This transcript was created using speech recognition software. While it has been reviewed by human transcribers, it may contain errors. Please review the episode audio before quoting from this transcript and email [email protected] with any questions.

I’m Kevin Roose. I’m a tech columnist at The New York Times.

I’m Casey Newton from Platformer.

And you’re listening to Hard Ford. This week, Casey reports from Google’s AI bonanza. Then we hit the streets of San Francisco in a fully self-driving car. And finally, Cruise CEO Kyle Vogt stops by to tell us why the autonomous vehicle revolution is taking so damn long.

Casey, where are you?

I am in a booth in the press area at Google I/O at the Shoreline Amphitheater in Mountain View.

And what’s going on there?

I mean, what isn’t going on, Kevin? Did you even watch the keynote?

[LAUGHING]

I didn’t. I had a conflict. How was it?

Well, people are milling around. It’s a few hours after Sundar Pichai, the CEO of Google, and recent Hard Fork guest —

Friend of the pod.

— spoke, along with his colleagues, for about two hours and laid out everything that the company has been working on over the past year or so. Google I/O is their annual developer conference where they show off what they’ve been cooking.

And as you might expect, most of what they showed off today had a lot to do with artificial intelligence and how they’re going to try to use that in their products going forward. So compared to other years, this felt more eventful.

Yeah. I mean, I’ve missed this years. But I’ve been to previous Google I/Os. And it’s sort of like a combination product reveal and, like, party. Can you just describe the pageantry of Google I/O? Because it is a pretty big budget involved event.

Oh yes. It’s quite a spectacle. They’ve taken over the Shoreline Amphitheater, which is an outdoor concert venue here in Mountain View. They’ve filled it up with colorful tents and drawings. I wish I could tell you that there were, like, fire dancers or sword swallowers. But it’s —

[LAUGHING] It is a tech conference at the end of the day.

This is still a corporate conference.

Yeah.

So then I imagine, as happens most at these things, a series of executives come onto stage and make their presentations about the thing that they are announcing that day. So what were the big announcements from Google I/O?

Well, I think the biggest one is just that generative AI results are coming to Google. So things that you might have been using ChatGPT for up until now, you’re going to be able to start using in Google itself.

So one of the examples they showed off was asking, “Of two national parks, which one would be better for a family with children and a dog?” And that’s not something that you would normally ask Google, because the core Google search engine isn’t good at parsing that many different variables. But this generative AI model that they have is able to say, well, of these two parks, you might want to go with this one because at that one dogs actually aren’t allowed.

Got it. So this is coming to Google search when and for how many people? And how drastically do you think it will change what people see when they go to google.com?

So they’re starting small. It is going to be in the United States first. And it is going to be rolling out in the coming weeks. They’re keeping some of the details vague, I think because they want to give themselves the flexibility to slow it way down if something goes disastrously wrong. But my expectation is that I will probably be able to use it within the next few weeks, and I bet you will too.

Wow. Yeah, what could go wrong? OK. We’ve got this Google change to update its search results to have more generative AI answers. What else did they announce today?

Well, I have to tell you about the thing that I thought was the coolest. Because I did have a moment where I basically wrote down curse words in my journal because I was so taken aback. Have you ever heard of Project Starline?

No.

So this is essentially the fanciest Zoom you have ever seen in your entire life.

I’m intrigued. Go on.

OK. Do you remember the metaverse?

I have a vague recollection that in the year 2022 people were talking about something called the metaverse. Yes.

Right. And one of the big ideas was what they called telepresence, the idea that people in your life, you would be able to interact with them in a way that felt very realistic. And the main thing about this vision, which was mostly being advanced by Meta, was that you had to wear a giant helmet on your head. You remember this?

Yes. We in fact did an episode where we actually wore these clunky VR headsets on our heads and did a podcast in the metaverse.

You can actually go back in the archives and hear the episode where we have headsets on. OK. Here’s why I’m telling you about Project Starline. Imagine if you could have an experience that was 100 times better and you didn’t have to wear a headset.

I’m sold. Tell me more.

OK. So I go into this demo. And you sit down in front of what looks like a TV, like, a big screen TV. OK. And it has little speakers on the sides and something on the top that’s clearly some cameras.

And right before I had gone into this booth I had met this man who works on Project Starline for Google and saw him in the flesh. And he’s like, “I’m just going to go into my booth and you’re going to go into your booth and we’ll start talking.” I was like, “Great.”

I go and I sit down. And then this man shows up on the screen. And I am not kidding when I say the resolution of his face is as high as if he were standing in front of me. OK? So that was the first thing that was cool.

The second thing he does, Kevin, he reaches through the screen and gives me a fist bump.

No. Come on.

Yes. There is something 3D about this technology. And so you can — it looks like you’re reaching through the screen.

No, no, no. I’ve seen “The Ring.” I don’t want anything coming through my screen. [LAUGHING]

The horror movie potential is high. After he did that, he had an apple. He reaches through the screen and he shows me an apple. He starts turning the apple around. And I’m moving my head around. And I can, like, see different sides of the apple. I am not kidding you. I have not seen a tech demo that I think was this cool in a super long time. I’m all in.

That’s amazing. And it also reminds me of, like, do you remember the stories from when, like, motion pictures first came out? And people would run screaming from the theater because they thought the train was going to jump out of the screen and hit them?

Yes!

You were the person —

Yes!

— in the theater, running from the train.

Completely. Completely. If you put a horror movie on this thing, you’re going to traumatize people. It was so amazing. So, look, I mean, the technology — of course I’m asking. I’m like, how expensive is one of these rigs? Well, you’ll never believe it but they don’t want to talk about it. But, you know, I was asking around afterwards. And they told me that the number one issue with getting this thing into more people’s hands is the cost. So let’s just assume it’s like some astronomical amount.

