Select your localized edition:

Close ×

More Ways to Connect

Discover one of our 28 local entrepreneurial communities »

Be the first to know as we launch in new countries and markets around the globe.

Interested in bringing MIT Technology Review to your local market?

MIT Technology ReviewMIT Technology Review - logo

 

Unsupported browser: Your browser does not meet modern web standards. See how it scores »

A hundred years ago, higher education seemed on the verge of a technological revolution. The spread of a powerful new communication network—the modern postal system—had made it possible for universities to distribute their lessons beyond the bounds of their campuses. Anyone with a mailbox could enroll in a class. Frederick Jackson Turner, the famed University of Wisconsin historian, wrote that the “machinery” of distance learning would carry “irrigating streams of education into the arid regions” of the country. Sensing a historic opportunity to reach new students and garner new revenues, schools rushed to set up correspondence divisions. By the 1920s, postal courses had become a full-blown mania. Four times as many people were taking them as were enrolled in all the nation’s colleges and universities combined.

The hopes for this early form of distance learning went well beyond broader access. Many educators believed that correspondence courses would be better than traditional on-campus instruction because assignments and assessments could be tailored specifically to each student. The University of Chicago’s Home-Study Department, one of the nation’s largest, told prospective enrollees that they would “receive individual personal attention,” delivered “according to any personal schedule and in any place where postal service is available.” The department’s director claimed that correspondence study offered students an intimate “tutorial relationship” that “takes into account individual differences in learning.” The education, he said, would prove superior to that delivered in “the crowded classroom of the ordinary American University.”

We’ve been hearing strikingly similar claims today. Another powerful communication network—the Internet—is again raising hopes of a revolution in higher education. This fall, many of the country’s leading universities, including MIT, Harvard, Stanford, and Princeton, are offering free classes over the Net, and more than a million people around the world have signed up to take them. These “massive open online courses,” or MOOCs, are earning praise for bringing outstanding college teaching to multitudes of students who otherwise wouldn’t have access to it, including those in remote places and those in the middle of their careers. The online classes are also being promoted as a way to bolster the quality and productivity of teaching in general—for students on campus as well as off. Former U.S. secretary of education William Bennett has written that he senses “an Athens-like renaissance” in the making. Stanford president John Hennessy told the New Yorker he sees “a tsunami coming.”

The excitement over MOOCs comes at a time of growing dissatisfaction with the state of college education. The average price tag for a bachelor’s degree has shot up to more than $100,000. Spending four years on campus often leaves young people or their parents weighed down with big debts, a burden not only on their personal finances but on the overall economy. And many people worry that even as the cost of higher education has risen, its quality has fallen. Dropout rates are often high, particularly at public colleges, and many graduates display little evidence that college improved their critical-thinking skills. Close to 60 percent of Americans believe that the country’s colleges and universities are failing to provide students with “good value for the money they and their families spend,” according to a 2011 survey by the Pew Research Center. Proponents of MOOCs say the efficiency and flexibility of online instruction will offer a timely remedy.

But not everyone is enthusiastic. The online classes, some educators fear, will at best prove a distraction to college administrators; at worst, they will end up diminishing the quality of on-campus education. Critics point to the earlier correspondence-course mania as a cautionary tale. Even as universities rushed to expand their home-study programs in the 1920s, investigations revealed that the quality of the instruction fell short of the levels promised and that only a tiny fraction of enrollees actually completed the courses. In a lecture at Oxford in 1928, the eminent American educator Abraham Flexner delivered a withering indictment of correspondence study, claiming that it promoted “participation” at the expense of educational rigor. By the 1930s, once-eager faculty and administrators had lost interest in teaching by mail. The craze fizzled.

Is it different this time? Has technology at last advanced to the point where the revolutionary promise of distance learning can be fulfilled? We don’t yet know; the fervor surrounding MOOCs makes it easy to forget that they’re still in their infancy. But even at this early juncture, the strengths and weaknesses of this radically new form of education are coming into focus.

