Tech Brew Ride Home - (Bonus) Self Driving Car Wagers With Timothy B. Lee @binarybits

Episode Date: February 3, 2019

Ars Technica's Timothy B. Lee gives me odds on my self-driving car wager. Sponsor: Eero.com/ride promocode: ride Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to another bonus weekend episode of the TechMeme Ride Home. I'm Brian McCullough. The motivation behind this one was simple.

Continuing to tease out where exactly we are and where we're at and where we might be in terms of self-driving cars. I wanted to reach out to someone that covers this space all the time. So today we're going to talk to Ars Technica's Timothy B. Lee, not only to figure out where we're at, but also because Tim came up with some super interesting angeles. about the reality of putting AVs on the road recently that I hadn't even considered. But first, okay, Timothy B. Lee from Ars Technica gives me odds on my self-driving car wager. Let's start with a little bit of basics. I keep reading that of all the players trying to make self-driving cars happen. Everyone says that Waymo is like the furthest ahead of that whole group. Is that correct?

Yeah, I think, I mean, it's hard to tell from the outside, but I think all the signs point in that direction. I mean, so way more obviously spun out of the Google self-driving car program, and so they have been working on this for several more years than anybody else. They started in 2009. A lot of the competitors you see started in the mid-20 teens, so they had that big head start. They've logged way more miles than anybody else. And in California, there's an annual process where companies are testing California. foreign to say how many miles they're different, how many times there's been, quote-unquote, disengagement, which are, you know, cases where a driver has to take over.

There's some kind of fuzziness about that because companies have some freedom to define, you know, what counts as a disengagement? So you want to take that with a little bit of grain of salt. But Google has done far more miles of testing and has had far fewer disengagement per mile than anybody else for several years in a row. So they've been doing it longer than anybody else, but are they doing it differently as well? Or is generally everybody kind of, is the technology sort of the same for everybody? Like, you know, we're trying to do machine learning to get the autonomy there,

and then we're using LiDAR and all sorts of sensors and things like that. Are they doing something functionally different with their technology than others are doing? At a high level, it's very similar. So pretty much everybody, except maybe Tesla, is using cameras, LIDAR and Radar, that kind of trio of sensors to understand the world. everybody's doing a fair amount of machine learning. I think there's some subtle differences in the kind of machine learning and the ways that companies use machine learning.

My impression is that some of the startups that are a little bit more started more recently maybe use machine learning for more of the autonomy stack, whereas the rumor is that some of Google's technology is a little bit more kind of hand-coded, where the human being is kind of figuring out what rules should the car use, but it's hard to know the details because obviously they're not talking about those details, and I would expect Waymo to be picking up kind of the best ideas that come from elsewhere. So broadly speaking, I think everybody's taking similar approaches in terms of the machine learning, the sensor, the computing hardware, et cetera.

So this conversation was prompted by a couple of specific tweets. I'm going to read the first one. You tweeted recently, until recently, my mental model of Waymo was that their technology was basically ready to go in late 2017 and that they were doing a last few months of testing out of an abundance of caution and to give them time to build out non-technical stuff like customer service and maintenance but you said that you recently changed your assessment of that what what made you change yeah so for all year well so so 2017 the big thing they did is they claimed that they had stopped they'd taken the drivers out of the driver's title running fully driverless vehicles

and they never claimed that we're doing that with all cars but my assumption had been they would kind of gradually shift to doing that with all cars. And then throughout 2018, they've been saying they're going to launch a commercial ride hailing service in Phoenix by the end of the year. And so they kind of sort of kept that promise in December when they launched a service called Waymo One. But it's kind of a weird commercial service because, number one, it was only open to people who are already in Waymo's closed beta testing program.

The only really changed for them was that now they're allowed to talk about it publicly, that it's the same program they were already in. And they also not only have one safety driver, but they've gone back to having two safety drivers, you know, a safety driver and also somebody in the passenger seat who's like taking notes and making sure the driver's paying attention. And this is not, it's hard to really regard this as a commercial service because they're charging like Waymo and Lyft prices,

but they're having their costs have to be at least twice as much because they're employing two drivers rather than one. And there's been, the anecdotal evidence that come out is that they really have not, even before this, you know, official launch, they really were not testing that many cars without drivers. That the driverless thing seems to have been kind of a more of like a kind of temporary experiment rather than something they were kind of ready to roll across their entire Phoenix fleet. It's almost like this official launch was just to meet some arbitrary deadline that they had put in front of themselves years ago. Yeah, I have not gotten them to admit that. But yes, they've been saying all year we're going to have an official launch. They technically had an official launch.

