Like good prison guards, the teachers’ unions are trying to stop this transformation from ever getting off the ground. In state after state, companies like K12 have had to fight costly battles to have the chance to compete. The prison guards didn’t try to make Sal Khan’s math lessons better, and they aren’t trying to figure out how we can improve virtual schools. They’re trying to kill them, depriving millions of students of a potentially better future for the sake of their own privileges. Whatever challenges virtual charter schools face, students in thousands of traditional schools are performing far below grade level and eventually giving up. Yet the unions are suing to keep those schools open. One in four students nationwide does not graduate high school, and in some places the rate is much higher.11 Nearly 40 percent of students in Chicago drop out of school; among African Americans, that number is over 60 percent.12 One out of three fourth graders cannot read, scoring “below basic” on literacy tests. One-third of eighth graders and 38 percent of twelfth graders read below grade level.13 The unions have nothing to boast of, and the prisoners of their schools could hardly do worse in a virtual school.
We haven’t begun to see the real breakout that virtual schools could offer. Their potential to outperform traditional schools is impressive, but the breakout will come only when they integrate the next big leap in education: learning science.
Bror Saxberg and Big Data
If you glanced at Bror Saxberg’s résumé, you’d probably assume that he’s a cloistered professor at an elite university. He holds two bachelor’s degrees from the University of Washington, one in mathematics and one in electrical engineering. As a Rhodes scholar, he received a degree in mathematics from Oxford, and he washed it all down with a PhD in electrical engineering and computer science from MIT, which he earned simultaneously with an MD from Harvard Medical School.
Despite his twenty-four-karat academic credentials, Saxberg is no snob about education. Indeed, he has chosen a career many of his peers from Harvard, MIT, and Oxford might (wrongly) view with disdain. The chief learning officer for Kaplan, one of the largest for-profit education companies, Saxberg is responsible for improving the learning of more than 1.5 million14 students worldwide. I got to know him while working with Kaplan on a series of short courses for NewtUniversity.com.
Google can determine traffic conditions on thousands of roadways by analyzing where millions of Android phones are. The Obama campaign optimized its donation form by live-testing hundreds of variations. Saxberg thinks that “big data”—generated by millions of students online—can likewise transform education.
With the zeal of an entrepreneur, Saxberg talks about presenting every concept to every student in the way that is best suited for that student. He recently tested Kaplan’s approach to preparing students for the logical reasoning problems on the LSAT, which many find particularly tricky. The company had invested heavily in an hour-long video that was loaded with fancy animations to illustrate the problems, but Kaplan’s “learning engineers” (people trained in learning science but applying it to practical problems at scale) thought there was room for improvement.
Using Amazon’s “Mechanical Turk” platform,15 the Kaplan team created a task that directed part of his test group to the hour-long video and then gave them a few of the LSAT questions. The second part of the test group read through a couple of “worked examples”—annotated text versions of how someone solved the problems correctly, shown by Australian learning researcher John Sweller to improve problem-solving—instead of watching the video. A third group got no explanation at all.
For a few thousand dollars, by the end of a day or two of testing, Kaplan had a randomized, controlled study comparing its video approach with a new alternative—approximating research that traditionally would take years and, in traditional university settings, cost tens of thousands of dollars or more. Saxberg’s team quickly discovered that there was little difference between the performance of the students who watched the video and the performance of the students who had no preparation. The video was ineffective.
The subjects who read through the worked examples, however, performed considerably better, and their preparation took only about fifteen minutes, compared with a full hour of video instruction. Kaplan is adjusting its method for teaching the LSAT problems accordingly.
Saxberg’s results reflected what he says science shows about how we learn: the most important factor is usually not the teacher or even the student’s innate talent. “What really makes the difference,” he says, “is deliberate practice.” By and large, people master the multiplication table or factoring quadratic functions the same way they become great basketball players or excellent airplane pilots: many (not all) of the key decisions and tasks need to be practiced and practiced, with feedback, until they are hardwired into their brains. An hour spent watching a video makes little difference, but an extra thirty minutes of practice problems helps dramatically.
Although this discovery, in itself, has nothing to do with technology, it points us to a more technology-based education system, Saxberg argues. An online environment, he told me, enables a “competency-based model of instructional learning rather than a schedule-based model, one based on seat-time.” In such a system, students advance not when the class schedule dictates, but when they have actually learned the material—and they get sufficient practice and feedback, made efficient by technology, for each of them to get there. The same technology, Saxberg said, “enables things like … data-gathering modules and diagnostics that tie into banks of hopefully usefully designed training materials in ways that just weren’t possible before, when you just had to work with an individual instructor who was in your face because there was no other way to get the learning that you needed.”
Like Salman Khan, Saxberg sees the explosive potential for personalized learning. “You can start to track how people are doing and you can speed up or slow down the pace at which you’re giving them examples. There is a lot of great stuff to explore that hasn’t even begun to be explored.”