Rise of the MOOCs

“I had no clue what I was doing,” ­Sebastian Thrun says with a chuckle, as he recalls his decision last year to offer Stanford’s Introduction to Artificial Intelligence course free online. The 45-year-old robotics expert had a hunch that the class, which typically enrolls a couple of hundred undergraduates, would prove a draw on the Net. After all, he and his co-professor, Peter Norvig, were both Silicon Valley stars, holding top research posts at Google in addition to teaching at Stanford. But while Thrun imagined that enrollment might reach 10,000 students, the actual number turned out to be more than an order of magnitude higher. When the class began, in October 2011, some 160,000 people had signed up.

The experience changed Thrun’s life. Declaring “I can’t teach at Stanford again,” he announced in January that he was joining two other roboticists to launch an ambitious educational startup called Udacity. The venture, which bills itself as a “21st-century university,” is paying professors from such schools as Rutgers and the University of Virginia to give open courses on the Net, using the technology originally developed for the AI class. Most of the 14 classes Udacity offers fall into the domains of computer science and mathematics, and Thrun says it will concentrate on such fields for now. But his ambitions are hardly narrow: he sees the traditional university degree as an outdated artifact and believes Udacity will provide a new form of lifelong education better suited to the modern labor market.

Udacity is just one of several companies looking to capitalize on the burgeoning enthusiasm for MOOCs. In April, two of Thrun’s colleagues in Stanford’s computer science department, Daphne Koller and Andrew Ng, rolled out a similar startup called Coursera. Like Udacity, Coursera is a for-profit business backed with millions of dollars in venture capital. Unlike Udacity, Coursera is working in concert with big universities. Where Thrun wants to develop an alternative to a traditional university, Koller and Ng are looking to build a system that established schools can use to deliver their own classes over the Net. Coursera’s original partners included not only Stanford but Princeton, Penn, and the University of Michigan, and this summer the company announced affiliations with 29 more schools. It already has about 200 classes on offer, in fields ranging from statistics to sociology.

On the other side of the country, MIT and Harvard joined forces in May to form edX, a nonprofit that is also offering tuition-free online classes to all comers. Bankrolled with $30 million from each school, edX is using an open-source teaching platform developed at MIT. It includes video lessons and discussion forums similar to those offered by its for-profit rivals, but it also incorporates virtual laboratories where students can carry out simulated experiments. This past summer, the University of California at Berkeley joined edX, and in September the program debuted its first seven classes, mainly in math and engineering. Overseeing the launch of edX is Anant Agarwal, the former director of MIT’s Computer Science and Artificial Intelligence Laboratory.

The leaders of Udacity, Coursera, and edX have not limited their aspirations to enhancing distance learning. They believe that online instruction will become a cornerstone of the college experience for on-campus students as well. The merging of virtual classrooms with real classrooms, they say, will propel academia forward. “We are reinventing education,” declares Agarwal. “This will change the world.”

 e-instiutions chart

 

Professor Robot

Online courses aren’t new; big commercial outfits like the University of Phoenix and DeVry University offer thousands of them, and many public colleges allow students to take classes on the Net for credit. So what makes MOOCs different? As Thrun sees it, the secret lies in “student engagement.” Up to now, most Internet classes have consisted largely of videotaped lectures, a format that Thrun sees as deeply flawed. Classroom lectures are in general “boring,” he says, and taped lectures are even less engaging: “You get the worst part without getting the best part.” While MOOCs include videos of professors explaining concepts and scribbling on whiteboards, the talks are typically broken up into brief segments, punctuated by on-screen exercises and quizzes. Peppering students with questions keeps them involved with the lesson, Thrun argues, while providing the kind of reinforcement that has been shown to strengthen comprehension and retention.

Norvig, who earlier this year taught a Udacity class on computer programming, points to another difference between MOOCs and their predecessors. The economics of online education, he says, have improved dramatically. Cloud computing facilities allow vast amounts of data to be stored and transmitted at very low cost. Lessons and quizzes can be streamed free over YouTube and other popular media delivery services. And social networks like Facebook provide models for digital campuses where students can form study groups and answer each other’s questions. In just the last few years, the cost of delivering interactive multimedia classes online has dropped precipitously. That’s made it possible to teach huge numbers of students without charging them tuition.