Yeah. Your $5 million dollar fancy Zoom call is maybe not coming to an office near you.

But they told me that they have given some of these to Salesforce, T-Mobile and WeWork. And they are testing them at Google. And they told me that right now they’re just trying to figure out, basically what they’re good for.

And I’m like, bro, they’re good for everything. Like, put it in my house now. This is how I want to talk to my family now and my friends. I’m not even kidding when I say, like, you could have a beer with a friend on a Friday evening and you would truly feel like you hung out with them instead of having a Zoom call. It was —

Wow.

Anyway. This thing was amazing.

I got to try this. I mean, between the fully self-driving car and the $5 million fancy Zoom, I feel like we are really living in the future this week.

Truly, we’ve gotten a glimpse at some of the happier parts of the future, which is nice to do every once in a while.

But as fun as your fancy Zoom was, it is not, strictly speaking, an example of generative AI. So —

No.

— bring us back to generative AI, which was the sort of focal point of this. What else did Google show off that it has been working on in some of those areas?

Sure. So they have something called Help Me Write in Gmail. It’s exactly what you think it is. You give it a few keywords of what you want to do. It’s going to write your email for you. I think within a year lots of people are going to be using this to draft lots of emails. So.

And it happens right in the Gmail window itself? Or do you have to go to an app and write and then copy and paste it into Gmail?

It happens right in Gmail. And then on the dorky side, they showed off this presentation that consisted only of slides. And they took the slides and they said, “Write speaker notes for the slide.” So essentially write the presentation based on the slides. Snap your fingers and it showed off a bunch of things.

So think of how lazy you can be as an office worker now who has to deliver — you don’t even have to write your own speaker notes anymore.

Right. I mean, I can see how this is all potentially very exciting and lucrative for Google. At the same time, some of these features have already existed in Microsoft products now. Is this Google playing catch up with what Microsoft has done with its products and generative AI? Or were there things that happen today that struck you as, like, oh, man, they actually are breaking new ground here?

You’re absolutely right. Microsoft has been rolling out similar features in their office products. They call their set of features Copilot. Google is calling its features Duet. And there is a lot of tit for tat that’s happening here.

But the honest answer to your question is that my life is in Google products, not Microsoft products. And so this stuff is just hitting me harder. Because I look at this stuff and I already know how I’m going to use it.

Another issue that we talked about, actually with Sundar Pichai is, like, Bard just didn’t feel very powerful when it was released. By comparison to ChatGPT it wasn’t giving people the kinds of answers and problem-solving abilities and coding abilities that ChatGPT and other generative AI products were.

So did they talk about Bard at all at this event today? Are they making it smarter? Is it changing in any way? And do you think it’s actually going to be competitive with ChatGPT?

Yeah. So a couple of things. Number one, they have added buttons, or I should say, they are adding buttons to this that lets you export the output to Sheets and Docs. And I know what you’re thinking. You’re, like, Casey, that is the most boring thing you’ve ever said on this podcast. But I’m telling you, like, this is the actual secret sauce of what is going to make this successful.

The past six months have just been a Lollapalooza of copying and pasting. Everybody’s going to ChatGPT and they’re generating text and they’re bringing it over to some other app. And the smart and obvious thing that Google did was said, hey, come use our thing and click one button and we will pre-populate a Google Sheet for you, a Google Doc for you.

The second thing that they’re doing is they’re upgrading the language model that powers it called PaLM. And so PaLM has now gone from PaLM to PaLM 2. Now can I tell you on a query by query basis what that’s going to mean? No. But in recent weeks Google has said a lot about how much better they think Bard is at coding challenges, at mathematics, which we’ve talked about in the past, as these models being really bad at doing. So it is improving over time. Their positioning that as a major upgrade.

OK.

Wait. I want to make a point.

Yeah.

Kevin, I read a super interesting story today in the information about where Microsoft is at with Bing. And there’s a research firm called YipitData. And they told the information that Bing’s share of global search has grown 0.25 percent since it launched the new Bing.

That’s it?

That’s it. Meanwhile, the OpenAI chat bot, which launched six months ago, already draws more visitors on desktop devices than Bing does. And Bing is 14 years old.

Wow.

Now why am I bringing this up? This is Google’s advantage, right, is you’re already using Google’s products. And so as the AI starts to creep into m I think they’re just going to have this tailwind that Microsoft doesn’t, because people aren’t used to using Bing. Right. And so to the extent you want to think about this as a horse race, this is where Google really could not just catch up, I think, but pull ahead is that they just have the stuff that everyone’s already using.

Huh. It’s basically, like, this idea that where AI is, is just as important as how good is is. Right?

Absolutely.

Maybe more important. You know, I was of the mind, when these chat bots first came out, that all that mattered was the quality of the AI model. That people were really going to only use the state of the art thing that they had access to, and that you would want the most powerful AI.

But I think there’s a convenience trade-off to. Where if I’m asking about something that’s sort of the kind of question that you don’t actually need the state of the art for, you’re probably just going to use the AI that is closest to you, right. The one that is in your Gmail already or your Docs already or your Microsoft Word already. And you’re not necessarily going to go shopping for the most powerful model that can maybe refactor a code base, but maybe that’s not what you’re doing anyway.

Absolutely. I think the most natural home for all of this stuff is inside the tools, the operating system that you’re already using. And I think Google realizes that too and they’re moving full speed ahead.

