Sebastian Thrun

You are currently browsing articles tagged Sebastian Thrun.

Sebastian Thrun is a brilliant guy who was positioned at the starting line of the recent boom in driverless technology, but he’s no stranger to irrational exuberance. A couple years ago, the Udacity founder earnestly announced that “if I could double the world’s GDP, it would be very gratifying to me.” Yes, that would be nice.

The computer scientist and entrepreneur is now employed as CEO of Larry Page’s Kitty Hawk, engaged in trying to perfect the flying car, a vehicle of retrofuture dreams that seems exceedingly unnecessary. Wouldn’t it be far better for society if he and others like him were engaged in innovation aimed at more practical public transportation solutions for the masses? The thing about childhood dreams is that most of them are childish.

Steven Levy held roughly the same view last month when he sat down to interview Thrun for Backchannel (now housed at Wired). The opening:

Steven Levy:

Why do we need flying cars?

Sebastian Thrun: 

It is a childhood dream. Flying is just such a magical thing to do. Making personalized flight available to everybody really opens up a set of new experiences. But in the long term there’s a practicality to the idea of a flying vehicle that takes off vertically like a helicopter, is very quiet, and can serve short range transportation. The ground is getting more and more congested. In the US, road usage increases by about three percent every year. But we don’t build any roads. And countries like China that very recently witnessed an explosion of automotive ownership are suffering tremendously from unbelievable traffic jams. While the ground infrastructure of roads is one-dimensional, the sky is three-dimensional, and it is much, much larger.

Steven Levy:

But it you build flying cars, won’t the air be just as congested?

Sebastian Thrun: 

The nice thing about the air is there is more of it. You could have virtual highways in the sky and stack them vertically. So you never have a traffic intersection or similar.

Steven Levy:

But highways have lanes. You can’t have dotted lines in the sky.

Sebastian Thrun: 

Yes, you can, it turns out. Thanks to the US government we have the Global Positioning System that gives us precision location information. We can paint virtual highways into the sky. We are actually doing this today. When you look at the way planes fly, they use equipment that effectively constructs highways in the sky.

Steven Levy:

Still, the number of planes is tiny compared to cars, which you want to put in the air. Plus, everybody is buying drones. If you folks get your way, the sky is going to be completely full.

Sebastian Thrun: 

Every idea put to the extreme sounds odd.•

Tags: ,


MOOCs feel mostly like misses for the moment but so did the phonograph and automobile originally. Outsize ambitions still abound, with Sebastian Thrun of Udacity having recently said, “If I could double the world’s GDP, it would be very gratifying to me.” Yes, that would be nice. In John Thornhill’s smart Financial Times piece about online education, the former Google driverless guru has a more sober quote: “It is not clear that the existing universities are the right places to create education.”

Higher education’s endless layers of administration, insane sticker prices and pauper professors have left an opening for MOOCs, but this nouveau learning industry will likely be only as successful as its products are good. Thornhill opens his piece about EdTEch with a story about French education innovator Xavier Niel:

With no teachers, timetables, or exams, Ecole 42 is a strange kind of educational institution, open 24 hours a day, seven days a week. Students shuffle into this tech-enabled school whenever they want and work as hard as they need.

Is this the future of education?

Xavier Niel, the French internet and telecoms billionaire who founded the coding school for young adults in Paris in 2013, certainly thinks so. He chose the school’s number for a reason. As fans of Hitchhikers’ Guide to the Galaxy know, 42 is the answer to the ultimate question of life, the universe and everything.

So sure is Mr. Niel that he has found the answer that he has committed himself to funding Ecole 42 for the next decade and is spending a further $100m on a new school in San Francisco. There are several ironies in a French entrepreneur teaching Silicon Valley geeks how to code.

Mr. Niel argues that smartly designed online courses are more effective than traditional classroom teaching methods. Students learn best by pursuing online projects by themselves and by interacting with each other. Peer-to-peer lending may be going through a rough patch, but peer-to-peer learning may be on the rise. “We are preparing people to learn together,” he says.•

Tags: , ,

When you’ve been one of the leading minds behind getting cars to drive themselves, you tend to shoot for the moon. Sebastian Thrun certainly is with Udacity, his education start-up which currently offers nanodegrees in things like Data Analyst and Android Developer. Not exactly a Stanford or Harvard or even community college curriculum, but Thrun believes he’s just at the beginning of reimagining higher ed. From a Smithsonian Q&A Roger Catlin conducted with the Google X lab founder:


What is your dream for Udacity?

Sebastian Thrun:

If I could double the world’s GDP, it would be very gratifying to me, measuring it not by the company itself but by the impact it would have. We are launching an education system that Google has undersigned, a joint education for entrepreneurship. It’s a niche to some extent, but if you bring this to the Middle East, if you bring this to Africa, if you bring this to Bangladesh, to developing countries, to China and India, I think it can have a huge impact on their ability to participate constructively in the creation of wealth and prosperity. Specifically the Middle East, at this point, suffers from the fact there is no path for young people to participate constructively, so some of those, as a result, may choose other paths, like terrorism.