It’s hardly a coincidence that Udacity, Coursera, and edX are all led by computer scientists. To fulfill their grand promise—making college at once cheaper and better—MOOCs will need to exploit the latest breakthroughs in large-scale data processing and machine learning, which enable computers to adjust to the tasks at hand. Delivering a complex class to thousands of people simultaneously demands a high degree of automation. Many of the labor-intensive tasks traditionally performed by professors and teaching assistants—grading tests, tutoring, moderating discussions—have to be done by computers. Advanced analytical software is also required to parse the enormous amounts of information about student behavior collected during the classes. By using algorithms to spot patterns in the data, programmers hope to gain insights into learning styles and teaching strategies, which can then be used to refine the technology further. Such artificial-intelligence techniques will, the MOOC pioneers believe, bring higher education out of the industrial era and into the digital age.

To fulfill their grand promise, MOOCs will need to exploit the latest breakthroughs in data processing and machine learning. Delivering a complex class to thousands of people simultaneously demands a high degree of automation.

While their ambitions are vast, ­Thrun, Koller, and Agarwal all stress that their fledgling organizations are just starting to amass information from their courses and analyze it. “We haven’t yet used the data in a systematic way,” says Thrun. It will be some time before the companies are able to turn the information they’re collecting into valuable new features for professors and students. To see the cutting edge in computerized teaching today, you have to look elsewhere—in particular, to a small group of academic testing and tutoring outfits that are hard at work translating pedagogical theories into software code.

One of the foremost thinkers in this field is a soft-spoken New Yorker named David Kuntz. In 1994, after earning his master’s degree in philosophy and working as an epistemologist, or knowledge theorist, for the Law School Admission Council (the organization that administers the LSAT examinations), Kuntz joined the Educational Testing Service, which runs the SAT college-admission tests. ETS was eager to use the burgeoning power of computers to design more precise exams and grade them more efficiently. It set Kuntz and other philosophers to work on a very big question: how do you use software to measure meaning, promote learning, and evaluate understanding? The question became even more pressing when the World Wide Web opened the Internet to the masses. Interest in “e-learning” surged, and the effort to develop sophisticated teaching and testing software combined with the effort to design compelling educational websites.

Three years ago, Kuntz joined a small Manhattan startup called Knewton as its head of research. The company specializes in the budding discipline of adaptive learning. Like other trailblazers in instructional software, including the University of California-Irvine spinoff ALEKS, Carnegie Mellon’s Open Learning Initiative, and the much celebrated Khan Academy, it is developing online tutoring systems that can adapt to the needs and learning styles of individual students as they proceed through a course of instruction. Such programs, says Kuntz, “get better as more data is collected.” Software for, say, teaching algebra can be written to reflect alternative theories of learning, and then, as many students proceed through the program, the theories can be tested and refined and the software improved. The bigger the data sets, the more adept the systems become at providing each student with the right information in the right form at the right moment.

Knewton has introduced a remedial math course for incoming college students, and its technology is being incorporated into tutoring programs offered by the textbook giant Pearson. But Kuntz believes that we’re only just beginning to see the potential of educational software. Through the intensive use of data analysis and machine learning techniques, he predicts, the programs will advance through several “tiers of adaptivity,” each offering greater personalization through more advanced automation. In the initial tier, which is already largely in place, the sequence of steps a student takes through a course depends on that student’s choices and responses. Answers to a set of questions may, for example, trigger further instruction in a concept that has yet to be mastered—or propel the student forward by introducing material on a new topic. “Each student,” explains Kuntz, “takes a different path.” In the next tier, which Knewton plans to reach soon, the mode in which material is presented adapts automatically to each student. Although the link between media and learning remains controversial, many educators believe that different students learn in different ways. Some learn best by reading text, others by watching a demonstration, others by playing a game, and still others by engaging in a dialogue. A student’s ideal mode may change, moreover, at each stage in a course—or even at different times during the day. A video lecture may be best for one lesson, while a written exercise may be best for the next. By monitoring how students interact with the teaching system itself—when they speed up, when they slow down, where they click—a computer can learn to anticipate their needs and deliver material in whatever medium promises to maximize their comprehension and retention.