So one recurring theme that we’ve talked about on this show now for several months is this sense of panic inside Google about the generative AI arms race, and the fact that they are, by many accounts, behind OpenAI and Microsoft when it comes to getting these tools into people’s hands. Obviously they had their code red last year.

But I also read a story this week based on a leaked internal Google document that was written by a Google employee. And it was called, “We Have No Moat.” And it was basically an argument that, despite all of the progress that Google is making, despite the fact that they’ve invented a lot of this fundamental AI technology that’s going into these products, this person, this Google employee in this leaked document was basically sounding the alarm and saying, we don’t actually have an advantage here and neither does OpenAI.

Because there’s this third threat that we don’t have a strategy for. And that is the open source AI movement. That basically there are these open source models like the ones that have been coming out from Stable Diffusion for images, more recently like the ones that have come out of Meta, the llama models and the spinoffs of that.

And that basically these tools that people are downloading and running on their own computers, their own hardware, are rapidly catching up with the state of the art AI models from Google and OpenAI and other labs. And ultimately that Google is not positioned to stay competitive in a world where everyone can just download AI models and run them on their laptop.

So how are you thinking about that threat, the open source threat after listening to this Google event today?

Well, so that memo is super interesting and people should check it out. But I just fundamentally disagree with the person who wrote it. Kevin, are you familiar with Linux?

Yes.

Linux is an open source operating system. You could probably install it on your computer right now. Except you wouldn’t. And you want to know why? Because it’s a huge pain in the ass. And then after you install it, you spend the rest of your life wondering, can I install that on Linux? Is there a Linux version of that? That is the open source world. OK.

So for the same reasons that I don’t think Linux poses a threat to Microsoft Windows or iOS, I don’t think that these open source models pose a threat to Google, at least not in the way that we’re talking about, right, for all the reasons that I just said.

The underlying technology piece is a huge part of the problem that needs to be solved. We’re seeing a lot of progress there. But there’s also a product problem. You have to make something that is easy to use, that is pleasant to use. Right. And Google, by the nature of the fact that it makes these tools used by billions of people, just has a huge advantage there.

Now, look, could someone come along and take a bunch of open source components and make an incredible new company that winds up challenging an incumbent? Yes, absolutely.

And Meta, in particular, has released a bunch of open source stuff, which I think it thinks is going to benefit them hugely by just essentially helping people create a bunch of content that is going to populate Facebook apps. But does the existence of a bunch of open source models running around on phones and laptops mean that people are less likely to use Google search with generative AI? I don’t think so.

Right. And what about this other threat that people inside Google have been sounding the alarm on for months now, which is the threat of releasing generative AI to the world before it’s safe and harmless? What did Google say today about the safety of its AI tools? And were you reassured that they’ve actually done what they need to do to make sure these things are safe?

Yes and no. Yes, they did talk about it. They talked about their responsible AI principles. It came up several times during the keynote. I was briefed on a few of these things before the event. And everyone I talked to made sure to talk to me about how responsibly they’ve developed everything.

But I do feel like, on some fundamental level, nobody really knows if it’s safe or not. You know. They’ve done their tests. They’ve done their red team exercises. They’ve tried to identify obvious misuses and to mitigate those. But at the same time I think that they know that once you give these tools to a billion people, you’re just not going to be able to predict user’s behavior.

So the approach that they’re taking is to roll it out slowly. And I do think that if there is some immediate, massive, obvious harm, they will stop the thing in its tracks and they will pull it off the market.

But at this point, basically everyone I know — including I think a lot of high ranking executives at Google do wish that things would slow down a little, both to give society a chance to adapt to what’s happened so far, and to give them a chance to just kind of see, OK, like, you can do this with this now. Is there some obvious harm that we missed?

Well, but they’re clearly not that interested in slowing down or they wouldn’t be holding events and showing off all the generative AI they’re shoving into their products.

That’s exactly right. It’s a fair criticism. But at the same time, even us — both of us believe that very deeply. And both of us were also like, why is Google moving so slowly on this stuff? They’re losing.

[LAUGHING]

Right? So everyone talks out of both sides of their mouth on this, I think, except for maybe some of the most diehard AI safety researchers.

Yeah. So I’m curious, just taking sort of a step back at this whole event that you’ve seen today, what is the vibe there? Is it a company that was caught flat footed and is now mobilizing everything in its power to catch up with OpenAI and Microsoft? Like, what is the vibe of Google I/O this year? And how are you feeling? Have you changed your mind at all about where the company is headed?

So I would say two things about the vibe. One is constant reminders that they’ve been doing this for a long time. So basically everything that Google announced was prefaced by a statement that was something like, in 2017 we did this. In 2019 we released this.

And the point that they’ve tried to make is, we truly have been laying the groundwork. That, you know, you jackals and hyenas in the press who accuse us of being asleep at the wheel are ignoring the fact that we’ve been laying the foundation for years. So that’s kind of the first thing.

When it comes to the products themselves though, Google’s vibe is a little bit more understated. Right. They’re less likely to come out and say, we’re the most path-breaking, innovative company in history. Instead they’re going to tell you, we have made a thing that we think is really helpful and we can’t wait to show it to you. So that’s kind of the vibe.

There’s a third thing about the vibe that — maybe this is more of a tech event thing than a Google thing, but it just really struck me, which is that every time something would be announced during this keynote there would be — somebody would, like, scream. Like “Woo!” Like at the top of their — just like outrageously excited about a development in a language model followed by laughter from everyone else in the audience. Like, look at this nerd.

And it’s such a weird, kind of whipsawing between joy and shame. And you can really only find that at a tech event. I’m just so glad they’re back for that reason.

Wow. This is like the tech reporter version of Coachella. This is good stuff.

