Self-Driving Cars Might Finally Be For Real This Time

0:00You're being sold an AI future where you're obsolete or irrelevant. That vision is wrong. At Palantir, they're building AI that helps workers and unlocks their full potential. American workers are our nation's greatest strength. AI shouldn't eliminate them. It should elevate them. Palantir is here to tell their stories. From factories to hospitals, AI is freeing people from drudgery, letting them do what humans do best. Create. Solve. Build. Palantir, making Americans irreplaceable.

1:01mobile devices, message and data rates may apply. JPMorgan Chase Bank N.A. Member FDIC. Copyright 2025, JPMorgan Chase and Company.

1:22Hello and welcome to another episode of the Odd Lots podcast. I'm Joe Weisenthal. And I'm Tracy Alloway. Tracy, remember the hype about self-driving cars from like 10 years ago? Like that really died out. Can I tell you something? Yeah. I'm still holding out hope for the self-driving cars because I can't drive. And it was kind of acceptable when I was in my 20s. But now it's starting to get a little embarrassing. So I really need the self-driving cars to become viable options. So when we go on the road and like do podcasts like in another city. You have to drive. I have to drive all the time, don't I?

1:59Because I think I've asked you, I'm like, oh, Tracy, you're renting a car, and then you sort of change the discussion or you bring up something else. Oh, I think there's Ubers in that town or something. But this is the real reason, isn't it? Well, I mean, there are Ubers everywhere. You know what we should talk about? Whether or not Ubers have decreased enthusiasm for self-driving cars as a business model. Remember when people used to say Uber would never make any money if they still had to pay humans? But now they're making a little money. It's true. But I do think, like, generally, like, when people talk about, like, tech that didn't live up to the hype.

2:30And you see it now with, like, chatbots and stuff like that and whether they're really going to change. Like, people go back to the self-driving cars. Like, to me, that's the sort of quintessential example of the, like, modern times. Maybe 3D printing. You don't really hear that much about it. Yes, but don't you also find it weird to imagine a future in 200 or 300 years where there wouldn't be self-driving cars? It feels at once both inevitable and like hype, if that makes sense, like an artificiality. No, I mean, I definitely agree. 200, 300 years, that's a long time, like 50 years from now. You think 50 years from now?

3:07Well, so this is the question, which is like my, it's not an area I know that well. But my impression is it's like sort of classic thing where like tech got us like 95 % of the way there. and then there are some edge cases that make self-driving cars difficult. I don't know exactly what they are, but that getting that last 5 % or whatever is so hard that it renders the whole thing very difficult and that whatever that last percent is, is the difference between the tech being like, wow, versus actually changing the world. We are so close and yet so far. Yeah, it's one of those things. And I feel like, again, with chatbots and some of these other current artificial intelligence applications, it comes back to this question of like, yes, it's really great.

3:48and it sort of blows your mind, but are these hallucinations, another thing. And like, if it's not 99.9 % reliable, if it's not 100 % reliable, does that mean it really won't be as disruptive as people expected? Well, the other thing I'm curious about is whether or not that sort of last 5 % that you're describing, whether that's on the software or the hardware side. Oh, yeah. Because I think that has implications for, you know, if we do make huge leaps in artificial intelligence, maybe that solves a software problem, but maybe the issue is actually that But the sensors are too basic or too expensive, that sort of thing.

4:23I don't know the answer to any of these questions. The one other thing I'll say, too, is like there is a lot of car talk these days. We've been doing more and more on the podcast. It's entirely like on the sort of EV charging side. How are EVs and EV production and batteries, how are they going to reshape the industry? Or Chinese exports, how are they going to reshape the industry? It does not seem like, again, 10 years ago, the big question was like, who's ahead in the self-driving car race? Google, GM or Ford? There was much less talk then about EVs as the big disruptor. Right. So I think it is time for a checkup, right, on what's going on with self-driving cars.

