Super Hacker George Hotz: I Can Make Your Car Drive Itself for Under $1,000
Comma.ai aims to bring plug-and-play autonomy to the masses.
George Hotz, known online as GeoHot, became one of the world's most famous hackers at 17 when he was the first person to break into the iPhone and reconfigure it to be compatible with providers other than AT&T. He was also the first to jailbreak the PlayStation 3, allowing users to play with unauthorized software.
Now this 28-year-old technical wunderkind is up against Waymo, Tesla, Uber, and most of the auto industry in the race to build the first fully operational autonomous vehicle.
"I want to win self-driving cars," Hotz told Reason. Whereas Tesla and Waymo are developing complex systems with expensive LIDAR and other sensors, his company, Comma.ai, is trying to bring plug-and-play driverless technology to the masses. "We're running it on a phone," says Hotz.
He's taking an approach drastically different than his well-financed competition, and is operating with $3.1 million in seed money. Comma's dozen-member team, which works out of a residential house in San Francisco, has built technology that takes over the existing RADAR and drive-by-wire systems in modern cars, incorporates a smartphone's camera and processor, and then makes the car drive itself.
"Google is going to lose because there's no market for a $100,000 system," says Hotz. "For us, we're just going to push the software update. And then—boom—you don't have to pay attention anymore. Done."
Hotz has a history of taking on tech titans, with mixed reactions. After the iPhone jailbreak, Apple co-founder Steve Wozniak sent him a letter of congratulations. After he hacked the PS3, Sony sued him. Hotz quickly became a cause celebre of so-called hacktivist groups including Anonymous and LulzSec. They attacked Sony's network, despite Hotz's protests, igniting a firestorm of legal and media scrutiny.
Comma.ai is Hotz's attempt to take on the big players in a new way. The company makes an app called Chiffr that turns a user's phone into a dashcam and monitors its GPS and accelerometers. Now Comma is launching Panda, an open source, $88 dongle that plugs into the car, links it to the phone, and puts out fine-grain detail about every aspect of a drive. Hotz ingests all the data from Chiffr and Panda users and feeds it to his artificial intelligence system, which then learns how to drive.
According to Hotz, this approach gives him significant advantages over competitors such as Waymo. His network is entirely crowdsourced and running on some of the most popular cars in the country. He doesn't need to build another expensive, specially designed vehicle and employ a trained driver and an engineer every time he wants to add another data point. And all his data come from real-world experience.
Hotz says Waymo and others take a rule-based approach to driving that doesn't reflect the reality of how people operate cars. "The humans ain't changing to match the self-driving spec," he says. "In order to really get access to the full, diverse spectrum of what driving is, you need a huge crowdsourced database."
While Tesla's training model is more closely aligned with his, Hotz says the company will similarly be restricted to the high-end market. He got into a public spat with Tesla founder Elon Musk in 2015, after Hotz says the mogul changed the terms of a deal for him to build a better vision system for Tesla's Autopilot than the one supplied by partner company Mobileye. Musk claims Hotz bragged that he could build a better system, and then welched on the bet.
"All I said was I could build a better vision system than Mobileye, myself, in 3 months," replies Hotz. "And I kind of did that."
Despite his new gig, however, Hotz maintains a hacker's spirit. He has a bounty program that pays out $10,000 to customers who are able to port Comma's software and enable it to tap into their car's driving systems. Hotz, however, denies that this constitutes hacking, even though he offers the bounty instead of contacting carmakers to access their APIs.
"They're just such bureaucratic organizations that it'll take me less time to reverse engineer it," says Hotz. "I'm not even asking for forgiveness. There's no permission to be asked. It's your car."
Comma.ai's first attempt at consumer hardware was the Comma One, a full self-driving kit for $999. But Hotz aborted the product launch after getting a letter from the National Highway Traffic Safety Administration detailing safety concerns and regulatory compliance issues.
Comma's new self-driving kit for under a grand is called the Comma Neo. Hotz says it's similar to a Gear virtual reality headset in that it's powered by a smartphone. The Comma Neo mounts, cools, and connects a OnePlus 3 running a custom operating system to the car. Comma's OpenPilot software then taps into the phone's camera and the car's systems to auto-drive it.
In the future, Hotz predicts users will only need the Chiffr app running on their own phone, connected to their car via Panda, to make their vehicles fully autonomous.
"We get to leverage that billions of dollars of investment of companies in a way that no other self-driving company would even think [of]," says Hotz. "It's super exciting when you think about what's possible here. And the fact that almost nothing can stop it."
This is a rush transcript. Check all quotes against the audio for accuracy.
Hotz: We're the Android of self-driving cars, right? Think about Tesla as the iOS.
All these other manufacturers are hopelessly clueless when it comes to self-driving. How do we build something that just works on their cars, right?
Monticello: And so the idea is that you would then be able to plug and play your self-driving system in pretty much any production car.
