Imagine every student has a tireless personal tutor, an artificially intelligent and inexhaustible companion that magically knows everything, knows the student, and helps her learn what she needs to know. “‘You guys sound like you’re from the future,’” Jose Ferreira, the CEO of the education technology startup Knewton, says. “That’s the most common reaction we get from others in the industry.”
When I first met Ferreira four years ago, this kind of talk sounded like typical Silicon Valley bluster from another scruffy, boyish founder of a technology startup. Today, he can back up the kinds of breakthroughs he says his company can deliver: several million data points generated daily by each of 1 million students from elementary school through college, using Knewton’s “adaptive learning” technology to study math, reading, and other fundamentals. Peter Thiel, the PayPal cofounder, Facebook investor, and an early investor in Knewton, told Knewton’s staff recently that the company has two key characteristics he looks for in a deal. “Before they happen, everybody thought it was impossible. Afterwards it’s too late for anyone else, because they’ve already done it.”
Adaptive learning is an increasingly popular catchphrase denoting educational software that customizes its presentation of material from moment to moment based on the user’s input. It’s being hailed as a “revolution” by both venture capitalists and big, established education companies.
Starting this fall, Knewton’s technology will be available to a vast majority of U.S. colleges and universities and K-12 school districts through new partnerships with three major textbook publishers: Pearson, MacMillan, and Houghton Mifflin Harcourt. And Ferreira’s done all this even though he says neither his investors nor his competition, to say nothing of the public or the press, really understand what Knewton can do.
But here’s the vision. Within five or 10 years, the paper textbook and mimeographed worksheet will be dead. Classroom exercises and homework—text, audio, video, games—will have shifted entirely to the iPad or equivalent. And adaptive learning will help each user find the exact right piece of content needed, in the exact right format, at the exact right time, based on previous patterns of use.
In an age of swelling class sizes, teacher layoffs, and students with a vast array of special needs and learning styles, some reformers hail these software systems as a savior that could make learning more customized and effective and teaching more efficient. While battle lines are sharp in K-12 school reform over issues from charters to the Common Core national curriculum standards, digital innovations have fans across the political spectrum for their power to engage students and bring the classroom into the 21st century.
Here’s what Ferreira thinks this software-powered learning can do. “Right now about 22 percent of the people in the world graduate high school or the equivalent. That’s pathetic. In one generation we could get close to 100 percent, almost for free.”
Like a lot of technology entrepreneurs, Ferreira has a personal beef with the existing education system. “I found school very boring and frustrating,” he says, his low, quick voice barely audible over the roar of air conditioners on the roof deck of Knewton’s new Union Square digs in Manhattan. The company is hiring as fast as it can. Clad in flip-flops, a T-shirt, and shorts, Ferreira presides over an office stocked with an espresso machine and several brands of beer. On the day I visit, employees are having a communal lunch.
“The factory model of education is a gargantuan bureaucracy. Some kids are good fits—I wasn’t. The system gives you bad grades and tells you you’re stupid. You don’t think, if this kid’s not a good fit it could be the system’s fault.”
‘All the content behind education is going to move online in the next 10 years.’
He inherited his rebellious streak, but his family also put tremendous pressure on him to succeed. Ferreira’s parents were native-born white Africans who came to America when he was 2, after his father’s anti-apartheid political activism, in Ferreira’s words, “got us kicked out” of South Africa. They settled in Chevy Chase, Maryland, where he attended public schools that were high quality but, to him, dull. He was the type to torment the teachers with smart-aleck questions, skip homework assignments, and cram all the material the night before the test.
After graduating from Carleton College in Minnesota, where he studied philosophy and mathematics, Ferreira faced anemic job prospects in the early-’90s recession. Living in San Francisco, he made ends meet as an SAT tutor for Kaplan, the largest test-prep company, by day, and as a poker player in casinos at night. In both roles, he delighted in playing the numbers, getting an edge over the house. “Casinos are statistics in game form. And I had started teaching because I love brainteasers. At some point I had taken every standardized test out there—the SATs, the GREs, the GMATs, the MCATs. I just took them for fun. ”
To his parents, however, he was the black sheep. They hadn’t told the rest of the family what he was doing for a living. “One day my parents came into town to take me to dinner. My dad was worried I was a failure. My mom asked, ‘Why do you think you haven’t dated anyone in two years?’—because all my nights and weekends were spent at the poker table. Kaplan had offered me a full-time job, so I took it. I figured it was time to get serious.” Ferreira extended his love of brainteasers to the art and science of taking—and beating, and helping others beat—standardized tests. He never saw this as cheating, rather as righting an injustice for kids like himself, smart but restless, whose fates were being decided on the basis of an arbitrary three-hour test.
His confidence and love of risk, meanwhile, served him well. At one point while working for Kaplan, Ferreira discovered a flaw on the GREs that turned what was supposed to be a very technical math question into something a child could do. The vulnerability forced the Educational Testing Service, makers of the GREs, to delete an entire section of questions. They referred to the test-prep hacker privately as “the Antichrist.” “They had to pull this section of the test. They gave me credit publicly,” he says, grinning with pride. Kaplan turned the admission from ETS into the basis of an international marketing campaign for their test-prep services. “We put that on everything we sent out after that. I mean nurses in the Philippines were getting postcards about it.”
