Skip to Content

Sebastian Thrun on the Future of Learning

Udacity cofounder and CEO Sebastian Thrun says more AI is coming to online education, but we’ll still need humans to grade our English essays.
July 19, 2013

Sebastian Thrun has worn many hats in the tech world: Stanford research professor, founder of Google’s X Labs, where he oversaw the development of self-driving cars and Google Glass, and, most recently, passionate advocate for MOOCs—or massive open online courses—through Udacity, the online education startup he cofounded and runs (see “The Most Important Education Technology in 200 Years”).

Sebastien Thrun
MOOC master: Sebastian Thrun, cofounder and CEO of online education startup Udacity.

Started in 2012, Udacity’s goal is to make education affordable for everyone. All classes are available for free on the Web, and some also offer college credits for a fee of $150. It’s one of a growing number of startups offering similar services. And the timing seems right: the average cost of a bachelor’s degree has shot above $100,000, and high-speed Internet has made so-called “distance learning” a plausible alternative to the classroom. MOOCs have been hailed by many as the future of education, but they’re still in their early days, and they’re susceptible to growing pains (for example, one of Udacity’s partners, San Jose State University, said this week that it is suspending its online course offerings until the spring, according to an Inside Higher Ed report).

Thrun sat down to chat with MIT Technology Review IT Editor Rachel Metz at Udacity’s Mountain View, California, office.

San Jose State University is suspending courses it has been offering through Udacity that involved both high school and San Jose State students, due to low course-passing rates as compared to traditional classes, and plans to start things up again in the spring. How do you feel about this?

We felt we got these kids, they worked really hard, and they stayed with it, but they didn’t get the skills they needed to be proficient. We asked them why, and they said they needed more time. Literally, this is a truly joint decision; I’m totally behind it because I feel the objective must be to give students a great education.

How has online learning changed since you started Udacity?

We’ve evolved the MOOC concept into one that really helps people throughout the course to complete the course. The most recent completion rates in pilots we’ve been running have been 85 percent, as opposed to 5 percent or 4 percent, which is common in MOOC-land.

That’s a huge difference in the number of people starting and finishing an online class. How do you make that happen?

We mostly did this by two ingredients. One is to really add notable value to the certificates [students receive], and the value proposition we’ve chosen is core college credits slash degrees, which you won’t get unless you complete. We have a partner institution—San Jose State University. We are also now working with Georgia Tech to offer an entire degree [which will begin in January]. The degree is offered by them, because we are not accredited, but we can certainly supply all the infrastructure, and the secret sauce for making it work.

The second component would be we also provide fairly extensive student services now. We have people on the ground that help you along the way. It turns out, if you’re not just left alone to a computer system, if there are people talking to you online, that makes for much better completion rates. There’s a help line, there are mentors that shepherd you time-wise and remind you to do homework assignments and so on.

Do you feel like part of the overall low-completion rate now for MOOCS has something to do with user expectations? Do users expect too much and then become disappointed?

Even at the top-notch institutions you find that students often take the courses that are the least work. Some people are in there for learning, but some people are in there to say we want to have a certificate. Sometimes it’s because the courses are just too long. Many MOOCs are like semester-long classes, and they’re just a long time to sit in each. In this day and age, I mean, if you look at video games, that are much, much shorter, and tweets are shorter than e-mails, right, so everything is becoming shorter, faster. To some extent we have to adjust to that.

It’s not just students using Udacity—companies are using it, too, to train workers and reach potential employees, right?

We have a whole bunch of partner companies that … have supplied us with courses, have financed courses, built courses, or worked with us to build courses. The idea there has been that the modern workforce really needs to have a certain kind of knowledge, and it’s much faster to supply it through a platform like Udacity than it is through a conventional university.

First, companies get out of it that they help the world acquire some skills. Often the skills [relate] to their own products. Google, for example, chose HTML5, which is a Web language, as the course focus, and Google itself has a vested interest in more people being proficient in HTML5. But they also get employees and interns out of this. Any of these companies has a deal whereby they get access to the top students in those classes. We found in the past that the top students in those classes compete favorably with the top students at leading U.S. institutions. And very often those top students in our classes had no access to leading U.S. education institutions for various reasons—geographical, demographical. Or sometimes it’s just cultural.

What these courses find are these extraordinarily talented students, they teach them something very important, some skill, and then they become highly employable. We have directly placed several dozen students in jobs, but we know of hundreds of people who found jobs with our certificates. And this is just the tip of the iceberg.

What’s a technology you’re working on or thinking about that would make online courses more engaging or interactive?

We have to really work on artificial intelligence, really understand, like, where can we take a student if they have a certain kind of learning profile? We do some of it manually right now, so we analyze student profiles, we make predictions of what are the success rates, and then we intervene manually right now based on the predictions we get from students’ profiles. But we haven’t automated this yet. So eventually it’s going to be a big piece of artificial intelligence that sits there, watches you learn, and helps you pick the right learning venue or task, so you’re more effective and have more pleasure.

How long until we get there?

I think a year, at least.

What might that look like on Udacity?

You might come and it might ask you—okay, here’s a problem, calculate some interesting geometric proposition or solve a set of equations, and when you start typing in your answer it might realize, wow, you don’t even know what fractions are. Without being demeaning to you it says, okay, let’s take a little detour; let’s see if you can do fractions. And then you all of a sudden get a problem that’s much easier. Maybe you get it right, in which case the system might say congratulations, let’s move back to the original question. Or maybe you got it wrong, at which point it’s pointless to give you the hard questions—it’s much, much smarter to give you the easy questions. It’s kind of what a good tutor does, right?

Speaking of AI, how do you feel about automated grading and feedback? Can it become as skilled, or better, than a human teacher?

A grader for a computer program is called a compiler. It’s either right or wrong and there are computer programs that can help you. And even there it’s not completely trivial. If you want to prove a theorem, it’s not entirely obvious how to assess a theorem, but by and large it’s easy. Compare this to critical dialogue in philosophy, discourse in philosophy. There, it’s really the subtlety of their language that makes all the difference and more—it’s not just about assessment, it’s not about grading, it’s also about feedback. When someone writes an essay, you want to give meaningful feedback so they can improve. I’ve seen good progress on the assessment of essays; I’ve seen almost no progress on qualified feedback. And that’s where you have a very simple opinion—you just have people do it. Our classes right now require essay writing, and those essays are being graded by people and it’s just fine, in my opinion. Why not? There are a lot of unemployed people in this country. I don’t think it has to be all computerized.

Where do you hope Udacity is five years from now?

I think we’ll be just like a university, but we’ll be a university for the 21st century.

Keep Reading

Most Popular

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

OpenAI teases an amazing new generative video model called Sora

The firm is sharing Sora with a small group of safety testers but the rest of us will have to wait to learn more.

Google’s Gemini is now in everything. Here’s how you can try it out.

Gmail, Docs, and more will now come with Gemini baked in. But Europeans will have to wait before they can download the app.

This baby with a head camera helped teach an AI how kids learn language

A neural network trained on the experiences of a single young child managed to learn one of the core components of language: how to match words to the objects they represent.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.