One Friday morning last April, Bethany LaPenta, a sophomore electrical-engineering and computer science major, was in her room in Simmons, sitting at her computer and eating a bowl of Apple Jacks. She was in her pajamas, and electronic dance music was streaming from the computer’s speakers. But she wasn’t checking her e-mail or updating her Facebook profile. She was taking an exam for her Circuits and Electronics course, 6.002.
Circuits and Electronics was the first course offered through MITx, an ambitious project that aims to reproduce the MIT educational experience online for a worldwide audience, with video lectures, virtual lab experiments, and automatically graded homework assignments and exams. But the online students who began taking the course in March were about a month behind 20 MIT students—including LaPenta—who were test-driving the system on campus.
Advance troubleshooting was essential, as the MITx team was venturing into uncharted waters. “We had no idea what would happen,” says Anant Agarwal, a professor of electrical engineering and computer science (EECS) who teaches 6.002. “We would have been embarrassed if only 500 people signed up for it. But it would be a great success if 10,000 did.” When the course, dubbed 6.002x, went live in March 2012, more than 154,000 people enrolled. In the end, 7,157 online students would pass 6.002x—as many as might take 6.002 in 40 years at MIT.
In May, the already massive experiment in online learning grew larger still when MIT and Harvard announced the formation of edX, a joint initiative—backed by $30 million from each school—that would offer online courses not only from its founding partners but from other universities as well. The technological platform built for MITx would serve as the foundation of the edX platform.
The benefits of MITx to a worldwide audience were obvious. “MITx is animated by a sense of obligation to maximize human potential,” wrote the Chronicle of Higher Education’s Kevin Carey. “It is simply untenable to claim global leadership in educating a planet of seven billion people when you hoard your educational offerings for a few thousand fortunates living together on a small patch of land.”
The benefits to MIT, on the other hand, were more obscure. Administrators talked about the possibility of charging only a modest fee to MITx students who wanted certificates of mastery; the initial course was offered free of charge, and at press time edX was in the process of being incorporated as a nonprofit. MIT seemed bent on giving away a product—an MIT education—for which it charges about $200,000. “Will edX Put Harvard and MIT Out of Business?” asked a headline on Forbes magazine’s website.
Agarwal, however, believes that edX should improve the educational experience in a way that benefits MIT students no less than those who take courses online. “We really haven’t applied technology—computing technology, Internet technology—to education,” he says. “With online learning, we can truly reinvent education.” He says the program will actually enhance the value of an on-campus MIT education, letting professors and teaching assistants work more directly with students by liberating them from the time-consuming obligations to lecture and grade. Undergraduate courses could adopt what Agarwal calls the “apprenticeship model,” in which students pursue semester-long research projects under faculty guidance. “People on campus get a whole different experience,” he says. “You can create a much better, augmented experience than a pure online experience.” And only on-campus students can receive MIT degrees.
EdX could also expand the pool of applicants to MIT, Agarwal argues. Already, he says, it’s providing an object lesson in the rigor of an MIT education. On the online bulletin board dedicated to 6.002x, “I’ve seen discussions that say, ‘Wow, this stuff is hard,’” he says. “People talk about getting a new respect for MIT students.”
The OCW Precedent
MITx grew out of a series of meetings over a span of five years, in which Rafael Reif, then the Institute’s provost, consulted with faculty, administrators, and student representatives on how technology could improve the quality of on-campus education. The program launched in December 2011, and Reif formally appointed Agarwal its director in March 2012; in May, Agarwal stepped down as director of CSAIL to become president of edX. This fall, MITx (the name now refers to MIT courses offered through edX) is hosting two new online classes in addition to 6.002x: Introduction to Solid State Chemistry and Introduction to Computer Science and Programming. Harvard and the University of California, Berkeley, will each offer two courses on edX as well. From there, edX course offerings will expand according to professors’ interest and students’ demand.
But in many ways, MITx builds on the Institute’s decade-old OpenCourseWare (OCW) initiative, which put MIT course materials such as syllabi and lecture notes—and ultimately, in some cases, lecture videos—online. When it debuted, OCW too prompted speculation about whether it was in MIT’s interest.
