaned in at the same time. Safety is paramount. The cost of a mistake in the real world is so much higher than a hallucination in the digital world. So while they are leaned in, these are long dialogues that we're having with regulators to make sure that you can be leaned in, but safe at the same time.

7:51But so far, the conversations that we've had are very, very constructive, but we have to meter how we lean in here. I mean, what we see is that this solves so many problems for cities, but it's, of course, it comes down to the nuance of how you integrate and deploy this technology. and, of course, doing it in a responsible way is critical. How do you prove it to London, or what's the path to showing that you are safe? Oh, there's so many aspects to proving that. Endless regulation. And the key thing for us is, of course, we prove it before we deploy. And so, of course, that starts with very large-scale offline testing and simulation.

8:25It starts with proving that you can expose the AI to the right amount of data across the right behaviors, and then testing the integration. I believe the Texas approach is prove it after you deploy. We'll get more on that. Dara, I mean, you are partnering with other self-driving car providers. You're working with Waymo. Like, what do you think about sort of your partnership strategy, and how do you decide which partners to work with? Well, ultimately, we think this is a technology that holds the promise of making the streets of the world safer. That is by far the most important factor here. In the U.S.

9:01alone, there are probably 35 ,000 automotive fatalities per year. And if you take the opportunity that we have now, which is to hold a standard of safety of these AI drivers being multiple times better than humans, and Waymo certainly has demonstrated that that's possible, you get to have safer streets. And over a longer period of time, you can bring the cost of transportation down, increase demand, and hopefully take a chunk out of personal car ownership there. So that's what ultimately the opportunity represents. We don't think there's going to be a winner-take-all in the marketplace. And so we want to partner with the best and brightest.

9:42We get to meet everybody in the industry. We have a safety case as well that we want to make sure is fully satisfied. And Wave is just unquestionably one of our top partners that we've met. And we're just super, super excited about building the relationship. I mean, you're in sort of a funny position in that, like, a lot of people in this room and probably Alex all want the technology of AI to really be the differentiator in sort of pricing. And you sort of want to be the demand aggregation and sort of customer relationship. is the differentiator. Do you agree with that? Or what do you see as sort of your leverage in sort of deploying self-driving cars?

10:26I mean, listen, our leverage, if you want to call that, is that we're operating in over 70 countries. And so we can bring instant demand to any technology provider. And I think at least for the next five to 10 years, there's going to to be more demand for these robot cars, if you want to call it that, than supply. So 10 years from now, maybe there'll be competition, who's going to win, et cetera, and who's going to be the provider for various OEMs. But we are a long way away from that. And right now, it's about making sure that we introduce this technology safely in the streets of, you know, hopefully all the countries in which we operate.

11:04So I just don't think we're there yet at this point. Yeah, far away. Alex, there's a third mystery party in this partnership, which is the OEM. I don't know that you're going to give it up right here on stage, but what characteristics do you look for? Do you think it'll be just one? And how do you see the roles of the three different players? Yeah, that's a good question, Eric. I mean, we work with automakers around the world through Europe, Japan, and the United States. and again, I think this is going to be a technology that all OEMs will want to produce vehicles that have autonomy capabilities.

11:42There will still be some personal car ownership market and of course the ability to drive hands off and eyes off with that is going to be a really key thing for that product experience and that's what our product can enable but then the market is going to move to focus on level four robo taxis and other autonomous mobility applications and so what we are seeing is the very best automotive manufacturers are setting themselves up for that future. They're working with our AI, of course, building vehicles that are software-defined that enable you to... The amazing thing now is that we're seeing manufacturers bring out vehicles that you can get data off them.

12:13We can start to aggregate data, just like Uber aggregates demand. We aggregate data around the industry, and that enables us to build the most safe and scaled AI across the market. And then the vehicles, making sure that there are features you need to operate a robo-taxi platform. One OEM per city, or do you think there will be multiple? I think that consumers will want choice and we want to enable our platform to enable many OEMs. You think the cars can provide differentiation too? Absolutely. And I mean, if you look at our AI is flexible in that it can work with different sensors. So it learns what a sensor architecture can and can't see.

12:48But some OEMs may want to put a camera here or slightly in a slightly different position. Maybe they want to use a different compute stack. Or of course, the actual driving experience. The inputs to our AI is not just the sensor input, the navigation prompt, but also a style prompt. How do you want the car to drive? Are you going to be the scorekeeper of AI self-driving? Like how much do you think you'll, well, we can measure all the accidents. This one is slightly safer. We do have a good viewpoint as to the efficacy of the various AIs and the different approaches. And there are different approaches to solving the problem here.