4:59It is time for a checkup. And our guest says they're back, that they're happening for real. And I do believe him to some extent because I follow some people who live in San Francisco. And they're tweeting about it more and more that they see them on the road. and sometimes when I'm up at four in the morning to read the internet in the dark and drink coffee I see like people who are still out at night in San Francisco talking about all the self-driving cars around them so there might actually be it may not be totally over there might be they might be back well the things I see on the internet about self-driving cars are those edge cases where it's like a car flummoxed by a traffic cone in the middle of the street which they're simultaneously like impressive and amusing and disappointing all at the same time, if that makes sense.

5:42It's interesting. You're very pro self-driving cars. It hadn't clicked, but you really want this to happen. I have a personal self-interest in not having to learn how to drive. I figure if I'm super optimistic, maybe if I just hang on for like another 10 years, maybe. I don't know. Let's ask our guest. Let's ask our guest. We have the perfect guest, longtime tech journalist, a tech understander, someone who really delves deep into technology to understand how things work and what's really happening. I've followed his work for a long time. We're actually colleagues together 18 years ago, I think, at a site called TechDirt.

6:20Tim Lee, he is the author of the UnderstandingAI.org newsletter, longtime tech journalist, and he recently wrote a piece, The Death of Self-Driving Cars is Greatly Exaggerated. So, Tim, great to have you on the show. Hey, I'm great to be on. I'm a fan of the podcast. Thank you very much. Appreciate that. Let's start 10 years ago. And, you know, I think 10 years ago, there was a lot of self-driving car hype. And my impression was, and this is so vague and fuzzy, it's like, oh, most of it's solved, but this last part's really hard. Is that true? What was that last part that has proven to be very challenging to like turn these from like prototypes on a track or a very like organized grid-like suburb in Arizona to something that could actually be used on the road?

7:06So it is true that about 10 years ago, Google was the leading company, and they had vehicles that could go on certain routes with a fair amount of kind of preparation. And about six years ago, Google rebranded itself as Waymo, its self-driving car project as Waymo, and started testing a taxi service in Phoenix. and they've been plugging away at that ever since. There were a bunch of other startups that were started between about 2014 and 2018, say, and a lot of those failed or were forced to sell to some of the tech giants. And so there's many fewer companies operating in this space than there were five or six years ago.

7:45In terms of what the last little bit is, it's just a lot of little things. I mean, that's the thing about a long tail is there's a lot of stuff out in the long tail. One thing, for example, that Waymo and Cruise, the kind of industry leaders have been struggling with is when you deal with first responders. For example, if you come up to an active fire site, you're not supposed to dive over the hoses that firefighters are using. I mean, that's something you might only encounter every 100 ,000 miles or something. And so there's just lots of... But it's a really big deal when you do it. Yes, absolutely.

8:13There's another case where a cruise vehicle like drove through police tape in a crime scene. So there's lots of little things. I saw a headline. I haven't actually looked into this yet, but apparently a cruise like drove into wet concrete. So the real world is complicated and there's just lots of weird situations that a human being, because we kind of understand how the world works, we see, oh, that looks like we're in concrete. I shouldn't drive on that. But you just have to like, it's like whack-a-mole. You have to like hit every single like bad thing a vehicle could do. And that just takes a lot of, a lot, a lot of work.

8:40Yeah. I don't know why, but I find all the stories of like robotic self-driving cars behaving badly, absolutely hilarious. And I, not the ones where they hurt people. I should just caveat that, but the ones where, you know, something that we wouldn't even think about, you know, There's an object in the road, just go around it, and they seem to really struggle with. I want to ask you more about why that seems to be an issue and sort of get into some of the edge cases that Joe mentioned in the intro. But before we do, here's a basic question. Why have a lot of these self-driving car companies struggled?

9:13Because on the face of it, it would appear that there is a lot of money floating out there in venture capital land that often goes into unrealistic or unprofitable projects. So why has this been an issue for self-driving cars in particular? I mean, I think on some level, the basic issue is safety. A lot of other areas of tech, you kind of build a minimal viable product and you put it out in the world and it breaks sometimes, but that's fine. That gets you more feedback. And because you can kind of iterate rapidly, you can scale up very quickly and get to a profitable scale pretty quickly. That obviously doesn't work if the moving facet breaking things is like literally breaking things and killing people.