But it was like the smallest, least ambitious commercial launch that you can imagine. So you also tweeted that, like, if Waymo were confident in its technology as almost ready to go, you'd expect them to be doing way more test rides with real passengers to test out all the intricacies of the system, like down to, like, you know, getting in and out of the car and all sorts of stuff like that. But the fact that they're still in this really limited sort of scale of service right now, does that mean that it's seeming like we're a far way away still from maybe me being able to hail one of these on the street corner? Yeah, it's hard to tell. They are doing a lot of that kind of testing.

So they had this program called the Early Rider Program since early 2017, where they have a few hundred people who have been getting taxi rides initially. So they were free more recently have been charging for them. And so they're doing some of that testing, but I think it's just fundamentally, like, once you can actually take the drivers out of the car, then you can make a profit or at least break even on these things, which means it's economical to scale them up to much larger scales and then get the economy to scale that can kind of get the cost down. And, you know, if that were the case, you would expect them to open it up to the public. I mean, obviously, you couldn't, you know, there's millions of people in the Phoenix area, so you couldn't literally let everybody do it right away. but you could start accepting, you know, just kind of anybody, at least with the waiting list. And so the fact that they haven't done that, basically not, despite having a commercial launch,

have basically not increased the number of passengers at all, as far as I can tell, suggests to me that it really is still a, you know, it is a research prototype. You know, they're doing, you know, testing, but they're not really ready to run a commercial service. I read a quote on the show last month from somebody involved in autonomous vehicles, where he said that if everybody's at the point now where they're going into the real world, like this is really where the, I don't know, the pedal meets the metal,

I don't know what the analogy is, but like this is where it's getting real because when you're in the real world, if you're at 95% autonomy and you go to like, say, 96%, like that's an order of magnitude, different complexity. And then going up to 99% and 99.1 and 99.2.

So is it maybe that we're now, the stage where okay we're out of running the tests in the simulations and things like that but now that we are in the real world like there's so many new different variables that are popping up that like maybe this is an inevitable sort of slowdown that all of these players should have expected as they have to work out all of these kinks now in the actual real world yeah I think there's some of that so one of the fundamental tradeoffs you face when you run a service like this is you can you have a trade off between safety and convenience or you know, kind of the pleasantness of the experience. So it's pretty, really easy to build a

self-driving car that never hits anything if you don't mind, you know, going 10 miles an hour and slamming on the brakes anytime anything gets close to you, right? And on the other hand, it's really easy to create a smooth ride if you don't mind, like, occasionally running somebody over. And so every company has to kind of tweak that knob and say, you know, how conservative do we want it to be? And the evidence suggests that Waymo is super-servous conservative. And so there'll be some cases where, you know, there'll be a difficult merge or a difficult left-hand turn. And the car will just take a really, really long time to go because they're being super safe, not, you know, because especially if you think about it like when you're merging, sometimes in a busy

traffic, you just have to kind of nudge your car into the adjacent lane and expect the guy next to you to, you know, give you some room. And I think software, you know, programmers are just really reluctant to do that because if they screw it up, the consequences are really catastrophic. And so I think for the most part, most vehicles, as far as you know, they seem perfectly safe. But that last one for that sentence, I think, is largely dealing with congested or complicated situations and dealing with in some other way than just slowing down and hesitating, like, actually knowing when is it safe and how do you kind of navigate the, like, complexities of interacting with other vehicles.

The other thing I'll say, though, here is I think what we're seeing is you're starting to see different companies have different business models. And I've really started to think that Waymo might. be making a mistake by going straight for the kind of full-scale taxi service. There's been a long-running argument about taxi services versus manufacturing car to sell the customers. Where with that, you have to have something that can work absolutely everywhere. And sometimes as doing a taxi service is an easier problem because you can just do Phoenix,

which doesn't have snow and doesn't have really high-down city cities. But you see some other companies, other startups, they're doing even simpler problems. So you have, for example, a startup called Nero in the Phoenix area who recently started doing grocery deliveries with this little vehicle that it's on public streets, but it only goes 25 miles an hour, and it's small enough that it probably wouldn't kill you, even if it did hit you, at that speed. And I think you're starting to see startups that are going slower and going in even more confined areas where they might be able to get two driverless quickly, just because they don't have to solve some of these difficult problems, like how do you merge on the freeway?

