The End of Human Driving? with Uber CEO Dara Khosrowshahi | On With Kara Swisher

0:00So if an AV is provably 50 times safer than a human being, do you think we should allow human beings to drive?

0:35It's on! the truth is almost everything Uber does is powered by AI. I've known Dara for years. In fact, I was the first one to tell him he got the CEO job at Uber. And since then, he's steered the company through tough times and into profitability. And now he's leading the charge as they move aggressively into autonomous vehicles. Uber has partnered with almost every global AV company in the world. And although the transition will take decades, Dara expects that eventually all Ubers and potentially all cars will be driven by software. I want to talk to Dara because he's very straightforward and very clear about where he wants Uber to drive to, get the joke, but it's also critically important to understand other issues around AVs like safety and the importance of AI in creating the ability for these cars to do the things they do, which is miraculous in many ways.

1:25This is the fourth installment of the Discovery Series on AI, which is jointly presented by Johns Hopkins University and Vox Media. It was taped live at the Johns Hopkins University Bloomberg Center in Washington, D.C. Our expert question comes from David Plouffe, who was the campaign manager for Barack Obama's 2008 campaign, a senior advisor to Kamala Harris's 2024 campaign, and a former chief advisor and board member at Uber. Stick around.

2:06Support for this show comes from me, Kara Swisher, and Johns Hopkins University. This week, I interviewed Uber CEO Dara Khosrowshahi in the fourth live episode of On with Kara Swisher that I've recorded at the Johns Hopkins University Bloomberg Center in Washington, D.C. In every episode, I've hosted timely discussions on AI, policy, copyright, and intellectual property. more as a part of the Hopkins Bloomberg Center's Discovery Series. Listen to my conversation with Khosrowshahi in this week's episode, and stay tuned for more live discussions to come from our partnership with Johns Hopkins University.

2:44Thank you for joining me here at Johns Hopkins University Bloomberg Center in Washington, D.C. for this special live conversation. I want to focus with you on a beginning with artificial intelligence, because Uber uses it in a myriad of ways. There's so many applications where AI changes the equation of a business. And one of the things that I think isn't getting enough discussion is applied AI, where it actually works in companies that are not AI companies. Everything from ride matching, route optimization, surge pricing, improving search results on Uber Eats. You're doing a good job, by the way.

3:22I like Uber Eats a lot. You become a platform. Thank you. Uber's also become a platform for workers to do AI labeling, and it's partnered with almost every global autonomous vehicle company. So are you an applied AI company or a company that uses a lot of AI? Because a lot of companies are trying to work their way through this issue. I mean, AI has been part of our genetics for a very, very long time. Now, AI is a big word. So in the time, it was like deep learning or machine learning. but you're right, which is most of the aspects of how Uber touch you in terms of the price that you get presented with, the price that we pay drivers, whom we match you up with, the routing that you take, even the algorithms that recognize your driver's license to make sure that you are the right person.

4:09All of that has been driven by AI. And so we have been an AI, applied AI company, I guess you call us. One of the cool things about Uber that I love is we're a technology company, but we're a technology company that operates in the real world. Like Amazon. Like an Amazon. Yeah, yeah. We're not just digital in scope. And so the application of that AI is it's a physical manifestation of getting a ride or getting your deliveries, et cetera. But we have long been driven by AI in all things pricing matching etc what's changed is one is there are AI applications internally for internal tools and systems and developer productivity if you want to get into that we can and then with the large language models obviously autonomous driving is a big big application that we can talk about as well but the ability to just build larger models now So, for example, if you're on Uber Eats, you have a sort that's unique to you based on your past behavior and just patterns that we know about you, what your favorite restaurants are, et cetera.

5:18And we don't want to just serve you those favorite restaurants. We also want to help you explore new places, et cetera. These models, as they get larger and as we get better silicon, are just getting more and more powerful and capable. So that our everyday AI use cases that we used, all of that is getting more powerful. Well, you're also collecting enormous amounts of data. So you have driver data. You have food data. You have interest data. You have all kinds of. Yes. We have a lot of data about how you live your life. We use that data essentially to improve the service for you so that we can be an everyday part of your service.