Hotz: You need some tailoring to the car, but so what we're trying to do, is you take the top 20 cars in America, and that's like 50% of the cars sold. We want to support most of those.
Monticello: And so how does your system work in contrast to other systems that use a lot of expensive sensors like Lidar? You have to have a lot of equipment. You're just tapping into the car's existing capabilities, right?
Hotz: We ship a camera, and we use the RADAR that's built in to the car, and we use the sensors that are already in the car. Yeah, those systems are just not shippable, they're not feasible for passenger cars because there's no market for an a hundred thousand dollar, quasi-self-driving, not even full self-driving. It's just a money sink. So you know, this is just done by people who do not think about the feasibility of anything. So that's why we use cheap sensors.
Hotz: Thousand dollars, not a hundred thousand.
Monticello: And you think that that's sufficient to be able to accomplish everything that the other self-driving companies are doing with their Lidar systems?
Hotz: Oh, absolutely, right. The truth is, even with those Lidar systems, nobody is at level four, right?
There are 6 levels of self-driving ability. Today's offerings such as Tesla's Autopilot and Comma.ai's OpenPilot max out at a level 2, which means the car can auto-drive, but the driver constantly needs to pay attention and be ready to take over. The Holy Grail is level 4, at which point the car can handle all driving on most roads, the steering wheel and pedals can be removed, and the human becomes a passenger who can read, nap, or take in the landscape.
Hotz: But in order for that to be okay, your system better be a good bit better than humans. Not perfect, it's never, ever going to be perfect, but it's got to be better than humans. No system is there yet.
Monticello: Before you started Comma.ai, which is your current company, you gave an infamous interview to a Bloomberg reporter, where you took the reporter on a ride in this self-driving car that you had worked on for a month, and you just got to work that morning. So are you crazy, or, how did you know that you weren't going to die when you brought that out onto the road?
Hotz: Well, so it's the same safety model that we have today. The safety works like this, it's two-fold. One, the second you touch either the gas or the breaks, the system just stops doing anything, right? So you always have that, that user override.
But the other one, that you even need more than this, you need to make sure that the car's never going to do anything so quickly that you can't respond, and the way you deal with this is torque limits.
It's never going to go like that. It can't. It can't. There's hard limits preventing that, alright? And actually, I didn't even write them. I'm using the hard limits that were built in to the car.
Monticello: Gotcha. So the car has its sort of, own built in system where the power steering, which is mostly electric now, not hydraulic, will not be able to jerk it in front a truck or something like that.
Comma's first consumer product is an app called Chiffr, which turns your phone into a dashcam and uses its GPS and accelerometer. Now the company is launching Panda, an open source, $88 dongle that plugs into your car, links it to your phone, and puts out fine-grain detail about every aspect of your drive.
Hotz: You can actually use a Panda as the bridge between OpenPilot, which is the software that'll drive your car, and the car itself. Panda is a universal car interface. So when it's used by Chffr, it's read only, but when it's used by OpenPilot, it's connected over USB and it can actually drive your car.
Monticello: Gotcha, Okay. So that's what taps you into the car's system so you can move the steering wheel …
Monticello: … Okay. And so, what kind of cars … can I use this on my '89 Volvo or like, do I need a newer car?
Hotz: For Chffr you'll get something from every car manufactured after 1996. Now, if you want to get the fancy self-driving stuff, well, there's only a few cars that OpenPilot supports right now. We're going to start getting a lot more … like I said, I want … In 2018, I want to support the majority of the top 20 cars sold in America.
Monticello: And right now is there a baseline of cars that have, drive by wire, brake by wire, gas by wire, that you can tap into?
Hotz: So we support, it's Hondas and Acuras right now. We just bought a Toyota Prius, we're going to be doing all the Toyotas this year. A user, one of our users has ported it to the Chevy Volt. We have a bounty program. As soon as he cleans up the code a little bit and merges it in, we're going to pay him out $10,000. We have a bounty up for the Ford Fusion for somebody to do that. We have a bounty for the Tesla Model S, BMW i3. We want to support them all.
Monticello: Gotcha. So you're kind of hacking into the car makers' systems, essentially. In order to-
Hotz: I wouldn't think it about it like that. You're certainly not hacking into anybody's systems, because it's your car, right? You're not hacking. You're not changing the firmware, you're not jailbreaking. You're just finding … every car has a different API to get to the steering wheel, the gas, and the brakes. So you know, it's about finding the API's in the new cars.
Monticello: But the car makers, they don't want you to be able to have access to that, right?
Hotz: They don't care. Care manufacturers sell cars, right? What do they care?
Monticello: So why don't you just call up Chevy and be like, "Hey, can I get access to the Volt?" Why do you pay a $10,000 bounty?
Hotz: Let's even say they wanted us to get access to it, they're not opposed to it. I think they probably would rather us do it than not because at the end of the day, you sell more Chevys.
