By Carolyn Said
San Francisco Chronicle
WWR Article Summary (tl;dr) Sebastian Thrun, founder of Udacity discusses Udacity’s online courses and the future of self driving cars.
San Francisco Chronicle
Sebastian Thrun wears many hats. He’s an expert in robotics and artificial intelligence, a scientist, educator, inventor and entrepreneur. He’s best known for founding Google X, a lab for world-changing, “moonshot” projects, and helping to pioneer self-driving cars at Google and Stanford, where he is a research professor.
Now he’s focusing on workforce training through his 5-year-old Mountain View startup Udacity, which offers online courses to quickly prepare students for tech jobs.
Udacity’s 12 “nanodegree” programs, which include certificates in self-driving engineering and artificial intelligence, cost a few hundred dollars, take a few months and are tailored to needs outlined by employers. Backed by $163 million in funding, Udacity is valued by the private markets at $1 billion and has served almost 5 million students worldwide. Thrun sees Udacity’s mission as democratizing education. This interview has been edited for length and clarity.
Q: Why do people call you the father of self-driving cars?
A: There are many fathers and mothers: teams, not individuals. But the DARPA Grand Challenge was the pivotal moment. (The Defense Advanced Research Projects Agency offered a cash prize for creating a fully autonomous ground vehicle that could complete a certain course. In 2005, Thrun led a Stanford team in developing Stanley, a robotic car that won the prize.) At the time, it was a niche event. Automotive companies didn’t take it seriously. Now there is tremendous velocity; you could say 2016 is the year of the self-driving car in terms of tech coverage.
Q: What technical barriers remain?
A: When a human driver makes a mistake, he or she learns but nobody else learns. When a self-driving car makes a mistake, all the other cars learn from it as well as the unborn cars, future cars. The rate of progress is much faster.
The point has been reached when self-driving cars are safer than human driving. Look at Google: There are 300,000 miles between critical interventions when a human driver has to take over. That’s longer than the typical mean times between accidents for human drivers. The technology is there. What’s missing are business models.
Q: What business models do you think will work?
A: There are a couple of approaches. With owner-operated vehicles, you buy a self-driving car for your own benefit, and it relieves you of the need to pay attention.
There’s the Uber/Lyft/Didi route, where transportation becomes a service. That’s more disruptive, because so many things will change. Then there’s trucking, like what Otto (a San Francisco company recently purchased by Uber) is doing. There are multiple ways of doing business there, such as convoying or platooning.
Q: What are the implications of widespread robot taxis?
A: If transportation becomes a service, we will need many fewer cars and won’t need individual car insurance. Parking garages will be obsolete; trauma surgeons and tort lawyers will have less work. City design will be different; people can live closer together and waste less land. It will be uniformly better for the planet.
Q: When do you expect self-driving cars to be available?
A: I would be shocked if the first successful transportation-as-a-service business takes more than three years to launch — in limited geographic areas, not widely. Google’s technology is already at the place where it has to be.
Q: Are there psychological impediments to people accepting self-driving cars?
A: At any given time, you have more than 100 self-driving cars in the Mountain View area and people here are used to them. When I ran the Google team, we’d often put visitors behind the wheel. I felt people were too confident too quickly with their willingness to take their eyes off the road and trust the car. (For example,) when I’d take the Google car to Tahoe for the weekend, my wife would ask me to let the car drive because she felt safer.
Measuring the Google car’s performance against professional drivers, it was appreciably better at keeping in the center of lanes, doing less braking and smoother acceleration. It’s more defensive in driving than most people are. It led to an increase in trip time by maybe a percent or so.
Q: Do you think we should design autonomous cars so humans can still drive them?
A: I would skip the steering wheels and brake pedals in a heartbeat. But they may be necessary for a while to retrieve vehicles from situations where they might be stuck.
Q: What will they cost?
A: Per-family transportation costs are now $26 per day in the U.S. That can be cut in half if self-driving cars pick up and deliver you. The cost per car will go up, but the cost per consumer will go down because utilization will increase. Now cars are parked 90 percent of the time. A parking spot in a modern parking garage in Silicon Valley costs more than a car — about $100,000 to build.
Q: What should we do with all the garages and parking lots?
A: Reclaim the land. Imagine tearing down the garages and turning them into living space or office buildings. In Los Angeles, 60 percent of the landmass is dedicated to cars: roads, driveways, garages. We can’t replace all of this because we’ll still need roads.
Q: What about the impact on jobs?
A: That’s where Udacity comes in. There are a million or more people whose jobs might change or be in jeopardy if self-driving cars become a phenomenon of scale, found everywhere — which might take 10 or 20 years. Taxi drivers, insurance brokers, trauma surgeons, tort lawyers.
But the beauty of innovation is that new jobs open up as we develop technology to get rid of repetitive and mindless human labor.
I believe only 1 percent of interesting things have been invented yet. We love recent inventions: planes, cars, cell phones, Facebook, Google. Yet we often don’t realize it’s just the beginning.
Q: How did you come to found Udacity?
A: It happened by accident while I was building Google X, as the “moonshot” captain. I was still teaching at Stanford. A colleague and I decided to put a graduate-level artificial intelligence course online. Within weeks, 160,000 people signed up. Of them, 26,000 finished. We were teaching the same class on campus with about 200 students. The single best Stanford student ranked No. 413 in the (combined) class.
At that moment, I saw a huge inequity in education. Of the top students online, some were soldiers, some were mothers raising children, one was on dialysis. It was like a calling. I realized I could be at Google and do self-driving cars — or I could teach the world to build self-driving cars. It became clear to me: If you can truly democratize top-notch education, you can reach so many more people. It became a mission.
Q: How are you carrying out that mission?
A: We offer all our content free for people who can’t afford tuition. We’ve had almost 5 million students all over the world take our free courses, and 13,000 (people are currently enrolled in the paid) nanodegrees. We charge around $200 a month and when students graduate, typically give half the tuition back. We give a full refund if people can’t find a job they love. (About 3,000 students have completed a nanodegree. About 900 students have found related jobs, but that includes people currently enrolled.)
Two weeks ago I was in Saudi Arabia, a year earlier in Egypt. We’re organizing (courses) for Syrian refugees in Germany. Our message is: You can learn to be data scientists or machine engineers wherever you are.
Q: How do you help your graduates find jobs?