5:53whether you're going someplace or getting something, that's essentially what we do. So it affects every single part of our business. And you can debate whether or not there's a bubble, et cetera, in terms of valuations. What do you think of that? That's the obvious question. I mean, the spending is so massive. The spending on the data centers, et cetera, is massive. And I am happily not a stock market analyst. I used to be a banker, and I left that a long long time ago i can tell you though that the the value creation not in the space age stuff but in the very practical stuff of hey if you uh if you are shopping with us and you choose oat milk what's the next thing that we show you these larger models are enormously more effective than the last generation of models and they are translating into you know hundreds of millions of dollars of benefit to us.

6:48And so the spend for us in terms of AI has been well worth it and then some. Right. But you're riding on their rails for the most part right now, which is like you and Apple. There's other companies that are benefiting from the massive spending, but not necessarily spending. We are not building the picks and shovels, but we're riding on top of that spend. Absolutely. And the impact that we see on our business is very, very positive. Where is the most of the impact when you think about it? Is it having all this data or being able to get insights out of the data when you're thinking about AI? So I say, honestly, right now, the area where I would say it's got the most impact is actually our developer productivity.

7:30So 80%, 90 % of our developers, and they are by far the most expensive talent that we have in-house. they are using uh ai uh developer tools uh like cursor and and many others but it's not just the developers and coding but it's checking the code code documentation on call you know we we have hundreds of engineers that are on call and you know we operate in 70 countries 15 000 cities something is going wrong somewhere all the time and so you need engineers to be on call all the time to fix the issues. And essentially, now we have AI agents that are on call. And if there's something wrong, whereas previously the engineer had spent hours and hours, you know, you got these calls together, 20 engineers, what's going on?

8:20Where is it going on? These AI agents are constantly essentially looking at all of our systems. And then they come to our engineers with a hypothesis. Something went down here, pricing error, here's a hypothesis. And then the human can look over the shoulder of the AI agent. Does that mean fewer people? Is that the... It could. Now, my attitude in this is... I mean, I had someone, you and I both know well, at a travel company saying, oh, I had 5 ,000 quotas, now I shall have 1 ,000 next year. You know, the good news for us is we're growing really fast. And so we're growing top line 20, 21%. My attitude is if an engineer can be 20%, 30 % more productive, you could take one view, which is, well, then I need 20%, 30 % less engineers.

9:08I just think they become superhumans. So I want more engineers. So we are actually hiring more engineers because every engineer got more valuable to me. And to your point, most tech companies that I talk to right now are using the opportunity to essentially keep heck out flat. You know, I can grow the company, but I don't need to grow headcount. So basically, margins increase. You know, you see the technology margins are out of its world, and it's because people can do more with the same number of people. We're hiring right now. So where doesn't it work? So most companies are trying to figure this out.

9:43You've been deep into AI. But I talk to a lot of non-tech companies, and they just buy it because they think they have to have it. And then they're like, I'm not using it. Our people aren't using it. We build it. And I'll tell you, it can be hard. So one example, and there's a little involved, but customer service is another area that we use AI for. We have hundreds of millions of customer service interactions, whether if you get your delivery late, et cetera. And we've automated some of that stuff. But when it gets to an agent, if you break down what an agent does, agent, first of all, has to understand caracals.

10:24is she a good customer or is she a fraudster we have a lot of fraudsters you're a great customer terrific good customer um she said that her delivery was 30 minutes late uh that's your claim the agent then goes against our own fact base and you know there's an order place there was a courier who delivered it uh there was a restaurant who prepared it said it's ready was it really 30 minutes late or not okay it was 30 minutes late Kara's a great uh great customer Based on that, what's our policy? Yeah, 30 minutes late. Hey, you get your money back because you're one of our best customers. And then last is, how do we communicate that to you?

11:00Each of those elements, what kind of customer are you? What was the claim? Translate the claim into something that can be solved by logic. Compare that claim with a reality on the ground. Based on that reality on the ground and the claim, what is the policy that we have? and then how do I communicate that to my customer? All of that can be powered by AI. And essentially we are, and we will take a foundation model or a cheap Chinese open source model that's in the open source and we will use those models to build all these solutions. And initially what we're doing is, for example, using that to empower the agent.

11:46And so the agent, instead of like doing all this work themselves the ai agent goes and does all the work and comes and says carol's a good customer this is what she claimed um we thought this was going to be great right uh five percent of the time give or take the ai makes a mistake because they do hallucinate sometimes and if they can't find the data they'll make something up so now what we have is the agents are reading the ai recommendation they don't trust the ai recommendation so they go do their own thing. So as opposed to like saving time, it's actually the agents like kind of doing double work and it was not a net benefit.

