GEORGE HOTZ is hungry. He missed breakfast so we are prowling through a Lisbon mall on the hunt for espressos and Portuguese custard tarts. His hoody displays the logo for his new start-up, Comma.ai, a company arming a growing number of hobbyists in the US with the ability to turn standard cars into self-driving vehicles. We talk as we walk through the tide of shoppers, looking for a cafe.
“I define making an impact the way a meteor does,” Hotz says, with trademark modesty. He has yet to destroy any dinosaurs, but his approach to developing AI for autonomous vehicles is certainly making waves. His ultimate goal? To corner the software market for self-driving cars.
Advertisement
Hotz shot to fame in 2007 as “geohot”, a trash-talking 17-year-old who was the first to unlock the original iPhone so that it could be used by a variety of phone companies. Later, Sony threatened him with legal action when he released a hack of the Playstation 3. A dropout from Carnegie Mellon University – “There are only five worthwhile courses you need to take. I figure you take them, then get out of there” – his latest obsession is hacking cars to turn them into semi-autonomous vehicles.
Tech-savvy drivers in the US with, say, a recent-model Toyota, can buy hardware from Comma.ai’s website – a dashboard camera, an interface to the car’s electronic systems – and combine it with the firm’s open-source software, openpilot, to create a car that can automatically control acceleration, braking and power-steering systems in specific circumstances, such as driving on a highway.
In driving terms, the set-up essentially provides a sexed-up cruise control that keeps you in lane and a safe distance from the car in front. It is a modest degree of autonomy, so the driver must remain vigilant; every 6 minutes the system prompts them to move the steering wheel or touch other inputs to prove they are paying attention. What if the driver doesn’t respond? “Steering and braking are controlled [by openpilot] as before. It just stops accelerating, and slows until you take control back,” says Hotz.
In late 2016, Hotz cancelled his first project, a $999 self-driving kit called Comma One, shortly before it was due to launch when he got a in the US expressing concerns. The legal status of self-driving cars is a work in progress, notes Hussein Dia of Swinburne University of Technology’s Smart Cities Research Institute in Melbourne, Australia. “The regulators are still struggling with how to license or validate [self-driving] software.”
Presumably to avoid a regulatory swamp, Hotz is now at pains to point out that he isn’t selling a self-driving “system”, per se, but merely providing a variety of hardware and software components and what people do with them is their call. His fans enjoy this balancing act. Later, as he roams the stage in front of more than 1000 people at the Web Summit conference, he raises an audience chuckle when he talks about the software you can run on some of Comma.ai’s in-car hardware. “Maybe you can run open-source software that drives your car,” he says, with a knowing look and heavy emphasis. “It’s up to you.”
So far, . Hotz’s goal for 2018 is to have developed software for six of the 10 bestselling cars in the US.
Back in the cafe, I ask him what his motivations are. “I solve puzzles,” he says. “And self-driving is an incredible puzzle – probably the best use of applied AI there is.”
It’s a challenge that the entire automotive world is tackling, from to Tesla, and Apple to Uber, with varying degrees of success. Hotz’s set-up is , but his software is always improving, thanks in large part to a constant stream of data from drivers who have elements of Comma.ai’s system installed. In 2016, Comma.ai collected input from more than 3 million kilometres of self-driving on roads, Hotz says, mostly in the US. Alongside data from the tens of thousands of people who have downloaded the firm’s chffr (“chauffeur”) smartphone app – which turns users’ phones into a dashcam that records and uploads all the road action – this information allows openpilot to constantly train and be updated.
“Luddites don’t win. It’s just a case of how long you want to hold things back”
That’s why Hotz thinks his software will ultimately have an edge over the giant car manufacturers. “The automakers just don’t understand software, they don’t do updates,” he says. He doesn’t believe it is possible to write code from the top down that will create self-driving cars for the masses. “Driving is like a dance,” he says. It evolves. Within two years, he says, Comma.ai’s machine learning will be advanced enough so that users will only need to get their car onto the highway, then “hit a button and read a book while it gets you to your exit”.
From Hotz’s perspective, firms like Alphabet’s – aka Google’s self-driving car project – and Tesla are going about things the wrong way. Waymo uses a variety of costly lidar and other sensors on its fully automated custom-made cars. ձ’s Advanced system has eight cameras, 12 ultrasonic sensors and forward-facing radar to create a highly autonomous self-driving vehicle. “There’s no [mass] market for $100,000 self-driving cars,” he says.
But as Comma.ai develops openpilot to higher levels of automation, requiring more complex driving, “it will quickly run into the limitations of cameras”, warns Srikanth Saripalli, who researches autonomous vehicles at Texas A&M University. “I’m hoping they will combine their current system with other kinds of sensors.”
Tesla CEO Elon Musk need not be concerned about Hotz, if he ever was. Hotz likens Tesla to Apple’s exclusive operating system, iOS, and styles Comma.ai more on Google’s Android, which can be run on lots of different hardware. In other words, Tesla can do its thing, Hotz wants the rest of the market. Ultimately, he expects the big car firms to license Comma.ai’s software for their cars, just as phone manufacturers license Android. And while the lane-keeping, adaptive cruise control and other basic self-driving services will be free, Hotz envisages drivers paying a subscription to get the latest innovations. “You get to be a member of the Comma.ai premium club,” he says, magnanimously.
A kind of techno-idealism pervades Hotz’s world view and he gets heated when discussing anyone who dwells on the more negative aspects of technology. “Luddites don’t win,” he says to me, leaning forward. “You’re going to lose, it’s just a case of how long you want to hold things back.”
Hotz bucks the trend of many commentators on artificial intelligence because he is convinced that we will have some form of general AI in the next decade or two – and that the impact will be huge. “The real question will be what place in the world do humans have?” he says, casually, tucking into his prized custard tart. This vision doesn’t worry him. “AIs are like your children, you want them to surpass you.” And his role in that? “I’ve wanted to solve AI since I was 15. It’s only recently I thought I had a shot.”
Hungry indeed.
This article appeared in print under the headline “I’ll make your car drive itself”