What are the greatest obstacles of reaching that goal?

Sebastian Thrun:

Eventually, it will take broadening the course catalog. We work with computer science and software stuff, but not everyone wants to be a software engineer.

Where should I start? Obviously we are iterating the student experience, and in some courses we managed to get the finishing rate from about 2 percent to over 90 percent. And that was really hard work to make it really good. So think about it as a car that in the beginning drives about 10 mph, but with relentless engineering you get it to about 100 mph. That’s the product quality. The quality of the experience. The second one, honestly, is that education is such a slow growing field, so there is a trust element. Like, do you trust a new player? And to some extent education is owned by the degree-granting universities that have an efficient delivery model. So to gain the trust of our students means we’ll be placing them in jobs, showing the job records, to show how the teaching really empowers them. That will bring new students, but that’s going to take some time.•

Tags: ,

Driverless cars have been a goal of some for at least 85 years, but as recently as 2007 they seemed a pipe dream to most. John Tierney of the New York Times, though, noted in that year the astounding progress the sector had made in short shrift, and assumed that autonomous cars wouldn’t remain futuristic frustrations like the flying kind. From Tierney’s prescient 2007 Times article:

As the baby boomers cruise into their golden years, I have good news for them — and for everyone else in danger of being run over by these aging drivers. The boomers will not be driving like Mr. Magoo. An electronic chauffeur will conduct them on expressways, drop them at the mall entrance and then go park their cars.

If you doubt this prediction, I don’t blame you. The self-driving car ranks right up there with the personal hovercraft as the futurist vision that never comes true. In 1969, Disney unveiled Herbie the Love Bug; in 1940, Popular Mechanics promised a car that would chauffeur you across America in a single day to visit Aunt Lillian.

At the 1939 World’s Fair, the crowds at the General Motors Futurama exhibit saw traffic speeding 100 miles per hour thanks to electronic help. ‘Safe distance between cars is maintained by automatic radio control,’ a voice explained as visitors looked down on the vast diorama of the World of Tomorrow, complete with hangars for dirigibles and landing decks for autogyros.

‘Does it seem strange? Unbelievable?’ the announcer intoned. ‘Remember, this is the world of 1960!’

O.K., so they were a little off on the date. But today, finally, those electronically spaced cars are on the highway. You can buy cars with ‘adaptive cruise-control’ that automatically slow down if the radar or laser detects you tailgating. Your car can warn you when you stray across lane markings, and these kinds of sensors are already being used experimentally in cars that drive themselves.

These smart cars still have their bugs, but engineers have made amazing progress the past several years. In 2004, when the Defense Advanced Research Projects Agency held its first Grand Challenge for driverless cars, none made it more than seven miles. At Darpa’s next Grand Challenge, in 2005, five cars made it 132 miles to the finish. And then, last month, six cars completed a 60-mile course that was the grandest challenge yet because they had to deal with traffic along the way.

These empty cars drove themselves around an Air Force base in Southern California, finding parking spots, obeying stop signs, idling in traffic, yielding to other cars at intersections and merging into traffic at 30 m.p.h. There was one accident and a few near misses, but the cars’ engineers are so buoyed by the results that they’re hoping the next competition will be a high-speed race on a Grand Prix course.

‘Within five years, it’s totally feasible to build an autonomous car that will work reliably in several limited domains,’ says Sebastian Thrun, a computer scientist at Stanford and head of its racing team, which won the 2005 Darpa competition and finished second in last month’s. In five years he expects a car that could take over simple chores like breezing along an expressway, inching along in stop-and-go traffic, or parking in the lot at a mall or airport after dropping off the driver. In 20 years, Dr. Thrun figures half of new cars sold will offer drivers the option of turning over these chores to a computer, but he acknowledges that’s just an educated guess. While he doesn’t doubt cars will be able to drive themselves, he’s not sure how many humans will let them.•

Tags: ,

It’s not likely that legal issues regarding autonomous cars will be as much a hurdle as some think, but they will be somewhat of a story. In the New York Times article, “When Driverless Cars Break the Law,” Claire Cain Miller breaks down the potential future of civil and criminal culpability:

“In cases of parking or traffic tickets, the owner of the car would most likely be held responsible for paying the ticket, even if the car and not the owner broke the law.

In the case of a crash that injures or kills someone, many parties would be likely to sue one another, but ultimately the car’s manufacturer, like Google or BMW, would probably be held responsible, at least for civil penalties.

Product liability law, which holds manufacturers responsible for faulty products, tends to adapt well to new technologies, John Villasenor, a fellow at the Brookings Institution and a professor at U.C.L.A., wrote in a paper last month proposing guiding principles for driverless car legislation.

A manufacturer’s responsibility for problems discovered after a product is sold — like a faulty software update for a self-driving car — is less clear, Mr. Villasenor wrote. But there is legal precedent, particularly with cars, as anyone following the recent spate of recalls knows.