12:24So that's an example of our trying it, it not working now. We are now going to more of a pure AI solution, which is actually having AI solve the problem. Oh, so they made mistakes before. So you'll give them the whole thing. It's actually interesting so you know we we have customers we we are having them solve problems where the cost of a mistake isn't very high ah i see and then uh and then we will learn from that and then eventually i say that to a lot of people eventually a pure tech company can do what we're doing which is like we're putting the work figure it out iterate iterate iterate a traditional company is not going to go through because it is a journey this has taken over a year and they don't know what to do about it it's just it's just very hard to actually translate this stuff into the real world and and what's funny is that the latest that that we've had is if you ask the ai to follow rules um sometimes it makes mistakes we're now kind of freeing the ai and we're like treat your customers well you know go check what's going on we give them very general guidance and that is turning out to have the best results early on.

13:37So we actually throw away the rules and we give them like, if you're a smart person, as I care, use your common sense, here are some guidelines, here's what I'm trying to do. The AI seems to be working really, really well there. Yeah, until it kills you. No, I'm kidding. You said one of the things that I'm interested in is autonomous vehicles. And for many people, AVs might be the most visceral form of AI they I think it is a visceral feeling when you get into an autonomous vehicle. For those of you who haven't tried, definitely try it. We work with over 20 partners now, both in terms of our mobility business, but also for delivery and for trucking.

14:19And what we're finding is consumers love the product. It is there's something to be said for privacy. Like I love talking to my Uber drivers and all that and getting their stories. But it's also nice being alone in the car. The technology is working. And it is, I think, generally the industry is taking a safety first approach, which is wonderful. Well, some people are. Yeah. I mean, I hear some undercurrents there, but I won't comment. I mean, I think generally the industry is taking. It needs another fucking camera on those cars. But go ahead. I think the industry is being appropriately careful because the cost of.

14:56Waymo certainly is. A cost of a hallucination in this business is a disaster, right? For you too, if you're affiliated with it. Of course, of course. And that's why we have our own safety protocols. And we have to make sure that our partners pass those safety protocols as well. But our view is we will work with the AI ecosystem. And just like I want every great qualified driver on the platform, I want every great qualified robot driver on our platform as well. um and so we're taking a bit of a platform approach and i'd say it's very very early um you know we're doing millions of uh rides run rate a year av rides uh but we do you know 35 million rides a day yeah you know so it's a very small percentage but it's absolutely coming the experience is delightful and and it is it's a great example of how ai really can change how you live every day.

15:57Because the pricing stuff, it's like, cool, but who notices? It doesn't make a difference. But these cars are pretty incredible products. Yes, they are. So Dave Risher, the CEO of Lyft, recently said there was zero likelihood that self-driving cars will replace human drivers anytime soon. He thinks the technology isn't ready, including the AI. AVs are expensive and riders, well, they're mixed on human drivers or not. It depends on who you are. However, you said that AV is the single greatest opportunity for Uber. Why do you see it so differently? How do you look at the different feelings and why are you so bullish?

16:29And I will reference that quote that I got out of Travis Kalanick many years ago, where he said the best thing for Uber will be to get rid of the guy in the front seat. It was kind of cloddish at the time, although I appreciated him telling the truth. But talk a little bit about why you think it's an opportunity. So, you know, number one, I would say the opportunity is safety. I do think the industry overall understands that the cost of a mistake here is enormous. And the AI driver will be safer than a human driver. They don't get distracted. They don't text. They don't get tired. And the AI algorithm is getting better all the time.

17:09You and I are driving isn't going to continuously get better. The AI drivers are going to get better. They have better. And they've got cameras, lidars everything so one is this is going to be safer method of transportation over over a long period of time that that's a period um right now the business has to scale up so the manufacturing of these cars is in there at scale the cars are super expensive eventually um it will also be a cheaper form of transportation now that eventually is 10 to 15 years from now uh and i think david to some extent is right in that you know by 2030 maybe in the u.s this is 10 percent of for a mix people would prefer what do you want yeah i mean our belief is is is actually like a hybrid network is the best way forward because it allows you to responsibly transition over um to av over per of time.