But you know they got to get it through their lawyers, and you're talking to the business development guy, and you know he's a level five, he can't be in the meeting with the level fours; and they're just such bureaucratic organizations that it'll take me less time to reverse engineer it, than it will to get legal approval. Then, once you finally do get legal approval, oh, you can be sure it's coming with a 10 page contract that says, "You can do this, this, this, not this, and this, and not this, and definitely not that, don't even think about that. We'll sue you into liquidated damages," and I'm not playing with that. We're going to reverse engineer it. We're going to make it open.
Monticello: Right. So you're asking for forgiveness not permission, basically?
Hotz: I'm not even asking for forgiveness. I'm just not … there's no permission to be asked. It's your car. I don't need permission to use my things that I buy however I want, right? I think a lot of people should get out of this mindset of thinking, and I've always said this with like the jailbreaks and stuff. Get out of the idea that you're buying the Play Station 3 and it's a closed box, and actually SONY owns it and you don't. No, no, no, no, no. Stop thinking like that, right.
Monticello: So, you've described Panda as basically like a Fitbit for your car. You get all this information, all this data coming from your car about the driving and the driver. But you're also feeding this information that you crowdsource because you have all these different people who are voluntarily putting this software in their car. And you're feeding that into an AI, which then is learning how to drive. So, how do you think that this approach is different and why is it better than what Waymo's doing, what Tesla's doing, these sort of bigger players?
Hotz: Tesla understands more about what we're doing, but let's look at Waymo. So Waymo, when they want to figure out how to build a self-driving car, they sit four engineers down in a room and they talk about, "Okay, well we come upon a stop sign, we know that we should stop and we know that we should stop this distance, and wait, can you get the DMV handbook out? Let me see. Okay, we have to signal 15 meters—," That's not what driving is. This is the same failure of computer vision for many, many years.
When people wanted to build a detector to see if there was a chair in an image, they would write out the definition of a chair. They would say, "Okay, well a chair has a back, and it has four legs, and it has … Wait what about a bar stool? Is that a chair? Well I don't know we're going to investigate bar stools." This is how a lot of people are thinking about this problem and it's absurd.
You want to figure out if there's a chair in an image or not? You get a million images of chairs, a million images without chairs, and you use machine learning to train a classifier that says chair or no chair. Right? So that's the same approach we take to driving. There is no rigid specification or definition of driving. Driving is just what people do when they drive, and in order to really get access to the full diverse spectrum of what driving is, you need a huge crowdsourced database, and I mean, that's what we're building. Then we'll just learn what it means to drive from people who actually drive.
Because you've got to operate with humans, right? It's some leftist utopian fantasy, "Oh we're going to wipe the roads of all the cars, and we're just going to have electric cars that interact," look at a Google commercial if you want to see this. It's not what's going to happen. What's actually going to happen is there's going to be a bunch of self-driving cars, bunch of human-operated cars, and they all got to interoperate, right? And the humans ain't changing to match the self-driving spec
Monticello: So you're saying that one advantage that you have is that you're not using Google engineers, and Google cars, and everything to go out there. You're actually looking at what real people are doing in the field.
Hotz: I don't think about what driving is, I say, "Show me what driving is," right, and that's the difference.
Monticello: Gotcha. So that's more expensive than, if they want to add a new driver, you can add pretty much limitless people who have Panda essentially.
Hotz: We have a huge network, we have thousands of people on our network.
Monticello: And then the other advantage is that they're taking a rule-based approach. So like you're saying, they're defining a chair, rather than letting the computer figure out what a chair is. But don't they also use AI and sort of deep learning techniques? What's the difference in their systems?
Hotz: Well, so I mean, everybody is moving towards AI and deep learning, and obviously Google is going to get to level four before we will. Nobody doubts this. But the reason Google is going to lose, is they're going to get to level four with an $100,000 system, and then they have a whole thing to go through to deal with, "Okay, how do we actually get this in the hands of people."
If we get the AI problem solved a year after Google, Google's still going to be sitting there thinking, "Okay, well, maybe we could finance it to people." For us we're just going to push the software out there. I'm going to push the button and be like, and Elon knows this too, we're just going to push the button and then boom, all these cars are level four self-driving. We got the insurance company to underwrite the policy, you don't have to pay attention anymore. Done. And that's the Comma.ai plan. We're just going to solve the problem as we go.
Google can only iterate and build on top of what other companies have already done. Was Chrome the first web browser? No, but it was the best. Was Android the first non-iOS operating system for smart phones? No, but it was way better than Symbian, the Blackberry OS, but it's not … Google in some ways is not an innovative company, and when they try to innovate, you get things like Google Glass. The google self-driving car project is like Google Glass. I say, Google is the Xerox Parc of the self-driving car industry. All right?
You see all these good ideas, you bring them out, you fan them out to other companies.