Yeah, this is another segment I read recently that, you know, before I can actually commute in a self-driving car, I'm probably far more like. likely to get my groceries delivered by a self-driving vehicle. So you said just real quick that you think that possibly Waymo is making a mistake by going the taxi route. Is that because you're saying that that's like a more ambitious target right now? And so maybe they should like, you know, ramp up to things like that? Yeah, exactly.

So I mean, I think history suggests that often what you want to do when, certainly when you're a startup, but when you're any kind of technology companies, you want to get a product to market is like a real product. So like Waymo is not a real product in the sense that they're not making money or even breaking even, you know, per ride. And so by tackling the simplest possible problem, you can have like an actually useful service, which means you can expand more quickly and then you can learn from real customers and figure out how to make the product more affordable and more effective.

And then I think once you have a product, you know, that works, you can then tackle higher speeds and different geographic areas and so forth. Another company I'm really interested in in this respect is there's a company called Voyage that is focusing on retirement communities. They are offering a taxi service. But one of their initial locations is going to be a retirement community called the Voyage Villages in Florida. And it's really interesting because retirement communities tend to have just naturally much lower speed limits than full-sized cities. And they're also privately owned. and so you can make a deal with the retirement community as a whole.

And some of them are quite big. I think the village just has more than 100,000 people in it. And so their plan is just to build a taxi service for that enclosed community. And again, because they're only going 25 miles an hour, some of the trickier situations they don't have to deal with, like merging on a freeway and also just the safety. You know, it's just much, much easier because one of the big challenges for soft-dive cars is stopping distance.

If you can stop in two seconds, then you only have to see a few seconds out in the world. errors, you know, it doesn't really matter what's in front of you. You can just hit the brakes, whereas on the freeway, if you have, you know, five, six, seven seconds to stop, which you really have to know is that thing in front of me, you know, an obstacle I'm going to hit, or is that something on the side of the road or something that's going to get out of my way, et cetera. And so that's another example where I think in the long run, Voyage definitely wants to be running, you know, full-scale taxi services, but rather than immediately try to solve that

difficult problem, they're going to the easier problem of can we build a taxi service that never goes to 25 miles an hour. I mean, I think what they may be able to do is even though right now I think their technology is not as sophisticated, they might be able to get to the point where they can take the safety driver out and start to really scale that up to lots and lots of different retirement communities and end up with a lot more experience than Google, even though they don't have the kind of money that Google has. You mentioned again the merging onto highways thing, and that was the other tweet that, uh, tweet Storm that I wanted to talk to you about. So the analogy that you made is that if you think about like elevator doors, like they're programmed not to her. you. So like you think nothing of shoving your arm or leg into a closing door to get them open again. And so like you're saying, the existing self-driving cars are sort of programmed to be super,

super safe, right? And so we know that. And so as you were describing in your Twitter thread, other drivers will know that an autonomous vehicle isn't going to hit them. So if you want to change lanes in front of an AV, you can just do it. You know it'll stop. Conversely, if you don't want to let an AV in, you can just not make room. So it's almost like, I've thought a lot about this. Like the problem also in terms of entering into the real world is that there's going to be this long interim period where the AVs have to deal with us, have to deal with stupid people. You know, like in a perfect world, all the vehicles will overnight be AVs. But it's this long interim period where it's going to be, like AVs won't even be a large percentage of the cars on

the road for a long, long time. So is the problem going to be? dealing with us for the immediate future where like people are going to learn that you can be really aggressive around AVs because they can't be aggressive to you right no I think that's absolutely going to be a challenge and I think it's it's hard to predict what how that will work until it's kind of happened um so there I think there's a benign scenario I mean you think about elevator door it's like you know the elevator door is that it's not a big deal because the elevator door doesn't have anywhere anywhere to be right and um it's just that's just kind of how it works and so it might be that in the future, you know, a trip in a driverless taxi will just be a little bit slower