18:10And it allows us to identify there are certain customers who are going to self-select. I want an AV. And there are certain customers who are very, very happy with the person. And our kind of a hybrid network allows us to keep humans busy. But then essentially, as a transition happens, as we get more and more AVs on the road, introduce the technology to more customers. But because of the advances in AI, AVs are more likely to win out, correct? I think eventually AVs will win. Now, I think it'll be 20, 30 years from now. But incrementally, you're going to see AVs increase, and they will become more and more part of our everyday lives.

18:53And it won't be like, oh, my God, this is incredible. I'm getting an AV. And the experience that you have in Austin or Atlanta is you just choose an Uber. And once in a while, if you're in the city centers, and that's where AVs are starting, we'll match you up with an AV. We'll say, hey, do you want to take this AV or not? The majority of people say yes, not everybody. And the ones who say yes, rate the AV drivers at a 4.9. Well, nobody's there. Well, you know, it's a great driver. Perfect score. Exactly, five star. So you had tried to do this yourself. You had an accident or a self-driving Uber.

19:30Yeah, it was a terrible tragedy. Killed a pedestrian. You handled it. It was a difficult time, but the backup driver was also distracted and she was ultimately held responsible. But you abandoned attempts to do it yourself and you partner, as you said. Longer term, shouldn't you go back to that as difficult as that was? I don't think so because ultimately I think that the way that I see this technology going 10 to 15 years from now, every new car sold is going to be L4 autonomy capable. And I don't want to be in the car business. I don't want to be in the car manufacturing business. And there are many other companies that are developing this technology.

20:09So I can essentially benefit from the technology without the enormous investments. And the benefits that we bring to AVs are enormous in that these are pretty expensive cars. And we can essentially drive more business for these cars than anyone else in the world. And we can launch into 15 ,000 markets, assuming that the regulatory framework is there. Meaning Waymo is doing this to make proof of concept is what they're doing. They have their own reservation system. Yeah. And I would expect like a player like Waymo to listen they're always it's a great brand and so just like a Starbucks has their own app and has their own direct channel and they work with our marketplace because they want as many orders out of that box that's making coffees any AV company is going to want to maximize the revenue of that box with wheels but that doesn't mean that they're not going to have a direct channel.

21:10I fully expect Waymo to keep building their brand and hopefully work with us and continue to work with us and we'll build our business together. And many other AV providers will do the same. So you're partnering with a lot of autonomous vehicles, as you said, companies. You have investments or collaborations with Aurora, Lucid, Stellantis, Pony AI. You're reportedly considering partnering with Travis Kalanick, Uber's controversial founder and ex-CEO, to help fund his bid for an American subsidiary of Pony AI, which is a Chinese AV firm. Talk about how you're thinking about this, like in terms of partnering to just give you more choice or to lay bets.

21:49We want to help the ecosystem develop. And our investing in these various companies allows them to raise more money from external partners. It gives them credibility. And it gives us a better look into their tech development and their talent. So we get to know the product better. We get to know the companies better. We help them develop. And essentially, we are helping out the entire ecosystem because we want this technology to develop as long as it's developing safely. That's essentially what the goal is. So as the ride-hailing industry transitions in this way, it's a question of who bears the cost of owning cars, for example.

22:31Right now, drivers do. I know you used a quote that I wrote a column back in 2019 saying owning a car would be like owning a horse someday, right? Like horseback riding, essentially. But autonomous vehicles are much more expensive, as you know. So you've suggested that private owners will finance and operate large fleets of robo-taxis on Uber platforms. Explain that. Yeah, so it's— Meaning you don't want to own these fleets. We don't want to own those cars. Now, early on, as the marketplace develops, as the ecosystem develops, we will initially buy cars. So for example, our partnership with Lucid, it's with a company called Neuro, terrific technology company.

23:13They're the software driver. Lucid is building the cars. And we will buy the cars initially. Once we prove out the revenue model, and these cars, again, they don't get tired. They're going to do the average Waymo in Austin, Atlanta does more trips than 99 % of our Uber drivers. So these things bring in a good amount of money. They're working all the time. Yeah. Once you prove out the revenue economics of the car, then we will have financial companies. All this stuff is going to financialize. So just like you have these private equity and private debt companies that are financing these data centers and buying NVIDIA chips, that same ecosystem is going to be buying fleets of cars.