Looking toward the future, Kuntz says that computers will ultimately be able to tailor an entire “learning environment” to fit each student. Elements of the program’s interface, for example, will change as the computer senses the student’s optimum style of learning.

Big Data on Campus

The advances in tutoring programs promise to help many college, high-school, and even elementary students master basic concepts. One-on-one instruction has long been known to provide substantial educational benefits, but its high cost has constrained its use, particularly in public schools. It’s likely that if computers are used in place of teachers, many more students will be able to enjoy the benefits of tutoring. According to one recent study of undergraduates taking statistics courses at public universities, the latest of the online tutoring systems seem to produce roughly the same results as face-to-
face instruction.

While MOOCs are incorporating adaptive learning routines into their software, their ambitions for data mining go well beyond tutoring. Thrun says that we’ve only seen “the tip of the iceberg.” What particularly excites him and other computer scientists about free online classes is that thanks to their unprecedented scale, they can generate the immense quantities of data required for effective machine learning. Koller says that Coursera has set up its system with intensive data collection and analysis in mind. Every variable in a course is tracked. When a student pauses a video or increases its playback speed, that choice is captured in the Coursera database. The same thing happens when a student answers a quiz question, revises an assignment, or comments in a forum. Every action, no matter how inconsequential it may seem, becomes grist for the statistical mill.

Scholars who are skeptical of MOOCs warn that the essence of a college education lies in the subtle interplay between students and teachers that cannot be simulated by machines, no matter how sophisticated the programming.

Assembling information on student behavior at such a minute level of detail, says Koller, “opens new avenues for understanding learning.” Previously hidden patterns in the way students navigate and master complex subject matter can be brought to light.

The number-crunching also promises to benefit teachers and students directly, she adds. Professors will receive regular reports on what’s working in their ­classes and what’s not. And by pinpointing “the most predictive factors for success,” MOOC software will eventually be able to guide each student onto “the right trajectory.” Koller says she hopes that Lake Wobegon, the mythical town in which “all students are above average,” will “come to life.”

MIT and Harvard are designing edX to be as much a tool for educational research as a digital teaching platform, Anant Agarwal says. Scholars are already beginning to use data from the system to test hypotheses about how people learn, and as the portfolio of courses grows, the opportunities for research will proliferate. Beyond generating pedagogical insights, Agarwal foresees many other practical applications for the edX data bank. Machine learning may, for instance, pave the way for an automated system to detect cheating in online classes, a challenge that is becoming more pressing as universities consider granting certificates or even credits to students who complete MOOCs.

With a data explosion seemingly imminent, it’s hard not to get caught up in the enthusiasm of the MOOC architects. Even though their work centers on computers, their goals are deeply humanistic. They’re looking to use machine learning to foster student learning, to deploy artificial intelligence in the service of human intelligence. But the enthusiasm should be tempered by skepticism. The benefits of machine learning in education remain largely theoretical. And even if AI techniques generate genuine advances in pedagogy, those breakthroughs may have limited application. It’s one thing for programmers to automate courses of instruction when a body of knowledge can be defined explicitly and a student’s progress measured precisely. It’s a very different thing to try to replicate on a computer screen the intricate and sometimes ineffable experiences of teaching and learning that take place on a 
college campus.

The promoters of MOOCs have a “fairly naïve perception of what the analy­sis of large data sets allows,” says Timothy Burke, a history professor at Swarthmore College. He contends that distance education has historically fallen short of expectations not for technical reasons but, rather, because of “deep philosophical problems” with the model. He grants that online education may provide efficient training in computer programming and other fields characterized by well-established procedures that can be codified in software. But he argues that the essence of a college education lies in the subtle interplay between students and teachers that cannot be simulated by machines, no matter how sophisticated the programming.