Monticello: So in this analogy you would be the Steve Jobs going into the Xerox campus, getting the graphical user interface and everything like that. Seeing that they're messing it up-
Hotz: I'm a lot more like the Bill Gates. Maybe Elon's the Steve Jobs.
Monticello: You want to be the Bill Gates?
Hotz: Yeah, I'll be the Bill Gates.
Monticello: Wow, nobody ever wants to be the Bill Gates, that's cool.
Hotz: No, no, no, Jobs is … you know, he's just … yeah.
Hotz: I'm not an Apple kind of company, I'm a lot more like a Microsoft kind of company.
Monticello: What do you mean by that?
Hotz: They practically are more art than technology, right. They do have good technology backing it up, but for me, the end result is technology. Technology is the only thing in history that has had this compounding effect on human wealth. On everything. We live in houses, they're heated, we have lights. This is all technology and that's what I want to be a part of a lot more than well, art.
Monticello: And so you don't care about making things beautiful? Like in the same sense that …
Hotz: Yeah, I don't care about make things beautiful. I'm very, very anti-advertising. The idea of advertising like we're almost going to manipulate people into buying our product. And then you have like atrocious companies like Facebook. Facebook, you're not a user of Facebook. You're Facebook's product. Facebook sells your attention to advertisers, right?
Monticello: Facebook uses you.
Hotz: Facebook uses you. That's just how it works, it's not even in Soviet Russia, it's in Palo Alto, California—Menlo Park, right? Yeah, no, I never want to be that kind of company. We build technology and, and you know, we open source as much as we can as well. Because it's not about, "Here's the pie, I want my slice to be bigger," It's about, "We can build a really big pie, let's build.
Monticello: So Apple just recently announced that they are sort of scrapping their hardware initiative and they're just going software, right? So does that worry you?
Hotz: No. Again, when you think about Apple as a company, they're going to play a lot more in the Tesla arena. What about that other 70%? Most smartphones in the world are Androids. Most self-driving cars in the world, in five years, will be Comma.ai.
Monticello: So you basically want everybody who's watching this to be able to turn their car into a self-driving car in five years?
Hotz: Yeah. I really emphasize, no regressions. We're going to work to make this better, and better, and better, and more, and more consumer friendly.
Comma.ai's first attempt at consumer hardware was the Comma One, a full self-driving kit for under a thousand dollars. But Hotz aborted the product launch after getting a letter from the National Highway Traffic Safety Administration detailing safety concerns and regulatory compliance issues. Hotz then aborted the launch, drawing outrage both from those who saw the event as a government overreach, and those who thought the government had valid concerns and, instead of addressing them, Hotz was just picking up his ball and going home.
Hotz: When you look at what we actually did, not what the media said we did, so you know maybe we didn't release the actual, $999 Comma One on Kickstarter, but we released all the plans for you to build pretty much exactly what the Comma One was, and we released them for free. And you could build them for way less than $1,000 and self-driving car, the car can certainly drive itself, so yeah.
The thing about the Comma One, it was never going to be a mass market product. We were maybe going to sell 1,000 of those to car enthusiasts. So you know, let's even say we're making $400 profit a unit, that's $400,000 profit to Comma.ai. That's more than … the lawsuit would cost me more. So it just make economical sense for us to continue with the Comma One as planned. We had to think of a more clever way to avoid a lawsuit, but still basically achieve the same outcome and that's why we open sourced it.
You want to start a company in the self-driving space? Literally rip us off. Literally, please, it's MIT licensed. Just download our stuff, build it into your cars, don't give me anything. You look at the world like that, you realize open sourcing is the way to go.
Monticello: So what are the benefits to you of open sourcing all of your plans and your products?
Hotz: We can eventually work up to a consumer product, and we can kind of let the market figure out like … if we don't build the best localizer, if somebody else builds an open source localizer, we'll use that.
Comma.ai is an ecosystem. We're going to be the constellation of things surrounding Android just as Android's an ecosystem, right? It's Google, it's the app developers, it's the phone manufacturers. We're the Google for the ecosystem, but I want to see the Samsung's, and HTC's and the …
Comma's new self-driving kit for under a grand is called the Comma Neo. Hotz says it's similar to a Gear Virtual Reality headset, in that it's powered by a smartphone. The Comma Neo mounts, cools, and connects the phone to the car. Comma's OpenPilot software then taps into the phone's camera and the car's systems to auto-drive it.
Monticello: And there are people out there who have these, and they're actually posting on YouTube.
Hotz: There's 99 people right now running OpenPilot. We have thousands on the Chffr Network, and yeah. We'll see 10x growth in the next year, we'll have 1,000 where we're constantly growing. Right now, I'm not sure how many cars Waymo has running, so they might be the second largest network, but I think we're pretty close, us and Waymo, and then Tesla obviously has tens of thousands of these things.
