Laura Bliss is CityLab’s West Coast bureau chief. She also writes MapLab, a biweekly newsletter about maps (subscribe here). Her work has appeared in The New York Times, The Atlantic, Los Angeles magazine, and beyond.
George Hotz, the hacker-turned-founder of an open-source self-driving startup, has a different philosophy of autonomy.
If you follow the feats of hackerdom, you remember George Hotz. In 2007, he was the 17-year-old from New Jersey who made headlines for unlocking an iPhone’s carrier settings—a world first. A few years later, he broke into a PlayStation 3 and got sued by Sony (they eventually settled out of court). In 2015, he got into a tiff with Tesla CEO Elon Musk, who had previously tried to hire him, over vows to build a semi-autonomous driving software that essentially reproduced Tesla’s Autopilot.
That was also the year Hotz launched his own self-driving startup, Comma.ai, which recently raised $5 million in venture capital and employs 14 people out of their San Francisco offices. Hotz says he kept his word to Musk: He claims that Comma.ai’s “Open Pilot” system performs better than Autopilot.
Now 28, Hotz is preparing to scale the next peak of nerd ascendancy: By the end of 2018, he’s promising a semi-autonomous driving system that performs better than General Motors’ Cadillac Super Cruise, which Ars Technica recently called the best advanced driver assistance program on the market. He made the announcement last week in a keynote address at Locate, a mapping and location data summit convened by the mapping startup Mapbox.
The twist with Hotz’s new “product” is that, true to his hacker roots, he’s not selling anything at all. Comma.ai gives away free, open-source software that owners of some of the most popular car models on the market—most Toyotas and Hondas, certain Acura models—can install (and tinker with!) on their vehicles by themselves, transforming a stock human-driven vehicle into a DIY semi-robo-car. Anyone with the right coding skills can do it. “Our slogan is ‘ghost riding for the masses,’ which is hard but not impossible,” Hotz told the Silicon Valley-heavy audience.
That’s a far cry from the fiercely protective approach the tech giants take to self-driving development. Comma also sells various plug-ins and dash-cams that play the data-gathering role other developers have built into their proprietary systems; on the Comma.ai website, you can purchase a $699 navigation system that also makes sure your eyes don’t stray from the road when the car sort-of drives itself (after all, these systems aren’t fully autonomous yet) or a $99 dongle that acts like a Fitbit tracker for your car, gathering its mileage, location, and driver engagement data.
“Everyone working on autonomous vehicles is doing it wrong,” was the title of Hotz’ keynote at Locate, and his comments onstage did indeed convey the kind of boundless confidence you might expect of a former teen superhacker. Hotz is quite certain the future of self-driving technology is open-source. “Don’t go rent-seeking,” he said twice in the keynote—that is, don’t go asking for cash infusions without creating something of value. What is valuable? To Hotz, limitless growth, à la Uber, but through open-source products that anyone can take, manipulate, and improve.
Sounds utopic. But the flipside of Hotz’s unbridled belief in open-tech’s power is that he doesn’t seem to worry much about questions of why, or the potential social disruption that automation breakthroughs may trigger. He has a tendency to brush off the human factors involved in autonomy: the fears, the discomfort with far-off horizons, the tough challenge of preventing drivers in semi-autonomous systems from becoming fatally distracted. In his Locate address, Hotz was downright glib when speaking about recent fatalities associated with Tesla’s Autopilot system. “Those were ‘unplanned disengagements’,” he said in engineer-speak. “Avoid those.”
Then again, Hotz likes to troll. He demonstrated that when we caught up over the phone after the conference about what he sees down the self-driving road. The interview has been edited and condensed for clarity.
“Everyone working on autonomous vehicles is doing it wrong” was the title of your keynote address at Locate. Why are they doing it wrong?
They made that title up. I didn’t come up with that. I don’t care. I’m not big on controlling the message or anything like that. I’m just gonna get up on stage and say whatever I’m thinking about that. If that’s their message, let them brand me.
But “everyone’s doing it wrong” does seem on-brand for you, in the sense that you’re about open-source in an industry that’s largely fiercely proprietary.
No, [Chinese search engine giant] Baidu is saying the same stuff. If you wanna be an Android-like play in this ecosystem, you have to be open-source. We’re not alone.
Why are you so certain that open-source software is technology’s future?
It’s a simple concept. Throughout history, we’ve had good scarcity. Wars have been fought over good scarcity: I want this land, you want it too, so let’s fight. Now we fight over intellectual property, and that’s not scarce because anyone can make a copy. In the land debate we’d both just get a copy of the land. Instead, companies want to keep artificial scarcity around. But it’s the lack of scarcity that’s the better way to drive growth, and that’s how you drive economies, and that’s why open-source.