13:23And obviously we're going to make our bet on the approaches that we think are safe. And then also, obviously, like you said, there's a third party in here, which is the OEMs. I do think 10 years from now, I think every car sold is going to come with an L3 or L4 package, depending on what you're looking for. I think it represents a huge profit opportunity for OEMs. I mean, like the OEMs are understanding the beauty of software, zero gross margin, obviously huge upfront costs in terms of development. So, you know, every OEM that we talk to is very, very interested in this technology. But at the same time, it's very different tech than they typically build.

14:06There's a big debate. Do we build it internally? Do we go with a third party, et cetera? But I think 10 years from now, every single new car sold, if it doesn't have this package available, it's not going to sell. And you think you'll be using sort of consumer cars as part of an Uber fleet? Well, I think that, yeah, drivers, you know, they'll drive our most popular features or cars is a Prius, right? And so there'll be a driver driving a Prius, whatever the next generation is. There will be a software driver. Hopefully it will be a wave driver driving that make of vehicle. Any thoughts on the Tesla rollout in Austin and what it sort of?

14:51yeah, I mean, you need to worry about, I guess, the overall reputation of self-driving. Sure. And how much do you think sort of any one player acting more or less recklessly will sort of blow back on the rest of self-driving? Yeah, I mean, listen, I think that Tesla has been an unbelievable innovator in EVs. They've been an innovator in developing self-driving software as well. And certainly, I wouldn't call their start reckless at all. I mean, it's a very small operational domain. And they have safety drivers. Yeah. I think they're being actually appropriately careful. And you see the challenges in real-life driving there.

15:32So at this point, again, like this is, it's a new industry. We think that it's great to have so much excitement around it. I think when people experience it, it kind of blows them away. Then they start taking it for granted. For like two minutes, there's absolute delight. And then they're like, you know, texting on their phone. I mean, I trust it. Yeah. I mean, I've been mostly in San Francisco Waymo's, but I trust them more than the average Uber driver. Yeah, yeah. I'm curious as to what you think. I mean, you're in it. So I'm curious. Do you trust them more than humans yet? I think that's going to be the future is that, of course, they are going to be in so many ways.

16:10They're going to have more intelligence and be safer than what we can do when we drive cars. But the operation today that we saw deployed in Austin from Tesla, I think the key thing from here is that this is going to grow. It's a start, and there's going to be iteration from here. The critical thing is that as an industry, we need to make sure that we iterate responsibly and have the right checks and balances in until we prove that the level of safety can be better than a safe and competent human driver. Like talking about technology, like a core problem that I think we've seen with AI is that like getting very close, getting to like 95 % feels awesome.

16:48But then it doesn't like, if it doesn't solve the case, then you can't get rid of the humans and then it's not as valuable because they're human drivers. I mean, well, one of the interesting things there is to look at the difference between cognitive and embodied AI and language models. Actually, how do you define safety is quite nebulous, and there's a lot of debate around what general purpose AI regulation and standards should look like. But that's not true in self-driving. We have very clear industry understanding of what safe is, of what expectations are, and actually the way we evaluate and benchmark these systems is very well understood.

17:18So I think there is a push in AI. We're seeing more of a push towards safety and benchmarks in embodied AI, and I think we're seeing innovation about how do we understand and prove the levels of these systems come out through robotics that may advance us as a field here. But I feel good about this in self-driving, and I still think that conversation is evolving in the language model space. But, you know, I'm interested in this beyond self-driving, almost the lessons from self-driving to some of the pieces of AI where it feels like we're getting answers, but then we're also getting hallucinations.

17:51I mean, in the Uber case, I mean, you came in, you did spin out Aurora. Like, how did you... We merged ATG into Aurora, yes. Yeah, and then became its own company. How did you decide, I guess, that self-driving wasn't there in 2017 or that this shouldn't be something to invest in? Or how did you go into sort of all your technologists saying, oh, we're just tomorrow, just tomorrow, and say, I'm not convinced? Listen, I think Eric was a really tough decision at the time. And one is we were deeply marked by the fact that we had an accident, lost life there. And so the responsibility that you hold in your hands when you're operating in the real world hit us deeply.

18:35I think the other circumstance that affected the ultimate judgment was that we ran ATG very separately from mainline Uber because we wanted the opportunity to partner with the rest of the ecosystem again. Like, we don't just work with, you know, we don't just have Priuses on our platform. We got forged. We want to work with the entire ecosystem. And when I went and talked with some of the other AV developers, I'd say, you know, ATG, it's in the family, but it's separate. We're going to treat you fairly. We're not going to share data, et cetera. You know, when you're having a conversation, actually, like I usually do with you and I'm saying stuff and you don't believe a fucking thing I say, pardon me, that was the response.

19:19Like we were having conversations, but I think people didn't trust us because they also viewed us rightly so as a competitor. Right. So we had a decision to make. Do we go with a proprietary kind of vertical strategy or do we truly partner with an industry? We couldn't have the best of both worlds. And ultimately, we decided that a partnership model was the right model going forward. and based on basically all of the leading players, sans Tesla, but everyone else is open, working with us, et cetera. I think it was the right decision to make at the time, but this was not a simple thing. It wasn't on a PowerPoint.