Monticello: So the Comma Neo is a full self-driving system?
Hotz: Maybe someday—I probably shouldn't say this but like, maybe someday all you're going to need is Chffr and Panda right? You run the Chffr app on the phone, it talks to your car with Panda, and maybe it's just your phone that starts to drive your car.
Monticello: But right now you need a special phone.
Hotz: Right now you need a special phone running a special operating system, you need a cooling solution; because phones aren't quite at that point yet, but every year they get better, and every year, billions of dollars is poured into making smartphones better.
Hotz: And we get to leverage that billions of dollars of investment of companies in a way that no other self-driving company would even think, "Oh, they're running it on a phone?"
Hotz: Yeah, we're running it on a phone. Phones are really, really good. It's super exciting when you think about like, really what's possible here. And the fact that almost nothing can stop it.
And that's why it's opensource. It's not even like … You don't have to work with us, you don't have to sign some sketchy back room business development deal. It's opensource and MIT licensed. Please download it and use it.
Right? We're not one company, what NHTSA … you gonna issue a … can't do a recall, right? What are you recalling? We didn't sell you anything. Going to recall software? Yeah, good luck.
Monticello: So you mentioned Elon Musk, you guys got into a little bit of a public spat, I guess? Maybe it wasn't that negative. But in 2015 you kind of criticized him of his technology-
Hotz: I didn't criticize his technology, I criticized his choice of a partner in Mobileye.
Monticello: Okay, yes. You criticized Mobileye.
Hotz: And I criticized the fact that he renegued on a contract, yeah, not cool.
Monticello: Well this is what I want to ask you about.
Monticello: Because you guys have kind of different stories about this. He says that you wanted to bet that you could outperform his system.
Monticello: Right? And then you made this bet, and then you ultimately didn't want to agree to the bet.
Hotz: Well …
Elon Musk needed a vision system for his car, wanted to replace Mobileye, and I'm like, "Yeah, sure, I'll do it. Give me a new contract, and I'll get you something incredible," you know.
Elon changed the deal at the last minute, he changed it from a contract, which has rigid completion criteria, to basically an option for him to buy. He's kind of like, "Build it, and then I'll pay you if I like it." Well, that's not worth it for me, right?
Monticello: At the time, he said in an interview that you were underestimating the technical challenges involved, right? And he said, quote, "It's not like George Hotz, a one guy in three months problem, you know. It's more like thousands of people for two years." Well, it's two years later, right? Who's closer? Who's going to make this happen?
Hotz: Tesla's always going to be a little bit ahead of us because they started earlier, right? All I said was I could build a better vision system than Mobileye, myself in three months, and like, I kind of did that, all right?
So yeah, the whole self-driving problem, no. Of course that's going to be a … I'm not going to say thousand person problem, Comma.ai's a 12 person company. It's been about two years, I think it's going to be at least another three in order to fully solve the problem.
I've come to respect a lot more of what he said there, about starting a company, you need validation, unique engineering practices. He's right. I'm fine. At the end of the day, you know, I'm an egomaniac. Big ego. I love being right. But not at the expense of being wrong.
Maybe it was a legitimate misunderstanding. You should never attribute to malice, what you can attribute to a misunderstanding, and I have a ton of respect for Elon Musk. I really do, it's nothing like how I feel about Ford. Ford is a … Elon is doing so much good for the world when it comes down to it, I really do regret if I inconvenienced him in any way. I mean, you know, you punch me, I'm going to punch back a little, but like you know.
At the end of the day, Tesla's not our competition. We would love to see Tesla succeed. I would personally just love to see Tesla and SpaceX succeed because that's the kind of world I want to live in. It comes back to the pie argument. I'd rather Elon Musk have a much bigger share of the pie than me, as long as the pie tastes awesome.
Ford, on the other hand … like, those companies need to lose because what good are you doing for the world? You're holding the world back to make money, all right?
I'm a big fan of capitalism, but you gotta avoid rent seeking. You gotta avoid rent seeking and monopolistic behavior and where you're just making money because you have kind of …
Hotz: Elon is doing absolutely great things for the world, and if his capital was better deployed somewhere else besides paying me, then I think he made the right choice.
Monticello: And so you also bet him, last year, that your system could successfully navigate the Golden Gate Bridge before he could, with pedestrians.
Hotz: That one was played up by the media a whole lot. We can both drive a-
Monticello: I'm the media so I'd like to play it up some more, so, yeah.
Hotz: We can both drive a across the Golden Gate Bridge, without an intervention today. Let's just say that.
Monticello: So you both win? Is that …
Hotz: We both win. Everyone wins.
Monticello: Big pie?
Hotz: You know who doesn't win? Ford.
Monticello: I have to say though, I also really like your pie analogies, because I'm also from New Jersey, so I love talking about pizza. So I appreciate it.
Hotz: Yeah, good pizza, yeah.