Ferreira could have stayed at Kaplan, rising through the ranks, but the dot-com era was dawning, and he became increasingly captivated by the power of technology to transform teaching and learning. He enrolled in Harvard Business School, but continued to play poker for fun and money, sometimes sleeping in his car at 4 a.m. in the parking lot of an Atlantic City casino, unwilling to spring for a hotel room despite the several thousand dollars in his pocket. MBA in hand, he went to work at Goldman Sachs, then left to found a mapping-software startup that imploded in the 2000 bust. He worked as a strategist for the John Kerry campaign for a while (he’s a nephew of Teresa Heinz Kerry). He was a venture capitalist.
“I was like, I’ve had these ideas about education germinating for a long time and I just got to the point where I said, I have to do this.” He started Knewton in 2008 with more or less the same vision he espouses today: to enable digital technology to transform learning for everyone and to build the company that dominates that transformation.
“Look at what other industries the Internet has transformed,” he told me at one of our first meetings, at a technology conference in 2009. “It laid waste to media and is rebuilding it—print, digital, video, music. Travel, hotels, restaurants, retail—anything with a big information component. But for whatever reason, people don’t see it with education. It is blindingly obvious to me that it will happen with education.
“All the content behind education is going to move online in the next 10 years. It’s one giant Oklahoma land grab—one tectonic shift. And that is what Knewton is going to power.”
The recommendation engine is a core technology of the Internet, and probably one you encounter every day. Google uses recommendations: other people who entered these search terms clicked on this page, so we’ll show it to you first. Amazon uses them: other people who bought this book also bought that book.
The more you use one of these websites, the more it knows about you—not just about your current behavior, but about all the other searches and clicks you’ve done. In theory, as you spend more time with a site its recommendations will become more personalized even as they also draw on everyone else’s interactions within the platform.
Knewton, at base, is a recommendation engine but for learning. Rather than the set of all Web pages or all movies, the learning data set is, more or less, the universe of all facts. For example, a single piece of data in the engine might be the math fact that a Pythagorean triangle has sides in the ratio 3-4-5, and you can multiply those numbers by any whole number to get a new set of side lengths for this type of triangle. Another might be the function of “adversatives” such as “but,” “however,” or “on the other hand” in changing the meaning of an English sentence. Ferreira calls these facts “atomic concepts,” meaning that they’re indivisible into smaller concepts—he clearly also relishes the physics reference.
When a textbook publisher like Pearson loads its curriculum into Knewton’s platform, each piece of content is tagged with the appropriate concept or concepts. Let’s say your school bought the Knewton-powered MyMathLab online system, using the specific curriculum based on, say, Lial’s Basic College Mathematics 8e. When a student logs on to the system, she first takes a simple placement test or pretest from the book, which has been tagged with the relevant “atomic concepts.” As a student reads the text or watches the video and answers the questions, Knewton’s system is “reading” the student as well—timing every second on task, tabulating every keystroke, and constructing a profile of learning style: hesitant or confident? Guessing blindly or taking her time?
Based on the student’s answers, and what she did before getting the answer, “we can tell you to the percentile, for each concept: how fast they learned it, how well they know it, how long they’ll retain it, and how likely they are to learn other similar concepts that well,” says Ferreira. “I can tell you that to a degree that most people don’t think is possible. It sounds like space talk.” By watching as a student interacts with it, the platform extrapolates, for example, “If you learn concept No. 513 best in the morning between 8:20 and 9:35 with 80 percent text and 20 percent rich media and no more than 32 minutes at a time, well, then the odds are you’re going to learn every one of 12 highly correlated concepts best that same way.”
As a student reads, Knewton’s system ‘reads’ the student as well.
The platform forms a personalized study plan based on that information and decides what the student should work on next, feeding the student the appropriate new pieces of content and continuously checking the progress. A dashboard shows the student how many “mastery points” have been achieved and what to do next. Teachers, likewise, can see exactly which concepts the student is struggling with, and not only whether the homework problems have been done but also how many times each problem was attempted, how many hints were needed, and whether the student looked at the page or opened up the video with the relevant explanation.
The more people use the system, the better it gets; and the more you use it, the better it gets for you.
In a traditional class, a teacher moves a group of students through a predetermined sequence of material at a single pace. Reactions are delayed—you don’t get homework or pop quizzes back for a day or two. Some students are bored; some are confused. You can miss a key idea, fall behind, and never catch up.
Software-enabled adaptive learning flips all of this on its head. Students can move at their own speed. They can get hints and instant feedback. Teachers, meanwhile, can spend class time targeting their help to individuals or small groups based on need.
Ferreira is able to work with competitors because his software can power anybody’s educational content. But before it had any content partners, as a proof of concept, Knewton built its own remedial college math course using its software platform. Math Readiness was adopted starting in the summer of 2011 at Arizona State University; the University of Nevada, Las Vegas; and the University of Alabama.