The cars could make reconstructing accidents and assigning blame in lawsuits more clear-cut because the car records video and other data about the drive, said Sebastian Thrun, an inventor of driverless cars.

‘I often joke that the big losers are going to be the trial lawyers,’ he said.

Insurance companies would also benefit from this data, and might even reward customers for using driverless cars, Mr. Villasenor wrote. Ryan Calo, who studies robotics law at the University of Washington School of Law, predicted a renaissance in no-fault car insurance, under which an insurer covers damages to its customer regardless of who is at fault.

Criminal penalties are a different story, for the simple reason that robots cannot be charged with a crime.”

Tags: , , ,

From a Foreign Affairs interview that Gideon Rose conducted with roboticist Sebastian Thrun, a passage about the subject’s triumph in a 2005 driverless-car competition in the Mojave Desert:


Why did your project end up working so well?

Sebastian Thrun:

Many of the people who participated in the race had a strong hardware focus, so a lot of teams ended up building their own robots. Our calculus was that this was not about the strength of the robot or the design of the chassis. Humans could drive those trails perfectly; it was not complicated off-road terrain. It was really just desert trails. So we decided it was purely a matter of artificial intelligence. All we had to do was put a computer inside the car, give it the appropriate eyes and ears, and make it smart.

In trying to make it smart, we found that driving is really governed not by two or three rules but by tens of thousands of rules. There are so many different contingencies. We had a day when birds were sitting on the road and flew up as our vehicle approached. And we learned that to a robot eye, a bird looks exactly the same as a rock. So we had to make the machine smart enough to distinguish birds from rocks.

In the end, we started relying on what we call machine learning, or big data. That is, instead of trying to program all these rules by hand, we taught our robot the same way we would teach a human driver. We would go into the desert, and I would drive, and the robot would watch me and try to emulate the behaviors involved. Or we would let the robot drive, and it would make a mistake, and we would go back to the data and explain to the robot why this was a mistake and give the robot a chance to adjust.


So you developed a robot that could learn?

Sebastian Thrun:

Yes. Our robot was learning. It was learning before the race, and it was learning in the race.”

Tags: ,

No one fully knows how our technological revolution will change the landscape of higher education, but it’s clear that things will be different in the coming decades. Perhaps markedly different. Sebastian Thrun departed from a tenured position at Stanford to become a Google fellow and to begin the online university Udacity. If all attempts to alter higher learning are this intelligent and enlightened, we will be very blessed. The opening of Steven Leckert’s new Wired article about the technologist’s experiment, in which the author enrolls in a Thrun class:

Stanford doesn’t want me. I can say that because it’s a documented fact: I was once denied admission in writing. I took my last math class back in high school. Which probably explains why this quiz on how to get a computer to calculate an ideal itinerary is making my brain hurt. I’m staring at a crude map of Romania on my MacBook. Twenty cities are connected in a network of straight black lines. My goal is to determine the best route from Arad to Bucharest. A handful of search algorithms with names like breadth-first, depth-first, uniform-cost, and A* can be used. Each employs a different strategy for scanning the map and considering various paths. I’ve never heard of these algorithms or considered how a computer determines a route. But I’ll learn, because despite the utter lack of qualifications I just mentioned, I’m enrolled in CS221: Introduction to Artificial Intelligence, a graduate- level course taught by Stanford professors Sebastian Thrun and Peter Norvig.

Last fall, the university in the heart of Silicon Valley did something it had never done before: It opened up three classes, including CS221, to anyone with a web connection. Lectures and assignments—the same ones administered in the regular on-campus class—would be posted and auto-graded online each week. Midterms and finals would have strict deadlines. Stanford wouldn’t issue course credit to the non-matriculated students. But at the end of the term, students who completed a course would be awarded an official Statement of Accomplishment.

People around the world have gone crazy for this opportunity. Fully two-thirds of my 160,000 classmates live outside the US. There are students in 190 countries—from India and South Korea to New Zealand and the Republic of Azerbaijan. More than 100 volunteers have signed up to translate the lectures into 44 languages, including Bengali. In Iran, where YouTube is blocked, one student cloned the CS221 class website and—with the professors’ permission—began reposting the video files for 1,000 students.”


 Thrun talking self-driving cars:

Tags: ,

As education shifts further online, Stanford is offering an online course beginning in September, Introduction to Artificial Intelligence. The course will be taught by Sebastian Thrun and Peter Norvig, and they’ll be weekly lectures and homework. The course is estimated to take 10 hours per week of work and certificates will be awarded.


CS221 is the introductory course into the field of Artificial Intelligence at Stanford University. It covers basic elements of AI, such as knowledge representation, inference, machine learning, planning and game playing, information retrieval, and computer vision and robotics. CS221 is a broad course aimed to teach students the very basics of modern AI. It is prerequisite to many other, more specialized AI classes at Stanford University.” (Thanks MetaFilter and Marginal Revolution.)

Tags: ,

The Stanford and Google genius gives a TED talk. (Thanks IEEE Spectrum.)