Monticello: Last thing about Elon Musk, he's become sort of an alarmist about AI, he wants international bodies to regulate it, he thinks it's going to destroy humanity if we don't do this. What do you think of that?
Hotz: I'm not sure it presents any particular threat that previous weapons have not. Some people think that the AI is going to get out of control, there's going to be a Skynet Terminator scenario. This to me seems highly unlikely. And a lot of researchers in the field will agree with this.
What does seem likely to me is humans will get their hands on the AI weapon, and do what humans have done when they've gotten their hands on any other weapon that they got in the past. "We don't like those people, how can we use it to mess with them? It's just about how the people are going to use the technology.
Now, if Elon Musk is calling for an international arms control regulation, he should look at history and see how well that has worked.
Monticello: All right, so you're skeptical sort of intervening. You think it'll self regulate, or people in the field are better prepared to respond …
Hotz: I think that, yeah … listen to people in the field. I certainly think that it's incredibly premature to talk about any sort of government level AI regulation. It's not going to kill me tomorrow, there's things that are a lot more likely to kill me tomorrow, right? Like driving.
Hotz: No seriously though. Yeah the AI alarmism … I mean these are problems that we should be thinking about, but to believe that somehow in the next two years Google is going to accidentally build Skynet, is absurd.
Monticello: And so, you're not an alarmist about AI in that sense.
Monticello: But you have raised a few eyebrows because you've said in the past that, you think that AI is going to take everybody's jobs-
Monticello: … and you cannot wait for that to happen-
Monticello: … you think this is a good thing, right?
Hotz: Of course.
Monticello: But people have been saying that since before the industrial revolution, right? Elevator operators were saying that like the elevators going to take all the jobs. So what's different now?
Hotz: The industrial revolution replaced man's muscles, right? Very few jobs today, at least in first world countries, are people doing work with their muscles, right? You even think of like traditionally muscular work like mining. It's not people with pickaxes, it's people with ladders operating hydraulic machines. So the industrial revolution replaced man's muscles, well what's left? Man's mind.
So you know people have always said, "Oh, it's going to replace all the jobs." The car really did replace all the jobs and the horses, right? There aren't that many horses left. The horse population of the world peaked in 1917, this is just a true fact. So I think it's the same thing with humans and their jobs. The last refuge is man's mind, and once you start building minds that are super human, well, you know, there are no more jobs for humans. Great. This is a great world. Isn't the endgame of technology in general?
Didn't you sit there and be like, "Man you know, I'm really sick of hoeing this field. Wouldn't it be great if there was some mechanized hoe?" And then you've got to sit on it and operate it, and now, you go to modern farm and it's got a … it's a tractor with an Apple-ized GPS, I think like, maybe some union is why the person still has to sit there, but they just don't do that much. They watch the thing drive … and then this is a great world, this is a great world.
Monticello: So Comma will ultimately be a mechanized hoe company, is what you're saying.
Hotz: We work where are talents are.
Monticello: So, what does a world like that look like? Where nobody has a job except for AI programmers?
Hotz: Oh, man. I think a month in advance. I think … I think … I got a date tomorrow night, I'm excited about that. That's about as far in the future as I think.
But I really … maybe you want to date an AI at some point. I mean that-
Monticello: That's the only thing that's holding you back.
Hotz: I think that would be really cool, right? To replace … that gets awfully [inaudible]. People outraged … people are outraged about everything today. But yeah, I want to date an AI, of course.
Monticello: But you don't see this as a dystopian future? You're hopeful.
Hotz: Absolutely, yeah. No, I'm a complete optimist about the future. People are just outraged about things. I don't know. We're living in the best time ever. We are living in paradise practically. When's the last time you had to deal with, "Oh man, I want to go to the coffee shop but there might be lions?" No, we beat lions, right?
Monticello: Okay so, getting back a little bit to the driving. So the data that you're collecting, it's about things like people's locations, their driving habits, a lot of personally identifiable potentially information. Things that would be attractive to stalkers, hackers, PI's, government agencies. Being a super hacker yourself, you know that these things are vulnerable to outside groups coming in, so how are you protecting your customers' data and how do you view that, the privacy issues?
Hotz: It's not … here's the thing about privacy in general, and at the end of the day people might yell at me for this, but this is really how you have to look at the world. The NSA's a big problem, and the reason the NSA is a big problem because they have privacy and you do not. They have privacy, you do not. Be nice if we both had privacy, but I don't see that happening either. So maybe the real solution is, we don't have privacy, but they don't have privacy either.
Now how do we actually protect people? We use industry standards, same stuff as everyone else.
Monticello: So you just don't value privacy as sort of a … in your own life, like would you be upset if privacy went away?
Hotz: Okay, so first off let's talk about personally identifiable information from Chffr, right. We do not record either the microphone or the front facing camera. I don't want pictures of your face. I don't want to know your name. I don't want to know your age. I don't want to know your gender. This is not your data that I'm interested in, this is your car's data that I'm interested in.