[LAUGHING]

Casey, thank you for attending Google I/O on behalf of the Hard Fork podcast. Did they announce that they are renaming their chat bot Bard Fork?

They did not. But I’m told that Hard Fork is now available on Google Podcasts. So that’s an exciting development.

Your $5 million Zoom call from heaven could never approach that level of excitement.

Truly.

[ELECTRONIC MUSIC PLAYING]

When we come back, self-driving cars made by Cruise are now running around the clock in San Francisco. So Casey and I hailed a ride in one to find out if the driverless car revolution we’ve all been talking about and waiting for for years has finally arrived.

OK. And now we’re just going. Holy cow.

Oh. Oh. OK.

We’ll be right back.

[ELECTRONIC MUSIC PLAYING]

This is our podcast studio. And this is Casey.

Hi.

Hey, good morning.

How’s it going? Nice to see ya.

Kyle Vogt, welcome to Hard Fork.

Hey. Thanks for having me.

So we wanted to talk to you because Cruise has just recently announced that you’re are operating 24/7 in San Francisco. And we actually just this morning took a ride in one of your cars named Banana Slug.

Yes. I don’t know if that one. But —

I’ve seen Banana Slug out there on the roads. Yeah.

Yeah, yeah.

What was the naming convention? Like, how did you decide on the names for the cars?

I mean, one of our interns named one of the early AVs, like, for fun. I think the theme then was bird names. And that stuck. And so we really leaned into it and let people suggest names. And I have no idea where Banana Slug came from. But it’s not the weirdest one that’s out there.

What’s the weirdest one that’s out there?

I’d have to come up with a list. But people have a very emotional attachment to the first AV that they rode in. Like, Mary Barra, the CEO of GM, her first car was Tostada. And every time she comes to San Francisco she’s like, how’s Tostada doing?

[LAUGHING]

Actually it’s been a rough few months. She’s out for repairs. We don’t like to talk about it.

[LAUGHING]

It’s a sensitive subject. So, Casey, what were your impressions of the ride?

Well, I mean, overall it’s super fun. This was the first ride that we had taken in self-driving cars where there was not a safety driver. We’ve each probably taken half a dozen or so rides in self-driving cars in the past but there’s always been a human. And there is something amazing about just seeing the steering wheel turn itself. Kevin was sort of likening it to a player piano, sort of watching that for the first time.

You know, at the time it made some herky jerky movements and it drove, like, six miles per hour slower than the speed limit. So we do have some questions.

Yeah. And it did almost swerve us into a median at one point, which led the human driver on the other side of us to roll down his window and say, you guys should report that thing. So here we are reporting it.

Yeah.

You guys are hilarious. You are moving through San Francisco in a technical marvel that took you across the city with no one behind the seat, in probably half an hour or more, and you found three relatively small things to pinpoint. [LAUGHING]

Meanwhile —

This is what I told Casey. Because he wanted to give it three stars at the end of the ride. And I was like, we got safely to our destination during rush hour in a car with no driver. That’s a five star.

I’m just trying to help them improve. But you’re right. It is a technical marvel. Yeah. Yeah, we want to talk about that.

So first of all, I want to ask, like, what does it mean that you’re operating 24/7 in San Francisco now? Because as we discovered when we were trying to set up this ride, it’s actually like, not everyone can just whip out their phones and start hailing autonomous drives.

Yeah. So we’ve been rolling things out in phases. Our approach is — we call it envelope expansion where we operate in a small area. We basically have done the analysis. We believe it’s ready. And then we operate there. We check all the indicators. When everything is green, we expand it.

And so this can refer to the area, whether it’s the full city or a part of the city. It could refer to the hours of the day or the number of people that we allow on it. So right now we’re operating 24/7. There are a small number of cars that are out there all the time available to employees. And then over the next few weeks or months we’ll open that up to everyone else.

And how do prices compare right now to what people might be used to paying for an Uber?

Pretty similar. And, in fact, if for whatever reason the AV takes a longer route, we charge you for what the route should have been.

Got it.

Today though we’re pricing it at parity with rideshare as it exists. And we expect that — well, rideshare right now, it’s already about as cheap as it’s going to get. These companies have optimized it, and they’ve found that basically the minimum amount of money they can pay the drivers to keep drivers on the road.

But that’s limited by the cost of a human wage. Self-driving cars don’t have the human. So as we drop the cost of the technology and the cost of operating these cars and increase the size of the fleet so it gets more efficient, we can get way, way, way below the cost of what exists today.

And we also know from the data that people who use rideshare are very price sensitive. They want to save a buck on that ride.

Yeah.

What that means is when you drop the cost a little bit, the number of people using rideshare as their primary mode of transportation is going to go up, which we’re excited about, because I think it helps with congestion and traffic in cities.

So the last time I really paid attention, close attention to the self-driving car market was in 2017, 2018. At the time there was a huge wave of sort of hype and excitement about driverless cars. Uber was investing tons of money into it. Lyft was telling people that rolling restaurants, you know, self-driving hospitality vehicles were just a couple of years away. You know.

And Casey and I both went on a number of rides at that time. And the line that we got then was like, yes, they only work on a predetermined route during certain hours now. But just wait a couple of years. They’ll be everywhere.

So we waited a couple of years. And with the exception of a couple cities, they’re not everywhere. So what happened between, let’s say, five years ago and now to make you convinced that these vehicles are ready for prime time? Is it some technical breakthrough that made them safer and more reliable? Or is it just that you’ve had more time to kind of work out some of the kinks?

I want to answer that in two ways. First of all, they are scaling up rapidly and appearing in more cities now. Like, that’s starting to happen. And so we’re on the verge of that rapid scale. We launched two new cities last year. There’s going to be a lot more big announcements in news on expansion this year.