9:53And so you have to be very close to perfect before you can launch a commercial service and start making money. And so you had a bunch of startups that were trying to do this. They had all sorts of strategies to do that. Some were trying to operate in retirement communities or do like package delivery. They tried to find kind of less demanding applications than like drive of anywhere, anytime, but it's just really, really hard. And so the companies that have sustained are the ones that have Amazon, Google, GM, like big companies behind them who are willing to put like a billion dollars a year behind them for several years in a row while they kind of try to iron out these final little wrinkles.

10:28So zooming forward to today, and that makes a lot of sense. I hadn't really thought about that. It's like for many tech, it's okay if there are edge cases where it doesn't work because you just sort of like, well, you put out in the world and like, yeah, it's not a perfect product, but it's a minimum viable. It's free and we're refining it. It's free and we're iterating, but you cannot do that when there's big safety issues. And if it's a threat to other drivers or pedestrians, it's not really an acceptable way to do product. Going to today, and you are more optimistic, and we'll get to that about the prospects for their existence, but has there been a breakthrough in recent years?

11:06Or has it just been this slow iterative grinding away at the edge cases that makes it so that there are fewer and fewer edge cases? I would say the second one. I mean, Waymo's technology has worked pretty well. They started doing fully driverless operations in Phoenix in the fall of 2020 and have just very gradually expanded that service. Now, Waymo, Just a week or two ago, they got permission from California regulators to begin operating commercially in San Francisco after a year or two of doing practice driving there. And so, yeah, they've just been plugging away at it. And it's hard to tell from the outside because they're not super transparent about all the details of how many incidents they have or how much work they have to do on the back end.

11:51But yeah, it seems like they're just very gradual and making the technology better. And they seem to think because they're now talking about scaling up much more quickly, They seem to think that the companies seem to think that they're that this is ready for it to be a commercial product. Just a really simple question. If I were to go to San Francisco right now, could I go there and download an app or whatever and get it in a self-driving taxi? Like, that's it. I haven't checked that recently. So it was literally like last week or the week before the California regulators gave them permission to do that.

12:20But I think until like last week, I think there's a waiting list. But it's definitely a case if you go to certain parts of Phoenix, including the Phoenix airport, you can hail a taxi. And it's just like Uber or Lyft. You can go try it. I want to do it. I want to do it. Tracy, let's go. Let's go to Phoenix just for the one ride and then fly back. I'm sure. Yeah, and I've talked to people there. I mean, it works quite well. I've talked to several people who have ridden in those vehicles. And at least in most rides, they say it's flawless. It drives very comfortably. and yeah, the service, there just aren't that many rough edges.

12:51Can we talk a little bit more about the edge stuff? Because my impression is that, okay, computers learn from repetition and from modeling out various scenarios, but driving is such an infinitely unpredictable experience, especially if you're in New York. It's not that hard, Tracy. You could get it. Like if this technology had never taken off, I have confidence you could do it. I don't know. Yeah, I think I've missed the boat on that one. But anyway, okay. But there are all these different possibilities that a self-driving car could be grappling with. So, for instance, an animal runs out in the middle of the street and, you know, maybe after that happens several times, the self-driving car starts to realize, well, it's this animal and then it's going to behave in this way and keep moving or stop and I need to respond to it in a certain way.

13:41but that kind of seems to be the issue here as far as I understand it. Yes, absolutely. And there's a bunch of ways that the companies have tried to do this. So for example, Google has long had a big test track facility out in California about an hour. I went out there a few years ago where they have some fake roads and they'll create kind of fake scenarios. They'll have cars cutting other cars off or have somebody like moving boxes across the street, people in Halloween costumes, something like that. And so they try to think of what are all the situations that a self-driving car could run into and kind of anticipate that.

14:14And this is also why they started in Phoenix is one of their strategies was, okay, there's so many edge cases, we can't do them all at once. And so let's start in kind of easy mode. And so Phoenix has very nice weather, nice wide streets, well-marked, you know, not a lot of pedestrians, not a lot of bicyclists. And so that was kind of way most theory was that we'll do the easiest one first. The issue with that is that the economics of running a taxi service in Phoenix are not that great because most people already have cars. And so Cruise has kind of had the opposite approach, which is we want to see these educations as fast as possible.