“A lot of people raised similar questions,” says Agarwal. “But it was very good for MIT in many respects. One is that MIT came to be known as a great university of the world.” More specifically, he says, “it helped the MIT brand. It brought a lot of goodwill to MIT, and recruiting became a lot easier.”
Indeed, in a 2010 survey, 57 percent of MIT undergrads reported that they had been aware of OCW before applying to MIT, and of those, 31 percent said that OCW had a “significant” or “very significant” influence on their decision to attend. Among graduate students, the numbers were 45 and 30 percent, respectively. A separate faculty survey found that 34 percent of faculty who had posted material on OCW felt the site had improved their professional standing, and 58 percent of all respondents felt that OCW had enhanced their departments’ reputations. “Think of edX and MITx as the next big step,” Agarwal says. “A lot of the same kinds of things that made OCW good for MIT continue to apply here.”
From the outset, Agarwal says, MITx was envisioned as a platform on which to conduct pedagogical experiments. Although he’s reluctant to speculate about what those experiments might look like—for him, part of the appeal of the project is its open-endedness—he offers a few examples based on his and his colleagues’ experience with 6.002x.
The organization of 6.002x mirrored that of a conventional on-campus course, with lectures and reading assignments that followed a prescribed order, weekly problem sets that tested students’ mastery of the concepts presented in the lectures and reading, and lab experiments that made those concepts vivid. But, Agarwal suggests, MITx could vary that formula—for example, by presenting students with problems first and requiring them to dig up the information necessary to solve them. With thousands of students adopting each approach simultaneously, a comparison of the two would be statistically meaningful. Students’ exam performance would indicate the relative merits of each.
Similarly, Agarwal says, in 6.002x, students were allowed to check the accuracy of their answers to homework problems; if they got the answers wrong, they could rework the problems and resubmit. Some students in the on-campus test group felt that that improved their understanding of the material; they were able to identify and correct flaws in their reasoning. “You knew exactly when you were getting stuff right and when you weren’t,” says 6.002x student David Ku ‘12. “With a physical P-set, you just keep going until you can’t figure out anything else, and you figure, ‘This is 70 percent likely that it’s going to be right,’ and you turn it in. And you may not think about it ever again unless you specifically take your P-set and then compare it to the answer set, which I don’t think most people do.”
Other students, however, reported that they would simply modify their answers randomly—multiplying or dividing by 2π, say—until they were correct. But again, the scale of the online course is so large that thousands of students could be offered unlimited checks, thousands of students could be offered few or no checks, and their exam performance could be compared.
“The platform built by [edX chief scientist] Piotr Mitros, Dave Ormsbee, and the rest of our development team allows us to give different batches of students slightly different experiences so that we can do these kinds of tests,” Agarwal says.
Of course, developing new educational technologies need not mean releasing them, free of charge, to the wider world. But doing so, Agarwal believes, will give edX a size and global reach that will help it concentrate resources, establish technical standards, and realize economies of scale that will yield much more useful results much faster.
“We’ve been trying to do various on-campus initiatives for a couple of decades,” Agarwal says. “But there’s no standardization. Everyone’s done their own thing. Thinking big can serve as a lightning rod for people to come together and focus all their effort along one vector, so that you can do something much, much bigger.”
“We’ve teamed up with Harvard, and we’re also going to be working with other universities around the world,” he adds. “So I think this creates a force-multiplier effect.”
For prospective donors, a project that promises to improve the quality of education in the developing world is also more appealing than one whose beneficiaries are students lucky enough to attend MIT and Harvard. Agarwal says that edX is already talking with several foundations and private philanthropists. “Many of them contacted us,” he says. “A not-for-profit approach to democratizing education is something of great interest to them.”
And building an international community of users has the side effect of creating an international community of developers. The technical platform is still a work in progress, but once it’s stabilized, edX plans to release the underlying code under an open-source license. “The benefit of doing that rather than doing it on your own is that you can get the community to develop and improve upon it,” Agarwal says. “I mean, this is a huge task.”
Indeed, to Agarwal’s amazement, 6.002x students didn’t even wait for the release of the edX source code before pitching in. “As people have requested improvements, many of the students themselves are giving us pieces of code and technologies that we are now making available worldwide,” he says. For instance, says Piotr Mitros, one student wrote an online textbook viewer that worked better than the MITx version on mobile devices; another wrote a program that let students download the course videos from YouTube for off-line viewing. “Several students wrote user scripts for Firefox that added substantial functionality to the courseware—from small things like a volume control for the video player to bigger things like better visualization of which problems are graded, or integration of the social components,” Mitros says.