Alan Jacobs, a professor of English at Wheaton College in Illinois, raises similar concerns. In an e-mail to me, he observed that the work of college students “can be affected in dramatic ways by their reflection on the rhetorical situations they encounter in the classroom, in real-time synchronous encounters with other people.” The full richness of such conversations can’t be replicated in Internet forums, he argued, “unless the people writing online have a skilled novelist’s ability to represent complex modes of thought and experience in prose.” A computer screen will never be more than a shadow of a good college classroom. Like Burke, Jacobs worries that the view of education reflected in MOOCs has been skewed toward that of the computer scientists developing the platforms.

Flipping the Classroom

The designers and promoters of MOOCs don’t suggest that computers will make classrooms obsolete. But they do argue that online instruction will change the nature of teaching on campus, making it more engaging and efficient. The traditional model of instruction, where students go to class to listen to lectures and then head off on their own to complete assignments, will be inverted. Students will listen to lectures and review other explanatory material alone on their computers (as some middle-school and high-school students already do with Khan Academy videos), and then they’ll gather in classrooms to explore the subject matter more deeply—through discussions with professors, say, or through lab exercises. In theory, this “flipped classroom” will allocate teaching time more rationally, enriching the experience of both professor and student.

Here, too, there are doubts. One cause for concern is the high dropout rate that has plagued the early MOOCs. Of the 160,000 people who enrolled in ­Norvig and Thrun’s AI class, only about 14 percent ended up completing it. Of the 155,000 students who signed up for an MIT course on electronic circuits earlier this year, only 23,000 bothered to finish the first problem set. About 7,000, or 5 percent, passed the course. Shepherding thousands of students through a college class is a remarkable achievement by any measure—typically only about 175 MIT students finish the circuits course each year—but the dropout rate highlights the difficulty of keeping online students attentive and motivated. Norvig acknowledges that the initial enrollees in MOOCs have been an especially self-motivated group. The real test, particularly for on-campus use of online instruction, will come when a broader and more typical cohort takes the classes. MOOCs will have to inspire a wide variety of students and retain their interest as they sit in front of their computers through weeks of study.

The greatest fear among the critics of MOOCs is that colleges will rush to incorporate online instruction into traditional classes without carefully evaluating the possible drawbacks. Last fall, shortly before he cofounded Coursera, Andrew Ng adapted his Stanford course on machine learning so that online students could participate, and thousands enrolled. But at least one on-campus student found the class wanting. Writing on his blog, computer science major Ben Rudolph complained that the “academic rigor” fell short of Stanford’s standards. He felt that the computerized assignments, by providing automated, immediate hints and guidance, failed to encourage “critical thinking.” He also reported a sense of isolation. He “met barely anyone in [the] class,” he said, because “everything was done alone in my room.” Ng has staunchly defended the format of the class, but the fact is that no one really knows how an increasing stress on computerized instruction will alter the dynamics of college life.

The leaders of the MOOC movement acknowledge the challenges they face. Perfecting the model, says Agarwal, will require “sophisticated inventions” in many areas, from grading essays to granting credentials. This will only get harder as the online courses expand further into the open-ended, exploratory realms of the liberal arts, where knowledge is rarely easy to codify and the success of a class can hinge on a professor’s ability to guide students toward unexpected insights. The outcome of this year’s crop of MOOCs should tell us a lot more about the value of the classes and the role they’ll ultimately play in the educational system.

At least as daunting as the technical challenges will be the existential questions that online instruction raises for universities. Whether massive open courses live up to their hype or not, they will force college administrators and professors to reconsider many of their assumptions about the form and meaning of teaching. For better or worse, the Net’s disruptive forces have arrived at the gates of academia.

 


Nicholas Carr is the author of The Shallows: What the Internet Is Doing to Our Brains. His last article for MIT Technology Review was “The Library of Utopia.”

147 comments. Share your thoughts »

Credits: Brian Cronin

Tagged: Computing, Web, e-learning

Reprints and Permissions | Send feedback to the editor

From the Archives

Close

Introducing MIT Technology Review Insider.

Already a Magazine subscriber?

You're automatically an Insider. It's easy to activate or upgrade your account.

Activate Your Account

Become an Insider

It's the new way to subscribe. Get even more of the tech news, research, and discoveries you crave.

Sign Up

Learn More

Find out why MIT Technology Review Insider is for you and explore your options.

Show Me