After weathering a round of negative publicity, Udacity CEO Sebastian Thrun believes vindication is at hand.
"The thing I'm insanely proud of right now is I think we've found the magic formula," he said in an interview last week. "Had you asked me three months ago, I wouldn't have said that. I'm not at the point where everything is great. There are a lot of things to be improved, a lot of mistakes we're making, but I see it coming together."
Formerly a Stanford University professor as well as the founder of the Google X Labs, which created the famed self-driving car and the Google Glass wearable computer, Thrun co-founded Udacity in 2011 to explore the possibilities of massive open online courses (MOOCs). The success Thrun claims to be on the verge of is actually outside the realm of MOOCs, if you define MOOC as a free online course with a huge enrollment. Instead, he is claiming an early victory in Udacity's partnership with San Jose State University (SJSU) to offer $150 courses for which students would get credit for a passing grade, just as if they had attended on campus. The credit-bearing classes are much smaller, and in the latest round of classes the enrolled students got more tutoring and help.
Announced in January with great fanfare and the backing of California governor Jerry Brown, the SJSU partnership was also the source of Udacity's biggest public relations headache, after the university decided in July to delay offering more courses with Udacity because of poor student pass rates in the spring semester. The results from the summer term, which was already in progress at the time of that announcement, are coming in, and Thrun says they show his company is on the right track.
Meanwhile, Udacity is gearing up to offer an affordable online master's degree in computer science in partnership with the Georgia Institute of Technology, partially underwritten by AT&T as a means of recruiting employees with IT skills. Scheduled to launch in January, the program will cost students just $6,600.
What Udacity is creating, Thrun said, is "an online version of education that really works, that has great retention, great outcomes of education and really reaches people -- not just the world's most motivated 1% -- but can be made to work for many more people."
That would be a turnaround from the current prevailing narrative about Udacity. When SJSU announced its change in plans, the California Faculty Association also leaked copies of an internal report to Inside Higher Ed and the Chronicle of Higher Education showing that enrolled students passed the three courses at a rate of between 29% and 51%, with the worst results in a remedial math class. By comparison, SJSU would have expected at least a 74% pass rate for an on-campus class.
[ Why the invisible hand? Read Udacity: Creating A More Engaging MOOC. ]
Thrun laments that the data was published "in an incomplete form, with a very strong bias," failing to account for the different student population. "In the spring, we specifically went to inner-city high schools and signed up a lot of kids you normally don't see in college." In particular, he said, "we had a pool of students who had previously failed remedial math, and among those we got about a 29% pass rate." Even amidst those results, there were also encouraging signs, he said. The rap on MOOCs is that despite often signing up thousands or tens of thousands of people for a class, only a small number will complete the class -- typically less than 5%. For the San Jose experiment, the completion rate was more than 80%.
Thrun said he could not announce numbers for the summer term yet, but he expects the pass rate to be well over 50%. The completion rate was actually lower, maybe 60% or so (he threw out a few numbers, emphasizing that none were final). More importantly, he thinks Udacity is closing in on a formula that will deliver educational success to students who traditionally have not had access to higher education.
Thrun said SJSU's decision to take the fall semester to analyze the results makes perfect sense, but the summer course statistics ought to help get things back on track for 2014.
The Magic Formula
Udacity MOOCs have a distinctive style. The lectures in Coursera and edX courses are more likely to be video from a classroom or a university TV studio. Udacity courses feel more like a Silicon Valley product, like something that might have been created by Google. When I shared that impression with Thrun, he chuckled, "Good, good, I'm glad you think that."
VP of product design Irene Au previously ran Google's user experience and design teams, but Udacity has also drawn staff from other quarters. Co-founder and CTO Mike Sokolsky was previously a robotics research engineer at Stanford, where he and Thrun collaborated on the design of self-driving vehicles. Google director of research Peter Norvig is also an adviser. In 2011, Thrun and Norvig adapted a class in artificial intelligence (AI) they had been teaching at Stanford, offered it online for free, and attracted 160,000 students from around the world. This was the first MOOC on anything like that scale, and it inspired the creation of Udacity. Thrun said he still does some work with Google, but mostly as an adviser.
While Udacity is often mentioned in the same breath as Coursera, a MOOC startup, and edX, a nonprofit created by MIT and Harvard, Udacity's approach is significantly different. Rather than partnering with universities to distribute courses they produce, Udacity produces all its own courses and in most cases hires the instructors itself.