24:00And just like Marriott doesn't own any of its own hotels. There are these things called REITs, real estate investment trucks. Same thing. You're going to have fleets essentially. So it's the benefit of Uber in that case. Yeah, and Uber brings demand to these. And these financial owners are, again, they're going to want to maximize the revenue from the car. And the way that you maximize revenue from the car is put it on the Uber ecosystem because we've got more demand than anyone else in the universe. So as with hotels, a bunch of dentists will own the cars and then - You know, it could be. Yeah.

24:30I mean, I do think that there will be an institutional, it'll be big institutional players. And then I think in the suburbs, et cetera, it'll be smaller fleet operators who, you know, they'll buy 10 cars, five cars, et cetera. And we actually now work with 15 ,000 fleet operators, large and small, who are entrepreneurs in this business all over the place. So many of those autonomous AV manufacturers have their own, as I said, ride-hail platforms. You can use Uber to order Waymo, Robotex, and Phoenix, but in San Francisco, riders have to use the Waymo app. Tesla is starting to roll out its own AV ride share.

25:10It hasn't partnered with Uber. Talk about, is that a competition or will they eventually, because you do have more data, you have more information. I think it is competition and also coopetition, to use that overused word. And again, I'll use the food example. McDonald's is going to have its direct channel, and they're going to work with Uber Eats because that can drive the highest utilization. Ultimately, I think that's where the industry is going. There may be some players that are tech only that don't want to have a front end because they don't want to invest in a front end and they will work directly with us.

25:46But I think you will have an ecosystem of some players who want to build a brand and want to work with us. And hopefully we'll have everybody work with us one way or the other. We'll see how it turns out. So one of the issues is obviously regulatory, including around AI, privacy, safety, everything else. So every episode, we get an expert to send a question for our guest. And for you, we got someone who knows Uber very well, used to work for it. Let's play it. It's not Travis. Don't worry. Hey, this is David Plouffe. And my question for Dar is this. When I was in Uber, so many of the discussions with elected officials and regulators in the beginning was about whether ridesharing should be permitted.

26:22And if so, how? Usually having to think through how you force it into regulations that permitted taxis that were decades old. But the best conversations were the ones that happened with city officials, state officials, leaders and countries who thought, OK, we're going to have more people getting around the cities with ride sharing. How should we change? Should we reduce parking requirements for new construction? Should we think about how ride sharing connects to mass transit? So my question is, you think about the autonomous future where so much of the debate currently is still around whether to permit it and how.

26:54what do you think city leaders and state leaders and leaders of countries should be thinking about in terms of you know parking requirements obviously there's storage questions around these vehicles how should we think about changing our public transportation system if more people are getting around using autonomous in addition to current ride-sharing options so i'm just curious from a city design and vision standpoint what you think the best approach is to really get the full power of this new technology. I never got to work with David, but he has a great reputation. The issues that we are starting to talk to regulators around, and they're really good issues, are one is kind of accessibility.

27:35The other one is congestion. And I say that the last is the transition, all right? And just talking about each of them. You know, like very early on when And this is a little bit before my time, but I kind of came into it as well. When we were launching our service or expanding in various cities, there was this idea. And I think it's a really important idea, which is affordable transportation should be available to everybody. And the thing about taxis was taxis only serve city centers because that's where you have the most rides. And one of the magics of uber is because of the network, et cetera We are able to serve the whole city not just the middle of the city where the rich people live The outskirts, et cetera Especially transportation deserts where there is no mass transit, et cetera I think that's going to be one question and one dialogue that we're starting to have with as it relates to ride share And and autonomous ride share which is you see most of the av players start in the high-density areas where more wealthy people tend to live?

28:48Is that the right thing for society? You know, Waymo is expanding into the peninsula. They haven't expanded into Oakland yet. Yeah. Is that the right decision or not, right? Now, I can't make that decision for them. But those are, I think, important questions for society to ask. You know, if this technology is going to be introduced, Should it be for the next five years a wealthy person technology or should it be Made available to everybody we because we already have built Essentially density of both supply and demand in these areas I understand it's going to be hard for an av player To make a business work in kind of let's say less fortunate areas of a town Because we have density there we can actually do it affordably So we think we have a lot to bring, not just to the AV players.