14:46So let's start in downtown San Francisco, because that's where there's a ton of crazy situations. And so we'll kind of be harder in the first place, but we gather data very quickly, and maybe we'll master it. And it's not yet clear yet. I mean, both companies now seem to think they're ready, but I don't think we've seen them in the wild long enough to have a sense for kind of which of those strategies are working better. But yeah, it's really tough. And so I should say, like, for the first few years, both of these companies had safety drivers behind the wheel of every vehicle. And so the vehicle was mostly driving itself.

15:12But if it got stuck, the safety driver would have to take over. And the kind of big switching, the big risky point is when they take the first drivers out of the car, which Wimler did about two years ago and Cruise did, I think, maybe a year ago. And then, you know, and then it becomes much trickier because if the vehicle screws up, it's a big deal. I love the idea of having to train the self-driving cars by like putting people in Halloween costumes in front of them. And it reminds me a lot of socializing my dog because we used to have to like wear weird hats, right? Or like bring balloons into the house so that he would get used to them and not freak out.

15:46But this goes back to something that I mentioned earlier, which is, is the issue here the software? So like the actual modeling of the reaction to an unknown or unfamiliar event or stimulus, or is it more on the hardware side where maybe you need better sensors that are better able to appreciate the things in front of you? I would say it's more software and particularly more data, but yeah, the hardware has stayed pretty constant. I mean, the trio of sensors most of these vehicles have are cameras, radar, and then LiDAR, which is a laser range-finding technology that gives you kind of a 3D map of your environment.

16:26And so 10 years ago, Google's cars had those three sensors, And I think now those sensors are better, but I don't think anybody thinks that the main issue is that we need to upgrade in the quality of the LIDAR. Really, they just need they need examples of every possible edge case. And, you know, it's hard to get enough of that data because some edge cases happen very rarely, but can be very serious if you encounter them.

17:00Silicon Valley is selling you a future where you're obsolete or worse, identical. At Palantir, they're witnessing something different and revolutionary. From re-industrializing the nation's defense base to shipyard workers building faster and frontline workers boosting productivity, AI is transforming work across the nation. AI is not replacing American workers or flattening them into conformity. It's unleashing what makes each one irreplaceable. their judgment, their craft, their creativity. When American workers become more powerfully themselves, they own the future. Palantir, making Americans irreplaceable.

17:42Support for the show comes from public.com. You're thoughtful about where your money goes. You've got your core holdings, some recurring crypto buys, maybe even a few strategic option plays on the side. The point is you're engaged with your investments and public gets that. That's why they built an investing platform for those who take it seriously. On Public, you can put together a multi-asset portfolio for the long haul. Stocks, bonds, options, crypto, it's all there. Plus an industry-leading 3.6 % APY, high-yield cash account. Switch to the platform built for those who take investing seriously.

18:15Go to public.com slash market and earn an uncapped 1 % bonus when you transfer your portfolio. That's public.com slash market. Paid for by Public Investing. All investing involves the risk of loss, including loss of principal. Brokered services for U.S.-listed registered securities, options, and bonds in a self-directed account are offered by Public Investing, Inc., member FINRA, and SIPC. Crypto trading provided by XeroHash. Complete disclosures available at public.com slash disclosures. Can you give us a quick industry overview? You know, you mentioned Waymo. Waymo is Google. Cruise is GM. Cruise is GM, yes.

18:52And then obviously Tesla and Elon's, like, if you just like read Elon's Twitter feed, you would think that they've already had self-driving cars like in the wild. And I don't really think that's true, but I don't really understand what's going on. Can you give like a really just sort of quick, like, overview of who the big players are and like who owns them and just sort of like what their status is? Yes, absolutely. So Waymo is mainly owned by Google. Cruise is mainly owned by GM. I consider Tesla to be in a different market. And some of the Tesla fans get mad at me when I say this. But Tesla is building a driver assistance product.