"We also have the smallest number of courses," Thrun pointed out, "more than an order of magnitude smaller than Coursera." Coursera has partnered with 83 higher education institutions and offers more than 400 courses.
SJSU and Georgia Tech are Udacity's only two university partners so far. Udacity worked with SJSU faculty to create the courses for its program, which can also be taken on a non-credit basis by a Udacity user.
"The reason I'm staying kind of small is I want to solve the question of how to make education work and only scale up once I figure that out," Thrun said. "When we did the AI class, which was sort of the first large-scale MOOC, I think we did a fine job, but I don't think we did a great job. I felt if I now go out and ask a thousand professors to do an online class, it will look and feel very much like the AI class. It's very hard to innovate if at the same time you're trying to scale as platform."
Udacity typically has eight people involved in creating a class, supported by a larger team "developing procedures and strategies and a lot of data-driven improvements," Thrun said. "If we were hosting a few hundred classes, we couldn't do it. I really want to find the magic formula first before scale. I'm really proud that I think we've found the formula. We just need time to really make it work and get the numbers up to where we want them to be. We're not there yet -- I want to be very humble about that."
Thrun seems to be trying to inject a little humility into the way he talks about revolutionizing online education. "A medium where only self-motivated, Web-savvy people sign up, and the success rate is 10%, doesn't strike me quite yet as a solution to the problems of higher education," he told the Chronicle of Higher Education, when interviewed for a report on the underwhelming performance of MOOCs so far.
Thrun's magic formula is not a fully automated online class featuring prerecorded videos and Web-based assessments. In other words, it's not a MOOC at all. To get better results, he said, "We changed the equation and put people on the ground." By adding mentors and a help line, and making phone calls to remind students to do their work, Udacity found it could get more students to do the work, finish the course and pass. Longer term, he has some ideas about using adaptive learning software to eliminate some of this labor, but for now it takes manpower.
"When we look at the data, which we are still analyzing, we do find a whole bunch of people for whom online education doesn't work," Thrun said. "But we're now massively driving students through education with good outcomes -- where these are not the classic, highly self-motivated people."
Beyond The 1%
There is another side of Udacity's business where the MOOC model is working just fine, where it doesn't matter that only a small number of students complete a course. What matters is that those who are successful learn marketable skills, and Udacity has a chance to capitalize on that fact. In this case, its customers are employers seeking access to skills -- like Google underwriting courses on HTML5 and mobile Web development because it can't hire enough good people with those skills.
"I like them because they bring really contemporary technology skills that are very much career-related to pretty much everybody in the world, which is something that's never happened before," Thrun said. "It was really hard for someone in Africa or India to learn how to build a contemporary HTML5 website, and now they can do it."
There is no college credit associated with those classes, and they are not associated with any accredited institution of higher education, but no one cares, Thrun said. "The companies we work with, they don't ask, 'are these courses accredited?' They ask, 'are the people we graduate capable of doing certain things?'"
If students give permission, Udacity will introduce the top students to these potential employers. The fact that these students were determined enough to finish the course on their own initiative, when so many others dropped out, just adds to their cachet.
Yet Thrun said he wants to make a bigger and broader impact than that. When he started the company, he said, "the number one thing I wanted to understand, as a technologist, was can we build something that gives access to a good education to as many people as possible?"
By that measure, he has a long way to go.
One "decisive moment" came earlier this year when he ran into Bill Gates at the World Economic Forum in Davos, Switzerland. Founder and former CEO of Microsoft, Gates has made improving the quality of education one of the major missions of the Bill & Melinda Gates Foundation.
"I love Bill. He's a smart guy, and also a skeptic, which is good," Thrun said. "And he told me, look, Sebastian, what you're doing is teaching to the 1% most motivated people on the planet. To sign up you have to be self-motivated. To stick with it you also have to be self-motivated. Those people, they can learn from anything. If you gave them a book, they would learn it equally well. So what exactly are you changing?' I really took that comment to heart. I started thinking, 'wow, if my target group is the ones who make it through a MOOC, I might find some truly extraordinary people, but I will not have done anything to change education.'"
It was Gates who suggested the focus on remedial math education, and the Gates Foundation provided the funding for Udacity to offer its courses for free to inner-city high school kids, Thrun said. The success rate may not have been impressive, "but we tried it," he said, and learned some things in the process.