So why did that just happen now instead of five years ago? Two things. First of all, when you have a person behind the wheel and they’re in charge of taking over in all those tricky edge cases, the last one percent of the problem where the system isn’t ready, 99 percent of the time it feels like it’s done.

And so you can take an investor or a potential partner or someone on those rides and they’re like, oh, this is here. We need to invest now. Like, throw all the money at it. And that went crazy. It was like a gold rush, like maybe like crypto a couple of years ago or probably like AI and large language models right now.

Yeah.

Maybe people are a little more excited about that than they should be. And then the rest of the time between then and now was the hard work to continuously grind away at those edge cases, that last one percent of the problem until you can safely remove that human out of the car. And that’s just what it took.

So it really was mostly a data collection issue then.

Data collection was a piece of it. But there’s just a lot of engineering challenges that go into things like, what if someone leaves a car door open after they get dropped off? Or, like, what happens if the lane markers have been repainted and there’s a police officer waving around traffic? What’s the car supposed to do? Like, these things that don’t happen that often, but when they do, the car just can’t sit there. You actually have to find a way through it.

So wait, what does happen if somebody doesn’t close the door at the end of a ride?

We just launched this last week. We have auto-closing doors.

Nice.

So the doors will just like — someone logs in with a camera. When the vehicle says, hey, my door is open, they check that the coast is clear and then the door will slam shut.

What other things did you see in your testing? What other edge cases have you seen that maybe threw you for a loop or caught you by surprise?

Yeah. So, I mean, there’s a lot of things that are weird, like situations. We see sideshows in San Francisco when people will block down an intersection and just do donuts in the middle of an intersection. Didn’t have that in our test cases early on. Right. Had to think through that. A lot of stuff with swarms of people rollerblading in the street or segues or bikes.

But that stuff, in a way, is kind of expected. Like you’d expect a large city to have a diversity of stuff, including some weird things. What I didn’t see coming is that a lot of people would essentially risk their lives and try jumping in front of a moving car to see if it’s going to stop.

We literally talked to someone today who did this.

Yes. We were at a coffee shop and there was this guy there who heard us talking about self-driving cars. And he says, oh, I’m a retired cab driver in San Francisco, and starts telling us about how he once jumped in front of a self-driving car just to see if it would stop.

And great news, it stopped.

Yeah. I mean, it’s terrifying to me. Because we want to improve road safety, not create situations where people are creating new opportunities for themselves to get hurt. So I hope that this is a short lived thing and that we all understand that these cars are going to stop and don’t need to be being reckless like that.

Have you ever thought about just, like, putting like some sort of device that shoots a beanbag at people if they’re —

We thought about water guns.

Yeah.

[LAUGHING]

Maybe those, like, exploding ink things.

Yes! Like if you’re trying to rob a bank. I’ll tell you, if that happened to you, that’d be the last time you ever mess with a Cruise robotaxi. [LAUGHING]

I want to ask a series of questions about safety. Because I think this is first and foremost on people’s minds. And in San Francisco most of the rides, maybe the vast majority of rides go off without a hitch. But there have been some incidents.

Earlier this year officials in San Francisco sent letters to state regulators asking them to either halt or slow the expansion of Cruise and Waymo because some of the cars had blocked traffic or gotten in the way of emergency vehicles. How do these incidents affect public support for driverless cars?

You know, when there aren’t that many driverless cars, and the only thing you hear about are these relatively rare incidents. Remember, we’ve gone almost two million driverless miles at this point. On a typical night we’ll do 1,000 rides for people. There’s a lot of stuff happening without incident. But when we do have a stumble or a misstep we’ll apologize for that. Or if appropriate we’ll do a recall and inform members of the public about what we changed to improve these systems.

And so, if that’s the only thing you hear about, I think that’s going to slow things back. But on the other hand, at this point we have tens of thousands of people who have ridden in this and are starting to question, OK, is this what’s happening all the time? Or is this sort of the rare events that get sensationalized and picked up and get lots of coverage?

And our view is that over time I think people will start to see more that these things do happen, but they’re rare. And I think more importantly, we’re starting to see evidence that the presence of these vehicles on the road actually is having a positive impact on road safety. And that’s priority number one for us.

These cars — not on our ride thankfully, but in some cases they are making very high stakes decisions. Like, do you swerve out of the way of an oncoming car if it means swerving into another car? If it means choosing between hurting one passenger or maybe the driver of a different car?

So this is the classic trolley problem where you have to decide who to run over in a hypothetical trolley. But you’re actually dealing with a real version of this with your car. So is there any programming in your cars that has to make decisions about which humans to protect in the event of an accident and which not to? How are you approaching these more philosophical debates that you actually have to then build into real products?

So, a fair question. It’s not one we’ve seen yet in two million miles of driverless operation or in the several million that we did in testing before we went driverless. But I accept that it is a possibility.

Right now we program based on trying to minimize injury on the road. And so that generally means we want to give more space to pedestrians and vulnerable road users, even if that means getting closer to a vehicle. And that’s because occupants in a vehicle generally have better protection mechanisms, like airbags and seatbelts and other things, than people on the road. So that’s a good starting point. And that generally has led to good performance and good decision-making.

We have not — because there hasn’t been a need to engage in these ethical philosophical debates on if you are in this situation where it’s equally necessary to go one way or the other, which one are you going to do, which lever are you going to pull? Because the reality is, in almost all the situations we’ve come across so far, there’s a clear choice that you can make that minimizes risk of injury. And it hasn’t really been a choice that we need to make.