because you'll get cut off a lot, but you will eventually get to your destination. Or conversely, it might be that you end up with like complete gridlock where, you know, you have a driverless vehicle that can't merge into the next lane and that backs up an entire lane and then the whole kind of, you know, traffic jams become much worse. And yeah, I think in that scenario, which I think is pretty likely, it's going to be a big problem. And I can think of a few different ways to deal with it. One would be, you know, these cars do have cameras and LIDARs and so forth. So you could imagine kind of beefed up enforcement where you could, you know, if you cut off

a self-driving car, it will send, you know, the kind of full video details to the police. And then you'll get a ticket in the mail. And obviously people are going to hate that the same way they hate red light cameras, but maybe for extreme situations, at least that would help with that. I mean, I think you'll see the same thing with pedestrians. I think especially once there's a lot of self-driving cars on the road, I think pedestrians will become even more aggressive about, okay, if I see it a little lighter on the top of the if I can just like step in front of the car whenever I want.

So I think it's just really hard, you know, you can kind of try to imagine what it's going to look like, but these kind of complex systems can take on a life for their own. And so we'll just have to kind of wait and see what it's like when there's a lot of people. What do you think of those recent stories that have just started popping up about actual acts of Ludditeism popping up of people like, you know, taking their anger out on the AVs when they see them out in the wild and vandalizing them and attacking them? Like, do you, I suppose, is that a trend maybe we can look forward to over the next couple years? You know, I don't know. I mean, my guess is I don't think that that's going to be a large scale problem in the sense that like they're so new now that I think people with a lot of time to see it in Phoenix really feel, you know, like they have an opportunity to maybe strike a blow against this kind of thing.

When it's like national and there's like lots of them, I think it'll kind of become normalized. I do think that one of the weird things about the, you know, the kind of introduction of this technical. technology is that there really has not been much debate about kind of the level oversight that these things should have. And I do think that these companies might be being a little bit foolish. And I think they really haven't tried that hard to either be transparent about the safety efforts they've done or about the problems they've had. And I think that probably does fuel a certain amount of public concern.

So definitely the violence we've seen is not acceptable. I wouldn't ever condone it. But I think it would maybe make sense for some of these companies to try a little harder to just, you know, explain what they're doing, explain how the technology works, you know, show the data that shows the technology safe, et cetera. All right. Final two questions, a bit of prognostication. Do you expect that Waymo will expand beyond this Arizona test this year, 2019, maybe to California or other places? I would not be surprised if they start an early rider program similar to what they did in Phoenix last two years ago. In 2019, I will be very surprised if they launch any kind of commercial service. I mean, because as we were discussing before, like, it's not clear that this really is a commercial service. It seems like they have a lot more work to do, and it would not be surprised to me if they, you know, have made very little progress.

And we're still roughly where they are now a year from now. But it's hard to say. All right, so this one's a little more unfair, I think, possibly. But there's a long-running sort of bit that I do on the show about like my bet of, you know, I was promised that I'd be able to commute to work by 2020 and an AV. If you had to guess, my mythical AV commute to work, is it likely going to happen closer to 2020 or 2030? Okay, so your commute to work is in the New York area?

Right, which you're right is a factor. But let's say I'm a generic commute to work in Peoria or something. Well, no, I really don't think you can talk about it that way because it's going to be city by city. And it's going to matter, for example, I mean, right now, like Waymo is pretty good in Phoenix, but they really don't know how to do snow yet. They're testing. And so if you're in New York, I would expect you to be several years behind Phoenix, even assuming everything else goes perfectly.

And then I think probably larger metro areas will get service before a smaller ones. and who knows, there might be other factors. So I would say probably closer to 2030. I actually have a bet that I made eight years ago now with the economist Ryan Event. He had a newborn daughter back then, and we bet about whether she would need a driver's license when she turned 16 and 2026.

And I was on the pessimistic side saying that she probably would need a driver's license. And I've gotten a little bit, you know, I think there's a good chance I'll lose, but the last year, so I've started to think I might win again. So I still stick with about 2026. If I had to pick kind of the, you know, when like the average, you know, American commuter is going to be able to hail self-driving taxi, I would say 2026, but I think you'll see. I mean, I think Phoenix will probably have it before that

and there'll certainly be rural areas that might not be, you're not able to get one until after the end of the 2030s.