19:58It wasn't clear which way to go. But it was more about the ability to partner with other people than an assessment of the technology? I think it was, for us, it wasn't necessarily an assessment of technology, but it was assessment as to what we are great at. And Uber is a very strong software company, but we're not necessarily, we weren't great at developing self-driving. We were good, but I think it takes better than good to get there. Alex, I mean, this is not a monogamous partnership. I mean, he's talking about partnering with a lot of different people. How do you see the partnership strategy and how do you get comfort partnering with a company that's like, oh, we want to partner with everybody?

20:39Well, in a similar way, the platform that we're building is going to enable fleets and automotive manufacturers around the world to build autonomous products. But what we're seeing from Uber is leadership and speed that really outpaces other fleets and marketplaces. Actually, it's not just the demand. You think about all the aspects of running an autonomy service, how you own, operate, clean, collect the data to integrate it. There's so much work there to take a technology works and actually scale it. Now, like Dara, I'm a huge believer in focus, and I know that our expertise is in building the AI model.

21:13We don't want to build the car, we don't want to own and operate it. We want to enable this AI to be on as many robots as possible. And, you know, that's going to go beyond cars in the future, because I'm seeing things in manufacturing or other mobility spaces that are a bit like self-driving back in 2017. And, you know, it'll take some time before manufacturers figure out the right software-defined vehicle equivalent then. But when they do, there's going to be some manufacturers, like Tesla's a great example of this today, that can build their own vertical stack. But then there's going to be a wealth of manufacturers and operators that have the platform, that have the demand, that have the product expertise, but lack the intelligence.

21:50And the scale that we can build there as a platform by aggregating data and training a model that's larger than anyone can do on their own is going to unlock, I think, the broader autonomy ecosystem. So we see it in the same way, But I'm so excited about, I mean, Dara's leadership and the speed that Uber is pursuing here. I think we can move really, really quickly. Uber's worth$190 billion, which is great, right? I think like an all-time high. But Tesla, you know, is worth a trillion. You have to spoil the party, don't you? Do you think, I mean, how much do you think you're getting credit for your work in self-driving?

22:25And how much do you think the market should be saying Uber is a self-driving car company? I think we're getting zero credit at this point. But listen, when you're developing new technologies as a company, you have to be willing to invest aggressively for many, many years before the market understands or necessarily gives credit for what you're doing. So, you know, the market has bet against this before. $190 billion isn't bad. But I think if the reality that we see in partnership with Wave and others and self-driving becomes real, not just in passenger mobility, but other kinds of mobility delivery, I think $190 will just be a step-stamp for us.

23:14Alex, you know, there's a risk that Waymo becomes synonymous with self-driving cars, where a lot of people, technologists, are getting the Waymo experience first. Like, where do you see your differentiation with them? Like, how quickly do you need to move to sort of stay in the conversations they keep to deploy? And how do you assess what they have on the market today? Well, here's the interesting thing. Going city by city with a geofenced approach that relies on high-definition maps, This AV 1.0 strategy that the industry looks at is one play. But actually, I think this technology is going to see mass scale through personally-owned vehicles first with driver assistance.

23:54And that's how most people are going to get to experience it. And I'm not clear if you can actually build a general-purpose robo-taxi without that data, that manufacturing integration. and having our AI in millions of vehicles around the world for hands-off and eyes-off driving is going to provide the regulatory relationships, the manufacturing integration, the data, and, of course, the brand and the exposure around the world that we'll be able to take and use that to build an AI for a general purpose robo-taxi. And so I think that is an enormous advantage that we can actually get that scale first.

24:27And, you know, we're in it for the long game. We want to produce something that, if you think about what the future of robotics should be, it's not a robot that follows around infrastructure in affluent places, but we want to build a future where there's a level of intelligence that can drive in new places, it can, you know, respond to your demands, you can delegate tasks to it, it can interact with you through language. I mean, fundamentally, the way that's going to happen is with end-to-end deep learning, and that's the bet we're taking. And I think that, of course, time will tell, but I think that just like we've seen in many other verticals from drug discovery, game-playing agents, language models, robotics is going to be no different.

25:03And I think we're going to see that better lesson play out. So I'm all for end-to-end learning being the approach that does scale by leveraging the scale deployment in consumer vehicles. Dara, do you have a year where you think the Uber platform flips from human majority to self-driving majority cars? I am 100 % sure that five years from now, we're going to need more humans on our platform than we do today. Okay. Business issues are growing really, really fast. Ten years from now, I'm not sure about that slope of the curve. It's not going to flip. But the increase in drivers and couriers that we need on the platform may start metering out.