But what I think people should know is that we’re using all the best research that exists to try to make these cars as safe as they can be on the road, recognizing that there are situations where there will be collisions, and it’s unavoidable.

One of the things that I’ve been struggling with is figuring out what our societal fault tolerance is for self-driving cars. I think if you had asked me 10 years ago, you know, when will society accept self-driving cars, I think I would have said, they just need to be safer on average than human drivers. Right. If they cause a few fewer crashes or fatalities every year then society will say that’s a good thing.

It seems like, from the evidence we have so far, like, it’s actually got to be much better than a human for society to say, OK, I want to get into that thing. How do you think about society’s, sort of fault tolerance for driverless cars. And does a self-driving car need to be 100 times safer than a human driver before we will feel safe all riding in them every day?

Yeah, look. I think it’s a really interesting philosophical debate. And I’ll tell you, there’s a bunch of different lenses to try to look through to come up with an answer. One is, what do our federal safety regulators look at? And when it comes to any new technology, they’re looking for unreasonable risk and aggregate improvement on road safety.

Even our Secretary of Transportation, Pete Buttigieg, was out there saying exactly the problem you highlighted. Like, even if robots do result in far fewer people dying on our roadways, even a small number of them is going to have people go nuts.

And I guess the last way to look at it is, if you forget the fact that there is no person in the car, but think about it like segmenting different groups of the population, old drivers from young drivers or other things, like, if old drivers are 50 percent worse than new drivers, how do we feel about that? And I think the same could be said for just a different category or grouping of drivers.

So I think where we’re going to land on this is that we want safer roads. And it’s a really interesting thing to talk about now when humans and robots are maybe relatively close in performance. But over the next couple of years, given the rate of improvement that these systems have had — talking about five years ago they weren’t ready yet now they’re out there and scaling. I think this is going to be a conversation that’s in the rearview mirror. Because they’re going to be so much better than humans and we’re not even going to have this debate.

And when do you say rearview mirror, is that kind of a self-driving car joke, or?

That was a good automotive pun. I have too many of those programmed into my head right now. [LAUGHING]

Kyle, I have to say, like, I disagree with you. Because I think that the minute that Kyle can get me across town for $3, like, I’m not looking at the safety report for a Cruise. I’m saying, that is the cheapest way to get from point A to point B. And if the thing shows up with three wheels I’m still getting inside because it’ll probably be fine. Don’t you think most people are going to be that way?

I don’t know. If you got on a plane and there was no pivot in the cockpit, even if you knew intellectually that plane is just as safe as a human sitting in the cockpit —

Yeah.

I think you’re still thinking twice about getting on that plane.

Well, it is different for me because I have been driving in these things for admittedly only a handful of times. But because it’s gone back so far in my mind now that it weirdly has lost the novelty a bit. Which as you sort of alluded to earlier is an incredible moment, right, just in the history of technology that a thing that is driving itself feels a little bit boring.

But I truly think that after you take your first self-driving car taxi and it’s basically fine, you’re just going to take them a lot more. You don’t think that’s true?

I don’t think so. I think humans are not perfect risk assessors, right. Like that’s why we should all be a lot more scared to drive than fly, because flying is much safer statistically. But a lot more people have fears of being in an airplane.

So I just think you’re working on a problem that is as much about psychology and human, sort of comfort as it is about technology. Would you agree or disagree with that?

Yeah. I mean, there will be an interesting test for society to see what happens the first time that a driverless car gets into an incident and someone gets hurt, severely hurt. It’s not a question of if but when.

And even with all the data showing evidence of positive safety impact and other things, we as a society will have a choice at that moment to decide whether we ban something that could potentially save tens of thousands of lives per year over the next several years, or we decide that this is a horrible situation and something awful, but this is something that we need. And we’re going to learn from this and keep moving forward.

And I don’t think that’s really something that I can say as a CEO of a company in the space. It’s really going to be up to our society and our regulators and our government decide what we do in that moment.

When we were riding in Banana Slug today, I had the strong feeling that banana slug has been told, please be more cautious than the average car. Is my sense accurate?

Yeah. It’s easier to drive cautiously and slow than really fast while also maintaining safety. And so as the performance of these systems improve, you’ll see them start to drive in more human-like ways in terms of assertiveness.

There are moments when cruise AVs already are at that point. And there are moments when they are around a mid-block crosswalk or a narrow street with lots of parked cars where it’s hard to see around the corner, that they’ll go slower than a human. We’re working on that. But we’re not going to move faster than we’re comfortable.

You mentioned a second ago that you’re seeing evidence that your driverless cars are having a positive impact on road safety. Can you just flesh that out a little bit? Like, are your vehicles safer on average than a human driver? Or are they less safe? Or about the same?

Well, I’ll start by saying it’s a hard question to answer. Like, what exactly is human driving and under what circumstances? What we have done is partnered with VTTI and UMTRI, which are these transportation research institutes, and try to produce an estimate of how well the human drivers drive in San Francisco, specifically.

Horrible.

[LAUGHING]

I mean, actually yes. Because the commonly cited national statistic on collisions is off by a factor of 10 to 20 compared to what we see in San Francisco. And so what we have seen so far — and this is still early. It’s only across — we analyze our first million miles of driving. So that’s a lot, but also not a lot. Is that the AVs are getting involved in far fewer crashes, 50 percent as many crashes compared to what we observe humans doing in a similar driving environment.

What’s more exciting, though, is the more severe crashes that would be more likely to lead to injury. AVs are doing significantly better at avoiding those, I think something like 70 percent fewer.

I think you could probably drive way safer than a human just by never going 70 miles an hour.

Right. And never checking your phone while you’re driving.

Yeah.