The openness of edX paid more direct dividends for the MIT students in the experimental section of 6.002 last spring. The MITx site hosted two discussion forums dedicated to the course, one for the MIT students and another for those off campus. “The external course had a larger user base, so the online forum was a lot more responsive,” says LaPenta. “In the internal course, you’d post a question, and a few days later you’d get an answer. Whereas with the online course, you’d post a question and three or four responses would appear within 30 minutes.”
Because the external students were a month behind the on-campus students, they weren’t much help on problem sets, which had specific due dates. But LaPenta says she did rely on the forum to study for exams. “If there was a concept that I was unclear on, it turned out that five other people had the same question,” she says. “It was very helpful for exams.” Agarwal adds that students who took the time to answer questions on the forum also solidified their own understanding of the material. “You never really learn something until you have to teach it,” he says.
For the students in the experimental on-campus group, however, most of the benefits of the MITx platform had to do with the technologies themselves, not the fact that 154,000 other people were also using them. Several students cited the convenience of video lectures. “I was managing five classes this semester, so it would have been a lot harder to go to lecture, go to recitation, do all the stuff,” LaPenta says. “It was a lot more convenient for me than a restricted time that I had to go to lecture.” Besides, she adds, “I work better later at night.”
Ku says that unlike LaPenta, he typically watched the lectures at the same time every week. But “you can watch them at any speed, from .75 of the regular to 1.5,” he says. “1.5 seemed to work pretty well for me. They also modulated the voice, so it’s not higher pitched with the speed increase.”
“When you want a copy of the notes, it’s right in front of you already,” adds David Lawrence ‘14, who also took the pilot MITx class. “Everything that was written or said in the lecture is just already transcribed. The slides are provided.”
LaPenta says she appreciated the course’s virtual lab component, a Web-based simulator of a circuit kit. “The simulations online were really accurate,” she says. “You just drag the elements onto the circuit and hit ‘Simulate.’” As a student in 6.01, Introduction to Electrical Engineering and Computer Science, LaPenta had tried her hand at building physical circuits, and she didn’t miss it. “You don’t want to sit there debugging a circuit that you have right but one element is broken, which was a case that had happened in 6.01 a few times when we were prototyping our circuits,” she says. With the online simulator, “you didn’t have to deal with the technicalities of ‘Oh, is the wire in this resistor not working?’—technicalities that really slow down your learning.”
“And you don’t have to worry about anything catching on fire,” she adds, laughing.
During the launch of 6.002x, the professors were still presenting traditional lectures to those students not in the experimental section, and both they and the TAs were troubleshooting the platform itself, so the automated tools were not as liberating as they might have been. Still, Agarwal, the TAs, and Gerald Sussman, a professor of electrical engineering and one of the course’s teachers, found time to meet twice a week with the students in the experimental section. Naturally, some of the discussion focused on ways to improve MITx, which suited Ku well, given that as a second-semester senior, he had volunteered for the experimental section chiefly to participate in MITx platform development. But beyond the inevitable discussions about fixing MITx bugs, conversation seemed much more free-ranging than was typical at a recitation. Lawrence, for instance, recalls discussing the derivation of the complex impedance using the Laplace transform, something not covered in 6.002—as well as a conversation with Sussman about ham radios.
“Gerry spent some time showing me pictures of a large transmitter with an inductor the size of a small car,” Lawrence says. “It was very enjoyable talking to the professors there, often about areas of circuits that were not strictly within the course material—just interesting things.” Ku agrees. “I mostly just went to hang out with the professors,” he says.
Agarwal contends that MITx will lead to a more interdisciplinary undergraduate education as well. He points out that a number of participants in the experimental section were mechanical-engineering students who had long wanted to take the course but been unable to. “They said that in the past, the 6.002 course conflicted with a required mechanical course,” he says. “Within the department, we work rather carefully to make sure that typical courses that students take in a semester don’t conflict in terms of times, but we don’t do this across campus.”