The sea change happens when you plug tens of thousands of students into such a system, he said. The avalanche of performance data enables those “learning engineers” to observe and test patterns of what is working and what isn’t and for what kinds of students. Which students should have the video feed turned off? Who works best in a small group, and who learns best on her own? Who needs videos or practice types Q, R, and S as opposed to A, B, and C? What happens if you give both Q and B—does performance go up, and is it worth the extra time? For whom?
In a large population—say, a big state education system—educators can quickly gather evidence about the optimal ways to teach their material. “Having a lot of learners that are in a systematic learning environment gives you the potential to do a lot of work,” Saxberg said. “Right now at Kaplan, we are setting up to do … maybe north of one hundred randomized controlled trials each year. We have some courses that have more than a thousand students starting every month. So you think about that: a thousand students starting every month, you should be able to do ten hundred-student randomized controlled trials every single month, so even a single course might get you a pace north of one hundred controlled trials in a year.”
Today, little classroom teaching is based on solid evidence of how students learn most effectively. Computerized learning environments tied in with teaching will finally let educators systematically apply the insights of science to the classroom. “We are applying results that have been around for decades in the world of learning science research,” Saxberg says, “about media and images and texts and the structure of the pieces and how you do some of the feedback and worked examples. Good evidence from laboratories suggests that these things make significant differences to learning, but they are not known or used at scale.” Soon, he says, parents could demand to see the evidence that supports how their children are being taught.
In an online learning environment, you could apply new insights to the entire sys
tem, hundreds of thousands of students, just as fast as you discovered what works for various types of learners. If one teacher had an idea for teaching fractions that improves practice results by 5 percent in certain kinds of students, all of them could benefit from it the next week.
We could do all of this today. There is no good reason that every child in America can’t receive the effective education that the analysis of big data makes possible. But how will teachers’ unions react to the data-drenched reforms that Bror Saxberg describes? Judging by their responses to every other proposed reform—teacher evaluations, merit pay, and the abolition of tenure, to name a few—we can expect a fight to the death. The victims, of course, are the students, who will learn less than they need to compete in the modern world.
Saxberg isn’t pointing his fire hose of data at K–12 education, however. Kaplan, after all, runs the world’s second-largest online university. Despite the scorn with which many professors at the “elite” universities view for-profit colleges, Kaplan is in some ways far ahead of them. It has found a way to give people the education and skills they need without breaking the bank. As the pace of innovation will require a workforce of lifelong learners, this is essential. Just as Netflix allows viewers to log on and view what they want on their own schedule, online education enables people with full-time jobs to learn what they need on their own schedule.
Compare online students with those at traditional universities, who take on tens of thousands of dollars of debt to finance their education. What do they get at Harvard or Berkeley for that kind of money? They probably get a professor who views teaching introductory biology as nothing short of purgatory. He is likely to be far more interested in his research into gametogenesis or in his forthcoming appearance on PBS than in teaching college freshmen the basics of plant photosynthesis. There’s a good chance, then, that he will recycle the same PowerPoint he has been using for the last twenty years, reciting the same lecture he has always given, without much thought about whether it’s the “optimal” way to teach the material or not. Even if a few professors are dedicated to the success of their students, the research-based tenure system means they have to regard it as a hobby, not their primary job. In the clearest signal of its bias against teaching, the system rewards the “best” faculty by removing teaching from their workload.
There is a deeper problem as well: “Historically, we conflated being an expert with being an expert on how to teach,” Saxberg told me. “The two are totally different. Experts have been doing their work for decades, often, and 70 percent or more of what they do is done in automated ways by their minds.” Many of them “tend to only teach what they think consciously about, which means they are going to leave out 70 percent of what a novice actually needs to know,” he said. “So this is one of the problems with the current education system, where experts teach the class, and the Nobel Prize winner has no idea how to describe what he knows so deeply, it’s become non-conscious. Nobody quite understands what he just said, nor realizes what has been left out.”
If students are lucky, their professor might hold office hours twice a week. The bolder students can show up to annoy him if they didn’t understand what he said in class, in which case he will patronizingly direct them to the nearest teaching assistant. If that doesn’t help, they’ve got the textbook. (No wonder traffic is so heavy at Khan Academy.)
In contrast to the challenged state of traditional higher education, the nuts-and-bolts approach Bror Saxberg describes is downright refreshing. Unlike Kaplan, America’s elite universities usually have no one whose job is to make sure professors are teaching their courses in the most effective manner. They don’t try to follow their students from course to course and then after college to see if what they learned is helping them to succeed. Where are the “learning engineers”?
Even if Columbia or Cornell did have someone to compare the test scores of Professor Watson’s section of Calculus 101 with those of Professor Price’s section, what could that person do? If he told Professor Price, “You need to do it the way Professor Watson is doing it, since his students consistently score 30 percent higher than yours,” Professor Price would probably tell him to get lost. Everyone in the modern world of higher education understands that lecturing the debt-laden student body is what the faculty does to pay the bills but that it’s a secondary concern at best.