I mean, that was my main thought about San Francisco, why it has worse drivers than most cities, because everyone’s constantly checking for new Bluesky invite codes or whatever.

That really should be your whole marketing campaign is, like, Cruise robotaxi, check your phone all you want.

Right.

[LAUGHING]

There it is. There it is.

I mean, is that — that’s sort of a joke, but that’s sort of not. Like, I feel like a killer moment for the AV industry is when you can actually just watch Netflix in the car —

Yep.

— like, while you drive to work. Is that something that you’re thinking about? Are you putting in — I mean, we saw the trivia system in Banana Slug today.

Which we crushed. I crushed it out, I have to say.

It’s pretty hard. If you did, nice.

So anyway. Are you doing more with entertainment inside the cars?

Yeah. I mean, so the new car we’re building, it’s a six-seater. The seats face each other. It’s more of this living room on wheels configuration. We’re testing that in Austin right now. And that really makes it clear that, OK, if you don’t need the front seat and the seat’s facing forward and the ghost steering wheel that’s moving, this becomes a really social space.

And I think that is time that we get to reclaim. And whether you want to watch Netflix or do work or talk to your friend or whatever it is, I think that’s pretty compelling. Because these days so many people in companies are competing for our attention and our time and sucking it away from us. And I think the notion of giving that back to people is really exciting.

Kyle, I want to ask you about the other big question, in addition to safety that I think is on people’s minds when it comes to autonomous vehicles, which is jobs and job displacement. Millions of people drive for a living, whether it’s Uber drivers or cab drivers or delivery drivers, truck drivers. How do you think about the effect that autonomous vehicles could have on those people?

I mean, I think with any new technology like this you’re going to see a shift in the distribution of jobs that people have. And I think actually one of the nice things about self-driving cars, as compared to maybe like, LLMs or ChapGPT — that happened very fast and very suddenly and could be very disruptive in a very short period of time. I think with self-driving cars, you know, almost a nice side effect of the fact that five years ago we thought they were right around the corner and it took longer than people hoped, there’s a lot of people know this is coming.

And so, anecdotally, when I talk to an Uber Lyft driver in San Francisco, no one is saying this is my long term career. This is the thing I’m doing now. When the self-driving cars are here, I’ll do something else I think for maybe truck drivers and other things, it’s maybe a bit more work to enter a different job or change the distribution of jobs.

But on the other hand, like, I think we do have to take a step back as a society and say, do we wish the Industrial Revolution never happened? Or do we wish that computers didn’t come out and displace a whole bunch of manual data entry jobs? I can’t say that we do. And so I think each one of these things looks scary at the beginning, but on the other side it makes a ton of sense.

Yeah. I remember talking with one of the co-founders of Lyft many years ago about this. And he was saying, well, you know, yeah, people won’t be driving the cars. But you’ll have rolling restaurants and rolling gyms and you’ll need people to staff all of that. And so he was sort of — he did seem to feel some responsibility for making sure that the people who are currently driving for a living and may get displaced by this technology actually found other work.

Do you feel a similar sense of responsibility for the people who are currently driving for a living to find them new work or to create new work for them?

I think we have to contribute to it, for sure. You know, I think — and that could take the form of to the extent we’re able providing training programs or alternate jobs for people. But more than that, I think it’s interacting with our government and our regulators and letting them know this is coming, when and how, and giving them some notice so we can plan ahead a little bit.

There are a lot of things that can be done if people or states or the federal government makes us a priority to ensure that this is a smooth transition. But some of that’s beyond the reach of any one individual company.

Can you say more about these rolling gyms?

[LAUGHING]

This is like a box truck with a Peloton in it and it’s driving itself?

Yeah, I guess. I don’t know. I didn’t really ask for more details.

Kyle, I would maybe move that one down lower on your product roadmap. I don’t know if that’s where the consumer demand is at the moment.

How do showers work for that?

Right. That’s what I want to know.

[LAUGHING]

Like, they have fresh towels?

You hit a bump, all of a sudden the weight plate is crushing you. I want to do some future-casting. But first I want to ask about the business model of self-driving cars. So Cruise, your company, is a subsidiary of General Motors. Last quarter Cruise reportedly lost $561 million, or about $6 million a day in the last quarter. It’s a lot of money. A lot of self-driving car companies have lost billions of dollars over the years.

So how long do you think you can sustain those kind of losses and when do you expect to be profitable, if ever?

Well definitely not —

Definitely ever. Yeah.

Yeah, it’s happening. So it is a lot of money. And I think this is a really unique situation, because you have — typically, like, an independently funded startup would not be able to go after a capital intensive initiative like this.

And so what we’ve seen is that the only companies out there who are willing to invest the necessary capital, kind of the activation energy to make this new technology exist, are a company like the major tech companies who can afford to invest in something on a longer time horizon or an automotive company that understands that this is the future and wants to be on the right side of history when things inevitably shift to people being driven by their cars rather than driving them.

On the other hand, we do feel a really strong sense of, I guess, gratitude towards General Motors to getting us to where we are. And we want to return that capital and make sure that GM is wildly successful, financially for making this bet.

So we do have an eye towards profitability. And one of the key factors there is driving down the cost of the vehicles, which we’re doing through investments like building our own chips and machine learning accelerators to take costs out of the vehicle, and then also increasing the scale, which actually makes the cost per ride much, much less.

And so if it is the case that this is really hard to do well, only a couple of companies crack it, and you have that opportunity to go after that market, just in the US, that’s worth a big investment up front, I think, if you have conviction that we can get there.

I mean, my pitch is, like, I probably was never going to buy a GM car. Not because I have anything against GM. But just, like, I don’t want to buy a car in general. It’s annoying to have a car in the city.