19:24So pretty much any car you drive now, it has advanced cruise control where it stays in your lane and doesn't hit the car in front of you. In some ways, I think Tesla has a more advanced version of that. Although also in some ways, I think Elon Musk just has a lower risk tolerance. And so he's kind of pushing a technology that's... Anyway, but the key thing about the Tesla product is you are not supposed to crawl in the backseat and take a nap, right? You're supposed to be there making sure it doesn't break. Have people crawled in the backseat and taken a nap? I'm sure somebody has. There are videos of people doing inappropriate things while behind the wheel of a Tesla, but you're definitely not supposed to.

19:56And the vehicles, they have ways of monitoring the driver so that that doesn't happen. But anyway, so theoretically, Elon Musk thinks they're going to at some point get to the point where you don't have to be behind the wheel. But I did not think they're close to that or really laying groundwork. Because one of the things for any service like that is you need an operations staff because a vehicle is occasionally going to get stuck. And when that happens, it needs to be able to phone home and get kind of remote guidance about how to deal with it. And as far as they know, Tesla's not doing it anyway.

20:19So that's Tesla. And then the other two companies, there are a few other companies that I would say are a little behind. So Amazon has a company called Zoox that used to be a startup, but got acquired by Amazon a couple of years ago. And there's a company called Motional that is also, I think, close to being ready for driverless, but not to driverless. And then there's a company called Mobileye that supplies the hardware for most of these driver assistance systems. And they have been working on this technology. So that's another company. I say those four or five companies are the kind of the remaining players.

20:47Am I hallucinating this memory or was there a situation in which Uber hired every single member of the Carnegie Mellon University robotics department to work on self-driving cars for them? Yes, absolutely. That's a real thing that actually happened? Yes, that was – I mean I don't know if it was every member, but yes. Uber hired a bunch of talent in 2016, 2017, and then one of their vehicles struck and killed somebody in Tempe, Arizona in 2018. And that basically destroyed their program. And so I think the remnants. Oh, actually, I should say there's a startup called Aurora that is doing trucking.

21:25I think they acquired Uber's thing. But anyway, yeah, so Uber is now not a player in large part because they're really the only one of these fully self-driving programs that have had a fatality with their testing. So let's assume that self-driving cars become a realistic thing. How viable is that as a business model? Because on a first reading, it seems extremely expensive to develop, possibly extremely expensive to maintain if you have to provide operational support to all these robot cars out in the field. And then thirdly, it does seem like there's a big regulatory slash safety slash maybe legal liability risk if something were to happen.

22:10I mean, I'm pretty optimistic about it because you think about if you think about Uber and Lyft, about half of the cost of running Uber and Lyft is the labor of the human driver. And so if you take that out, then Waymo and crews need to get the new cost, the cost of the sensors, plus whatever operational stuff in R &D to be less than half the cost of the driver. And that's a pretty significant amount of money. And so I think it'll take them a while to get to the scale where it's profitable, because certainly Waymo and crews both have, I think, hundreds or maybe thousands of people working on this technology.

22:43And the sensors are currently pretty expensive. But one of the most predictable things in business is that mass manufactured technologies like LIDAR sensors and computer chips get cheaper at scale. And so I have no doubt that in the long run, this is going to be a viable business. And it's really, I think, a question of how much patience the big companies backing, you know, Google, GM and Amazon companies like that, how many billions of dollars they want to spend to get to this. But I think that in the long run, I think that the taxi industry will be operated by self-driving cars. And I think that in the long run, I also think it'll be cheaper and probably expand the market a lot.

23:17So my long run expectation is that this is going to be a big and profitable industry. Do you envision it just or primarily for taxis? Or could you have a situation where people like me are buying self-driving cars? Well, and just to add on to Tracy's question, because it sort of dovetails, could Tracy drive to work and then make some extra money by during the day when it would be parking for eight hours, have it be a taxi? And then could that impair total volume sales for the automakers because basically Tracy takes her self-driving car to work, but then also is serving the taxi industry at the same time?