You say that’s true from a mapping perspective, too. The high-definition road maps that underwrite the next edition of Open Pilot will also be entirely open-source. That’s a big difference from Google, Uber, and other mapping giants also working in the self-driving space.
It’s questionable whether HD maps can be copyrighted at all, since everything in it is a fact. A lane line is a fact. Facts can’t be copyrighted, so it’s unclear if they’re creative works and could even be protected by property law.
But the software and sensors that are building the map are still copyrighted, right?
Oh, absolutely. The thing that creates the map or the thing that uses the map is software and covered by normal property rights. But the map itself is unclear.
What does that mean for the companies pouring millions of dollars into the race to build the best 3-D maps for self-driving?
They shouldn’t be. They’re wasting tons of money. You can break mapping companies into two groups: the group that pays for expensive survey maps, like TomTom and HERE, and the group that’s making maps through crowdsourcing, like us, Tesla, and Mobileye. The second group, instead of paying for maps, they get paid for each car they’re entering into the network. Waze, too—they made their map through crowdsourcing and get paid for each user as well by showing them the map.
You’re talking about the feedback loop: You test autonomous cars out on the road, their sensors gather location data, then feed it back into the map.
Yes. Tesla is going to start to make maps, too, I’m sure. They’ll use their fleet of cars to make maps.
What kinds of biases could still creep into a fully automated map?
I don’t know. There might be bugs. But you’d fix them.
What has demand been like for Open Pilot?
Pretty good. We have over 250 weekly active users on the software—250 users out there, testing it for free. They’re sitting in the driver’s seat. Plus all of our employees have the cars.
Where are the users?
Majority are in the U.S. Some in Canada. Wherever people are.
Is there a standard consumer you’re selling to? Open-source enthusiasts, I imagine?
We don’t really track that. We just make the product and make it available for sale.
Sources have told me that the ability of artificial intelligence to properly read large trucks on the highway, which was a factor in one of the recent Tesla Autopilot crashes, is an industrywide challenge. Have the recent Tesla and Uber fatalities involving semi-autonomous cars given you pause in any way, or flagged technological challenges for you?
No. The issue with that crash was that the driver wasn’t paying attention. Tesla Autopilot is a level-two autonomous system. You have to pay attention like you’re driving. So it was like someone turned on cruise control and crashed into that truck. Autopilot is not a self-driving car. You have to pay attention. So the remedy for that would be driver monitoring. Making sure people who drive cars pay attention is a good idea in general. I don’t think people realize how distracted drivers are.
The problem is about when these things start to get too good. If your system is making a mistake every 10 miles, it’s no problem. But once it’s every 1,000 miles people trust it too much. So the way you combat that, you just make sure they pay attention. Watch the person and make sure they’re looking at the road and not the phone.
How do you watch?
With front-facing cameras. One of the products we’re shipping today has a camera pointed at the driver.
Why go head to head with GM’s Super Cruise? [Note: This is not to be confused with GM’s Cruise, the company’s developmental self-driving project.]
It’s a good benchmark. It’s a more restricted product than Tesla’s Autopilot [its maps are limited only to a certain radius] so it does what it does, but better.
By the way, saying you’re better than Autopilot is meaningless since they constantly update their software. But GM’s Super Cruise is good because it’s a more stable target. We’ll also be releasing our own maps to make that possible.
Where do Comma’s products stack up against those of Waymo, the company considered to be the industry leader by many measures?
We don’t compete with them. They compete with Uber. We’ll never offer point-to-point taxis. Those are moonshot ideas, way too far in the future. We compete with what car manufacturers are making now.
George, what is the promise of autonomous vehicles?
I’m not promising anything. I build technology.
It’s something to do. I think technology is generally a good thing.
Other companies tout the safety promise of autonomous vehicles. Is that disingenuous?
If some people care about their branding and narrative, fine. I think safety is a good thing. Don’t make things that are unsafe—that’s bad. But it’s not the reason I wake up in the morning. I like winning. I like winning and building technology. How do you win? Build a better product. A lot of technology is oriented around that.
How long before we have that fully autonomous future we’ve all been promised?
I don’t know anything about that. Whenever you hear some company give an estimate, it wasn’t made up by the engineers. It’s marketing.
In your keynote, you said that you believe that once we solve the AI issues in autonomous driving, we’ll also “solve” the need for human labor in other industries. Why?
I just said that to troll.
Because it’s funny. It gets people up in arms about bullshit. Have you seen what these things do today? Not that much. They make driving better. If you want a better tagline for what we’re doing, that’s it: “Let’s make driving better.” There are a lot of axes you can make something better on: faster, cheaper, safer, more comfortable. But building technology takes a long time.
What are you working on today?
I’m working on one of our secret new products. So we can sell more products on the store.
What are they?
Similar to the old products. But better.