But if you guys come up with some sort of subscription program for, like, I don’t know, a couple of hundred bucks a month and you give me 40 miles of rides, you’re going to make $200 a month off me for the rest of my life. So, I mean, to me it just seems obvious that there’s a lot of money in it. It’s just insanely expensive to get there.

I think a lot of people are going to say, why do we even bother owning a car? As long as it’s there when I need it and it can do everything I want to do with a car, I’m not going to own a car. Or at least I won’t have so many cars in my driveway.

I think one thing that we’re sort of very bad at as people who write and think about tech for a living is thinking about second order consequences. Right. Like, the car was invented, and we didn’t know it at the time but it was going to sort of invent the suburb and change the entire fabric of American civic life.

What are some changes that you think this era — if you are correct and autonomous vehicles are coming to cities all over the country, what other effects do you think that could have beyond sort of the obvious, like, it’ll change the way you get to work?

Well, I think from a city planning perspective there’s a lot of things that, I think, if you had a clean slate we would do differently than what we did decades ago when most of our city plans were laid out. I don’t think we’d have as much space for parking cars, like inventory or garage inventory on the streets for people.

I think we’d do a lot more public transit and walking paths and biking paths as a general matter. And I think that AVs will make us a little bit easier for us to do that in the future. If cities do decide that certain parts are AV only, just like the streets in Europe that don’t allow cars, walking only. If there are certain streets that are AV only, what that will let you do is get a lot more throughput for a lot less square feet of city space.

If you have roads that are only AVs that can travel in caravans, like maybe just three feet apart going at high speed, you can fit, maybe three of them in every two lanes of city traffic that you use today. And so I think we’ll be able to reclaim a lot of that space for other purposes, while still being able to move a large number of people through our cities per square foot of road area that we allocate.

So you think there will be some highways or something where it’s like, this is the robot only highway. You can’t go on that if you’re driving the car.

Yeah. It depends. If a goal of a city is to optimize the efficiency of its transportation system, the benefits of AVs is that they can sort of go at high speed on these AV only highways, but they can go point to point. They can show up at your doorstep, which a bus system won’t do and mass transit won’t do. So it’s a pretty powerful tool if we design cities in the right ways.

And one final question that I always like to ask CEOs, which is, if five years from now we call you up and things have not gone according to your plan, if Cruise was never able to reach mass adoption, if GM has shut you down to save money, if none of this comes to pass, why will that have been in your best estimate?

Is it because regulators didn’t make concessions to the industry? Is it because the hardware costs just never come down? If you fail, what do you think the reasons for that will be?

Well, I’m very confident in the cost coming down. I’m confident in the capabilities of the technology improving. I’m confident in how attractive and preferential this is going to be, in terms of a ride experience compared to the status quo. And I’m confident in the safety performance.

And so if it doesn’t get there, it would be for something that is valid but maybe not in the best interests of society. Like maybe we do struggle with how to rationalize collisions caused by robots instead of humans, even though there are fewer of them. Like, that could cause delays or cause, like a shift of this deployment to going to other countries first.

In particular, if you at China, they’re wide open. The government gets it. They’re going full speed. I think that is a risk to the US economy if we’re caught a decade behind on that. Or it could be, if not for that, then maybe one of these other forms of transportation, like electric vertical takeoff and landing plane comes in and we’re not really talking about cars anymore.

But, I mean, you tell me. Like, it really seems like at this point we’re past the critical gate. We’ve gone from R&D to these products scaling out there in the field. And it just seems very obvious that this is the way the world is going. And it’s a matter of time. And I think from a pure system of constraint standpoint, unless we as a society do something to block this or slow it down, I think it’s going to happen really soon and I think we’re all going to benefit.

I think that could be true. I mean, I also felt that way six or seven years ago. So maybe I was just too eager. But when I have friends come in from out of town or colleagues from other cities, one of the most reliable things they say about San Francisco is like, oh my god, there are so many self-driving cars. I didn’t realize these things were so ubiquitous in San Francisco.

Also, like, the only thing that prevented your enthusiasm from becoming a reality was just that the technology hadn’t caught up. Which I think happens to tech reporters all the time. We are constantly seeing stuff that is just not ready to be deployed. But now it actually is being deployed.

And I don’t know. I think the minute that people can take these taxis, and they’re cheap, they’re going to do it. I think the minute you can own a self-driving car, that becomes such a status symbol. Think of all of the people who are going to be thrilled to be like, oh, you’re still driving your car?

[LAUGHING]

Sucks for you. So the market is ready. You just got to get more of these things on the road, Kyle.

On it

All right. [LAUGHING]

Kyle Vogt, thank for coming.

Thanks for coming.

Yeah. Thanks for having me.

[ELECTRONIC MUSIC PLAYING]

Hey. Before we go, we wanted to say thank you to everyone who has sent in voice memos for our segment, Hard Questions. This, of course, is our segment where we hear your tech dilemmas related to ethics and technology, points in your life where you’re struggling to decide what to do with respect to tech. We’ve gotten a lot of great submissions already, and we’re eager to hear more. If you have one, you can send it to us at [email protected].

Hard Fork is produced by Davis Land and Rachel Cohn. We’re edited by Jen Boylan. This episode was fact checked by Caitlin Love. Today’s show was engineered by Alyssa Moxley. Original music by Dan Powell, Marion Lozano, and Rowan Niemisto. Special Thanks to Paula Szuchman, Pui Wing Tam, Nell Gallogly, Kate LoPresti, and Jeffrey Miranda. You can email us at [email protected]. And if you see us in a self-driving car, say hi.

[MUSIC PLAYING]