I'm not trying to sell your data to advertisers, right?
Monticello: Right. So I take your point, but I would say at the same time, the NSA makes a similar argument all the time. We're not listening to your phone calls. We're not looking at your Facebook profile, we just have metadata.
Hotz: Okay. Cool.
Monticello: But if you look at somebody's metadata, you can find out where they work, if they have kids, where they go to school, where they go to church, what their religion is—
Hotz: I don't have a problem with the NSA, just make the data public. I would have no problem with the NSA in metadata collection, I don't think these things are necessarily bad, I just don't like the idea that one organization has a monopoly on it. Let's make the data more public.
I don't want to have a monopoly on data. This is the old way of thinking. What if we could open our data up more, and really think about it, not as like, "Facebook owns this data, Google owns this data," but we all collectively own the data and you're contributing to a big collective pool of data.
Hotz: Now it's a market, it's not like, "Oh we're all going to do this for smiles and roses," or whatever communists have, but no we're going to do it for you know, [inaudible] the market. Create this data, contribute to this, all the data combined is a whole lot more powerful than any piece of the data alone and I think we can do incredible things with these sort of data sets, right? So yeah. Like I said, my problem with the NSA is not that the metadata's being collected, it's that they collect it and I can't see it.
Monticello: That's a pretty radical view, right? Like I don't think there's a lot of people who'd be very comfortable with all of their data being online.
Hotz: There's a difference between all of your data and all of your metadata, right? Right now data in terms of what you're saying on a phone call, maybe if you even make a phone call, but like where you drive? That's already public and we have to accept this, right? I could hire a PI to follow you around. It's not like it's … and that doesn't really violate any expectation of privacy. I think when you step out of your home into the world, you're entering a public space. This is what the laws say.
Well I mean, Google did it all with street view, right? And if you're outraged about street view, come on. And then Google spends all this money to do this, and then provides it as a free service for everybody. I love street view, right. You gotta think about whether the benefits just outweigh the costs.
Monticello: What regulatory issues do you see having to tackle as you kind of move forward into a semi-self-driving or a full self-driving system?
Hotz: I mean I've seen this NHTSA regime be a lot better than the last NHTSA regime. Mark, whatever his last name was, the head of the NHTSA, talked about like how his dad died in a motorcycle accident and how this is a personal crusade for him, and how, "If Tesla autopilot has two more accidents, we're pulling it off the road," stop. Stop. 30,000 people are dying a year, stop with that rhetoric. Let's use statistics. Let's use math. Let's use engineering, and let's solve these problems.
You're not going to be able to constructively prove this is what you do at a stop sign. The arguments that should be made for these sort of things, and they're the kind of arguments that Tesla's making, are statistical arguments.
If it turns out that people with the system get into 50% less accidents than people without the system, the system is good.
Monticello: So is what you're saying, that you're arguing for outcome based, versus process based regulation.
Hotz: Exactly. Yeah.
Monticello: So you said NHTSA, which is the safety administration, like are they actually adopting that sort of standard?
Hotz: Yeah, they do, they do. In fact the NHTSA Tesla Autopilot letter, which came out after the Comma One, was very encouraging for it. This is basically what NHTSA said, Tesla said "Here's our crash statistics before Autopilot, here's our crash statistics after," and you know at least rear ending collisions were reduced by 30%, as they should be. So this is … it's a great thing to see that NHTSA accepted that argument.
I'm certainly not fighting the regulators. I'm fighting Ford way more than I'm fighting the regulators. The regulators, it'd almost be criminal if you were to ignore technology that's saving tens of thousands of lives, right. Just anything that can work, outcome based, should be what regulators look at.
Monticello: And do you identify as anything politically?
Monticello: You just are?
Hotz: Highly tribal. No interest.
It's all about the war with nature—
Monticello: Sort of like an Adam-Smith-state-of-nature kind of guy?
Hotz: Someday, so I can eventually move into virtual reality and live in a great artificial world that everything is awesome in, and I can go on cool quests every day and slay dragon beasts, virtual dragon beasts. You know, no real dragon beasts have to die. Even vegans are okay with this, right? Let's stop fighting.
Monticello: At the bottom of your webpage you have the text, "we are actually about to change the world."
Hotz: Yeah. That mostly is a stab at other startups who use this rhetoric of, we're going to change the world. Right, yeah, I mean you know, Facebook's changing the world too, yeah, you know, hey.
Hotz: Hitler changed the world, you know, hey. It's not always right. Like we're changing the world, we're not like using this to sell you egg timers. Or whatever it is, whatever it is. Let's not-
Monticello: So you're serious about it. You're actually changing the world.