At ASU, students worked through the computer material in Knewton’s Math Readiness program on their own or in small groups, with instructors spending face-to-face time working on problem solving, critical thinking, and troubleshooting specific concepts. After two semesters of use, course withdrawal rates dropped by 56 percent and pass rates went from 64 percent to 75 percent. At Alabama, pass rates rose from 70 percent to 87 percent, and at UNLV, where entering students were given the chance to take the course online in the summer before they started college, the percentage who then qualified for college algebra went from 30 percent to 41 percent.
“Before this, I worked on the assumption that all students were at the same place. Now, because they progress at different rates, I meet them where they are,” Irene Bloom, a math lecturer at ASU, told an education blog about the pilot program. “I have so much more information about what my students do (or don’t do) outside of class. I can see where they are stuck, how fast they are progressing, and how much time and effort they are putting into learning math.”
The Knewton system uses its analytics to keep students motivated. If it notices that you seem to have a confidence problem, because you too often blow questions that should be easy based on previous results, it will start you off with a few questions you’re likely to get right. If you’re stuck, choosing the wrong answer again and again, it will throw out broader and broader hints before just showing you the right answer. It knows when to drill you on multiplication and when to give you a fun animated video to watch.
It turns out that personalizing in this way can speed up learning. In the first year, 45 percent of ASU students in a 14-week course learned the material four weeks ahead of schedule. “We’ve had students finish a semester course in 14 days,” says Ferreira, clearly psyched about setting free the kind of bored student he was. “For the whole history of the human race until now, she had to stay in that class the whole time.”
Better data is giving more options to the student who didn’t succeed as well. Students may not yet know enough to pass the final exam, but a close read of their answers shows that they are making slow and steady progress. “In the past, those students would have dropped out of school,” he says. In fact, the vast majority of students placed into remedial math at community colleges never get their degrees. “Instead, we were able to say, give them another semester and they’ll get it. Their whole life has now changed.”
Global spending on education is in the trillions of dollars, and demand still far outstrips supply. Ed-tech venture funding reached $1.1 billion in 2012. Rupert Murdoch has launched a $200 million tablet and digital curriculum brand; Apple and Google are building significant education businesses in devices and apps, respectively. Sebastian Thrun, Daphne Koller, and Andrew Ng, all Stanford professors, raised almost $100 million in venture capital to create online course platforms Udacity and Coursera, which deliver free video versions of courses created at the world’s most famous universities. The Khan Academy, a nonprofit library of free instructional videos and online exercises, has more than 6 million users per month and is used in 30,000 classrooms. Microsoft founder Bill Gates has perhaps done more than any single person to seed the growth of innovation in education, giving away hundreds of millions of dollars to both nonprofits and for-profits through the Gates Foundation’s education programs. At his keynote at SXSWedu, an entire subconference of the South by Southwest technology gathering in Austin, Texas, dedicated to ed-tech, he compared the digital revolution in education to the discovery of the polio vaccine.
These are early days, but the questions are mounting: Can all this innovation narrow the stubborn achievement gap between rich and poor or black and white? Can it lower the cost of college or make a dent in student-loan debt? And, as must be asked of all things tech, can it complement the human elements of education?
Research indicates that emotional qualities like grit, persistence, and motivation may be even more important to students’ success than the knowledge or skills they acquire, and they all depend heavily on human relationships. Knowledge acquisition is the only aspect of education that today’s digital technology seems especially well adapted to. So far, most software applications, platforms, apps, and games, including Knewton’s, have been optimized for transferring quantitative, bounded bodies of facts in domains like math, science, or engineering, as well as basic literacy and grammar. An adaptive-learning platform like Knewton’s is helpless to tabulate or analyze a student’s insight in class discussions, the special brilliance of an essay, or creativity in a group presentation; anything that complex requires human discretion.
Still, within that subdomain of knowledge transfer, or what we’re used to thinking of as “learning,” interventions like Knewton’s are having intriguing results. As the Common Core attempts to raise standards, there is a hunger for proof that students are achieving mastery, not just covering material.
Knewton’s launch of a retail platform planned for early next year will allow anyone to upload any piece of educational content from the Web—whether a teacher’s own lesson, a TED talk, or a set of Khan Academy exercises. (Until now only textbook publishers could engage with the software.) The laborious process of tagging the content by “atomic concept” will be crowd sourced. If the community gets engaged, Knewton’s recommendation engine will be helping to sort the universe of free and open educational resources to provide much richer learning experiences to anyone, again, for free.
The idea of learning as an unmediated, infinitely scalable experience, as customized and seamless, given sufficient bandwidth, as downloading a song or watching a YouTube video, may turn out to be a fantasy of the early digital age—the era we’re in now. But the promise of putting all the intelligence of big data, rich content, and analytics in the hands of talented teachers and learners and setting them free, together, to explore at their own pace, in their own way—that is far more compelling, and achievable given what we’ve already seen.
In a rare moment of humility, Ferreira agrees. “In the end,” he says, “Knewton is just a tool.”
From our Aug. 2, 2013; What If You Could Learn Everything?