Under the apprenticeship model, MITx could also encourage interdisciplinary study by giving students flexibility in their choice of projects. A project might, say, apply concepts drawn from electrical engineering to biological questions, or use physics to address issues in economics.
The Grading Challenge
The theory that edX will give TAs and professors more time to work directly with students assumes that grading can be automated. That may have been the case in 6.002x, where the answers to most homework and exam questions were numerical values or mathematical expressions, but the assumption won’t hold for all courses offered through the program.
How to automate grading is one of the major research questions that edX will address, but it’s by no means a new problem for MIT researchers. EECS professor Regina Barzilay works in the field of natural-language processing, which seeks algorithms for making sense of free-form texts. Barzilay’s group has developed systems that automatically deciphered an ancient alphabet and deduced the meanings of words in a video game manual by playing the game. But they’ve also proposed techniques for improving automatic essay grading. Last year a trio of researchers at the Educational Testing Service evaluated one of those techniques, using it to help score essay questions on the GRE General Test and the Test of English as a Foreign Language. They found that the agreement between its scores and those assigned by human graders was comparable to the agreement between human graders themselves. Agarwal says that he and Barzilay have begun discussing how her work could be brought to bear on edX.
Agarwal also cites the work of Rob Miller, who leads the User Interface Design Group at CSAIL. In recent years, Miller’s group has researched crowdsourcing, or breaking complex tasks into small, discrete units and farming them out over the Internet. In the classic crowdsourcing scenario, participants are paid a small amount—often just a few cents—for each task they complete. Miller’s group devises techniques for organizing tasks so that they require as little of the participants’ time as possible. The researchers also develop methods for verifying participants’ work, so that the results won’t be compromised by either incompetence or malice.
Miller also teaches a course called Elements of Software Construction—6.005—in which most homework assignments consist of writing short programs. Verifying that a program does what it’s supposed to is fairly easy to automate, but the course instructors also want to provide students with feedback on the clarity of their code. Last fall, Miller began using his group’s crowdsourcing techniques to perform code reviews. Tasks were farmed out not only to TAs, and to MIT alumni who volunteered their time, but to students in the class as well. “You want to draw on a crowd that consists of your learners—the students in the course, the students who have already taken the course—as well as, hopefully, experts in the field and teaching staff of the course,” Miller says. “The challenge is to get them to interact together in a way that provides good feedback to the students.” If edX courses continue to draw hundreds of thousands of students a semester, they will rapidly produce a huge population of potential graders.
Paying graders will probably be impractical, but Agarwal points out that MITx already provides so-called karma points to students who make regular, high-quality contributions to the discussion forums, and that high karma-point totals grant users privileged access to aspects of the edX site—say, the ability to moderate posts on the discussion forums dedicated to particular courses. Karma points could provide an incentive for crowdsourced grading.
Miller agrees. Other question-and-answer forums in the programming community, he says, provide similar ratings, which high scorers have been known to cite on their résumés. “That is evidence that they treat it like a real-world kind of grade point average—that it does mean something to them,” he says.
Ultimately, Miller says, grading “is probably going to end up being a mix of solutions, which will be partly automatic and partly human eyeballs.” For some courses, for example, an automatic grading system like Barzilay’s could make an initial pass, with verification of the results farmed out to thousands of former students.
But assuming that the grading problem can be solved and that online versions of more on-campus classes will be added, will students who have the option prefer them to the conventional versions?
Lawrence, who already had extensive experience building physical circuits and missed getting his hands dirty with actual resistors and capacitors, says that it would depend on the course content. “I would definitely take the X version of a purely theoretical course,” he says. “In fact, I think I would take any X course within MIT that was specifically tailored to include the necessary hands-on component.”
Ku and LaPenta are less circumspect. “I would,” Ku says. “Definitely,” says LaPenta. “I really enjoyed it.”
This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
The Biggest Questions: What is death?
New neuroscience is challenging our understanding of the dying process—bringing opportunities for the living.
Rogue superintelligence and merging with machines: Inside the mind of OpenAI’s chief scientist
An exclusive conversation with Ilya Sutskever on his fears for the future of AI and why they’ve made him change the focus of his life’s work.
How to fix the internet
If we want online discourse to improve, we need to move beyond the big platforms.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.