Hotz: We're serious about it, we're actually changing the world. That's mostly what it's about. Again it's like, I think of the world as like nature, and we're trying to build more accurate localizers and then better sensors, and put more data on the internet, and grow the internet, and that kind of stuff. That's what I mean by actually. Not, you know, I have a line in a song. I have a SoundCloud. Soundcloud.com/tomcruise. "Changing the world is just a euphemism for how can I get you to give more stuff to me," and I think that's how a lot of people view it. Like I say, I have enough stuff, a lot of stuff.
Monticello: But you are looking forward to a future in which you can sort of plug yourself into this virtual world and virtually kill beasts and that sort of thing, right?
Hotz: Absolutely, and here's my argument for this. I have the best phone. I have the best phone you can buy. And in order to get a better phone, I'd have to spend millions and millions and millions of dollars. I'd have to practically start a phone company, because you can't buy a better phone than this. There isn't like a phone for $2,000 that's any better. So we have to move the world forward in order to get this technology that I want.
I can't sit in my basement and build my own virtual reality thing. We need all the technologies that support that. We need to move the whole ecosystem forward. We need better display technology, better localization technology for head tracking, maybe we should start looking into, like neurological link stuff, so I can feel sensations. It's a huge, it's a bio, it's the biggest project maybe humanity's ever undertaken. Maybe AI's the biggest one, and then AI can build me all those cool things.
Monticello: And is this what open access is contributing to? That sort of communal knowledge gathering?
Hotz: Who is open access?
Monticello: Like even Tesla has gone to open access for their patents and some of their technology. You're open access.
Hotz: Oh, yeah. I'm not … intellectual property. Here's my view on intellectual property. It's kind of like, we had the problem of goods scarcity and a lot of wars are fought over goods scarcity. And like, how many potatoes are there. I gotta kill you because I need more … we build this intellectual property thing, it's actually not scarce. Intellectual property is not a scarce good. And we put … we use guns to make it artificially scarce? Really people? Really? This is the best thing we could come up with? So yeah, no, I'm all about just you know … we shouldn't really view it as like a monopoly on … 'I drew Mickey Mouse first, so Mickey Mouse is mine. You can't make Mickey Mouse porn, definitely not. Definitely not, and we're going to lobby to make sure you can't,' right? Come on.
Monticello: So you want the pie to be big enough that Mickey Mouse porn is a thing.
Hotz: It could be in a small sliver, maybe not a corner of the pie that I go to, but hey, you know, to each their own, right?
Monticello: That is the first example you thought of, though.
Hotz: Well, because that's the classic one. You think of the Sonny Bono Copyright Extension Act, you think about these copyright extension acts that have been passed. Copyright is designed to incentivize the creation of work. Why do you need to extend it to cover works that have already been created?
Monticello: For 70 years. Right, plus the life of the author.
Hotz: It's become absurd, right. And then patents are this absurd … it's not … They're just big companies fighting each other. Maybe its monopolistic, maybe it's not. It's not 'you're protecting your inventor' anymore. It's big companies who need a patent strategy, and … yeah right.
Monticello: Yeah. And just to wrap up here, so your first goal is to get your software good enough where it's a level three system, as you've said. So where the driver can fully disengage and isn't expected to have their hands over the wheel the whole time, right? For certain stretches of road at least, like maybe if you're in traffic, or if you're on the highway. How long do you think that is, how far away are we from that?
Hotz: So our hardware isn't quite good enough to do that now, so we're going to need the next generation of Comma hardware. Which we're looking at, and we're trying to be like the phone companies. We're trying to come out with something new every year, right?
Our AI tech is going to need to be better and then we're going to have to operate this thing in shadow mode for a while, right? Where people are using it but they still have to pay attention. Then if it turns out that all these people paying attention didn't need to have paid attention by statistical arguments, then we'll show these statistical arguments to an insurance company, or to a bank, or you know. Insurance is a pretty simple game. And then we say, 'On these stretches of highway your system is level three, feel free to disengage, you need to be ready to take over in 15 seconds.'
Monticello: So your pitch to the insurance company is, 'Hey, statistically our system is safer than if you let the driver drive themselves. So really you should want them to.' And how long do you think it'll be before … if you had to give a guess, before that'll come online.
Hotz: I think in some stretches of road, maybe two years? And then you see the full battle, three, you start getting more and then five I think the self-driving car game is going to kind of be over and you're going to see who the winners are.
Monticello: All right, so in five years I'll come back and see you and I won't have to drive myself, is that the …
Hotz: Yeah, probably. Probably.
Monticello: All right, well I'm counting on you to make this happen. I'm really excited about it, but I'm just watching, so.
Hotz: I want to make it happen. Hey, I might be wrong. I'm always humble in the face of nature, so we'll see. We'll see. I'm going to try my best.
Monticello: George Hotz of Comma.ai, thank you so much for talking to us today.
Hotz: Thanks for having me.
Monticello: For Reason, I am Justin Monticello.