Gov. Brown also "made a really convincing point that we should reach out to disadvantaged people, not just to the people who already had it all," Thrun said. "If you ask Sebastian the capitalist, was it wise to go into inner-city high schools with this? Probably not."
The summer-term program was different, based on open enrollment of any students who wanted to sign up, and Udacity added staff to drive success rates up. "What we've learned is the computer program alone, a MOOC alone is not likely to be a good educational medium for large numbers of people, except for the truly highly self-motivated. To be successful, we need people on the ground to do things, to provide educational services."
Similarly, when the Georgia Tech program launches, students will not be in a traditional classroom but neither will they be alone with just the computer to teach them. Georgia Tech will provide the faculty and the content and take 60% of the tuition money, while Udacity will provide "course assistants" in addition to the computer platform for the other 40%. The money will fund services for the students, including a help line, mentoring, student project grading, and exam proctoring, Thrun said.
Next: Adaptive Learning
For a company created by a former Googler and expert on artificial intelligence, Udacity has done little to try to apply big-data-driven adaptive learning techniques to its courses, which are organized around short videos and multiple choice quizzes. The idea that educational software can track student performance and adapt instruction to an individual's strengths and weaknesses is not new, but large-scale online education opens up new possibilities to analyze vast quantities of clickstream and assessment data when making those judgments.
"One of the holy grails we have not yet reached is, can each person find an individual path through the course -- we really want to get there," Thrun said. "We have a lot of data that shows forcing everybody down the same path is not a good idea." So far, he sees the leaders in applying that technology are companies like Knewton, which provides an analytic cloud service that is being incorporated into adaptive software from some of the major educational textbook providers, and Khan Academy, "where your path through a maze of math lessons becomes personalized."
The most adaptive resource Udacity has brought to bear are the tutors working with SJSU students, who will make up extra exercises on the spot to help students understand something they are struggling with. "With an optimal adaptive system, we might not have to do as much of it," Thrun said.
The educational technology industry focus on big data concerns some educators, who worry the technologists are trying to boil all of education down to a simplistic level, where recommendations to students can be issued as easily as book recommendations on Amazon.com or movie recommendations on Netflix.
Thrun's first reaction to this question was to object that those people have no idea how hard it is to create a good recommendation engine. As for whether similar techniques can be applied to education, he professes to reserve judgment. "I believe it's worth trying to solve this problem. When I look at the transcripts of our tutoring sessions, when someone gets stuck, I don't get the impression it always takes an Einstein to answer these questions. I get the impression there's specific ways students get stuck, specific gaps that they have. And there are also patterns in how you practice things, that are documented in the literature -- how you repeat practice things to have a sustained learning effect. I see a lot of hope that it is computer-izable. I think we can make great progress, especially compared to where we are today at Udacity where we don't have much of this technology."
Politics And University Politics
One reason Udacity has had to contend with controversy is that online education and educational technology in general have a political and emotional dimension. When the California Faculty Association helped publicize Udacity's setbacks at SJSU, it was reacting in part to the fears of its members that their job security was being undermined -- a fear intensified by the enthusiasm of some politicians for MOOCs as a replacement for traditional higher education.
Thrun recently told MIT Technology Review that in five years Udacity will be "just like a university, but we'll be a university for the 21st century."
Reminded of that, he insists he wasn't saying online education will replace the university as we know it, only that the next-generation university "will be much more pervasive, much more available throughout your life." With skills going stale much quicker due to the pace of technological change, people will need ongoing resources like Udacity, he said.
Thrun is conscious that journalists tend to see the transformation of higher education through the lens of their own industry, where the migration of news content from print to online diluted the power of newspapers and magazines, which found themselves competing for ad dollars with every blogger. It may be an easy comparison, but the differences are important, Thrun said.
When newspapers went online, there was no shortage of news content in the market, and the new distribution model did not lead to an increase in the public's appetite for that content -- the audience and the ad dollars just got spread thinner. He offers a different analogy: "After the movie industry started about 100 years ago, the amount of time people spent consuming fictitious play went up by some estimates by a factor of 28." Access to higher education is relatively "locked up," and online education provides greater access to what has been a scarce, expensive resource, he said.
"I think it's completely conceivable that, in the future, education could be 20 times bigger than it is today -- in which case, it's going to be great for the industry, not bad," Thrun said.