Udacity
Why are Harvard, Princeton, and Yale behind Kaplan and the University of Phoenix when it comes to personalizing their courses or making their lectures available to students at any time of day? Why have they largely failed to adopt competency-based adaptive learning systems so students can set their own pace?
Until recently, the accreditation guild that has fueled much of the student loan bubble protected America’s leading universities from meaningful competition. The average cost of a college education has increased twelvefold, or 1,120 percent, since 1978, according to Bloomberg—four times the rate of inflation.16 In 2013, two-thirds of graduating college students had outstanding school loans, with an average balance amounting to 60 percent of their annual salary.17
For-profit education companies (which are constantly under regulatory assault by the old order) are often accused of churning students through the system and giving them a degree without teaching them much, leaving them burdened with debt they will struggle to pay off. It is often more accurate, however, to level this accusation at traditional “non-profit” universities, including America’s leading institutions of higher education, than at their for-profit competitors. As the work of Salman Khan and Bror Saxberg suggests, this unsustainable model could be on the verge of collapse.
One pioneer working to bring down the cost and improve the quality of higher education is Sebastian Thrun, a vice president at Google and a professor at Stanford. He already has a few world-changing inventions under his belt. He helped create the self-driving cars that have safely navigated six hundred thousand miles of California roads—a breakthrough that I explore in chapter six—and he helped pioneer Google’s Street View technology, which has indexed images of almost every inch of public road in the United States and many countries around the world.
What could prove to be Thrun’s most disruptive project, however, began not at Google but at Stanford. Along with Google’s director of research, Peter Norvig, he taught Stanford’s computer science course on artificial intelligence. The pair decided to offer a version of their course online for free, unsure of how many people would sign up.
Before they knew it, more than 160,000 people from all over the world had registered.
As Thrun recounted to the Wall Street Journal last year, there was just one problem: “I had forgotten to tell Stanford about it.… Stanford said ‘If you give the same exams and the same certificate of completion [as Stanford does], then you are really messing with what certificates really are. People are going to go out with the certificates and ask for admission [at the university] and how do we even know who they really are?’ And I said: I. Don’t. Care.”18
By the end of the semester, twenty-three thousand of the original 160,000 had completed the full course. The best Stanford student was ranked 411.19 The results were stunning to everyone. As Thrun explained it to me, “The fact that we brought twenty-three thousand people to the finish line meant that in this specific quota I taught more people artificial intelligence than all the other professors in the world combined.” Reaction to this remarkable achievement was mixed, however. “To the world, it mostly meant, ‘Wow, we’re entering a new era of education where things become accessible, affordable, and also transparent, and as a result higher quality,’” Thrun said. “To some, it means, ‘Wow, did I just lose my job?’”
Thrun was amazed by the results of his experiment, which became one of the first so-called MOOCs (short for “massively open online courses”). He was so excited by the new possibilities that he cut his work at Google to part-time and founded an education company, Udacity, with the goal of reducing th
e cost of a college degree by 90 percent. “I think we found the magic formula,” he told me. “That may sound pretentious, but the magic formula for me is high-quality education at scale … I think we owe students a personalized, high-quality experience.”
In addition to a version of Thrun and Norvig’s original class on artificial intelligence, Udacity now offers college-level introductions to computer science, statistics, physics, and psychology, as well as a smattering of more advanced courses. In “Differential Equations in Action,” students discover how to use mathematics to “rescue the Apollo 13 astronauts, stop the spread of epidemics, and fight forest fires.” In another class, students develop sophisticated browser-based computer games.
The most-exciting courses on Udacity are those offered in conjunction with traditional, credit-granting universities. Udacity took a radical step in 2013 when it partnered with San Jose State University in California to offer introductory-level math, computer science, and psychology courses for credit, enabling San Jose students to apply their work toward a real degree. The cost? One hundred fifty dollars for each course.
It was a comment by Bill Gates that motivated Thrun to work with San Jose State. The early Udacity model, Gates said, worked only for extremely self-motivated people, maybe 5 to 10 percent of the learning population, since they had to work independently. “I took this to heart and thought about this,” Thrun said. The San Jose project was his first effort at solving the problem.
“We picked exactly the opposite” of the top 10 percent Bill Gates had in mind, “those students that traditionally are the hardest to reach and educate,” Thrun told me. “We picked inner-city high school students from low-income neighborhoods, and they come with enormous … handicaps. We deliberately even picked a group of students that not only had failed the entrance exam for college in mathematics, but also had failed the subsequent remedial education that they had to undergo to make up for the remedial math. Out of those kids, about 6 to 10 percent can stay in college. So these are basically kids on the way out from college. And we picked those because we wanted to understand how to make education work for everybody.”
Breakout: Pioneers of the Future, Prison Guards of the Past, and the Epic Battle That Will Decide America's Fate Page 5