Edly Spotlight Episode 4: Past, Present & Future of Online Education with Anant Agarwal

Yeah, frankly, I think education is one field that can stand to benefit the most from AI. I mean, every single child or student or learner in the world can have a super high quality teacher that is working with them throughout life. I mean, that is a game changer. When we launched EdX in 2011, I still remember in an opening speech at the announcement conference with Harvard and MIT, you know, I said, 'Look, I'd like to reach 8 billion people around the world.'

Oh thank you. Thank you so much for having me on your academic series and delighted to join you to have a conversation about cool stuff. So, speaking of cool stuff, can we talk about AI? Oh, cool stuff. I guess today if you want to talk about cool stuff, it's about AI. But even there, I guess AI is old stuff. I think for some reason everybody prefers to talk about Gen AI, but you ask what Gen AI is and I don't think people know what Gen AI is.

So I used to work in AI when it was called. I actually was involved with AI and so on when it was called AI, and then a lot of symbolic approaches to AI when AI was looking to mirror how humans thought, and then machine learning came about which is much later, which became the cool new kid on the block where they said, 'Look, let's not try to mimic how the human brain works, let's use statistical techniques to do or create intelligence.' And that was machine learning. And now the focus seems to have gone back to neural nets and deep learning which actually mimic more in how the brain works than machine learning. For me, I feel like we have been wrong in machine learning for like 35 years and we could have if we had just stayed course? I don't think so. I think all of these developments were critical. There's a lot that we learned in machine learning and frankly the neural net-based approaches also began to be called machine learning. So it's not like deep learning, which is really learning with neural nets, was different. It simply became a way of machine learning. And so anything that a school was called machine learning, and then after that everything that a school became to be called AI, and then everything that a school became to be called Gen AI. But really Gen AI itself is a form of machine learning which itself is a form of AI. So that's a question that I like to ask people, friends and family as well.

Well you know I'll give you two answers. One is that there isn't much you can do today that doesn't make use of AI. And so when you go and use any search platform, the search platforms now give you a summary of what you're looking for before they show you links. And so in that sense, search has changed completely and those search techniques use Gen AI to summarize answers to your questions. So in that sense, you're using Gen AI every day. But the second way I would answer it is that you know I often go to specific Gen AI tools like Chat GPT and so on and I use that in some of my teaching. I'm actually working on a course on how to use AI in education and learning. And so there I try to use Chat GPT as a teacher would to try to figure out how we can make education better, how teachers can teach better, how students can learn better. And so that's one way in which I go in, and as an example I might say things like, 'Here's a topic, give me a highly motivating example for this knowledge transfer topic that I can show students.' So that's an example of how I might have used Gen AI in the recent past.

Well, you know, certainly it will be published on edX and try to figure out, I mean it's still in its early infancy. I'm still thinking about it, dreaming about it. So there's a long way to go for the course. But part of what I'm doing now is trying to get a sense of what are the ways in which people can learn and teach better using AI and how AI can afford better tools for teachers. Those are the kind of things that I'm doing at the moment. So when you dream about that course, are you dreaming that you are teaching this course online or onsite? A bit of both planned? No, it's pretty much online. Yeah, you know, what I deeply believe in these days is that AI will be only as good as the data that you provide, both in real time and past data history for AI to learn and then personalized teaching to you, and digital learning or online learning is a way where you have very rich data collection happening all the time. So on edX, for example, if you're taking a course on the edX platform, we are capturing every single mouse click, every single activity that a learner is doing. And that huge event stream is captured and that event stream can be used to really train intelligent forms of interventions and teaching that can enable adaptation or personalization to people. So that data stream is really important. I've also taught for 36 years at MIT, three decades of classroom teaching, and there it's very hard to get data while you're teaching. You get some data: you can look at the students and see some students going 'Huh, what's that?' So you know based on their expressions that they are not understanding things. You get some real-time data, but a large majority of students are usually asleep in the class or doing social or other stuff, so you're not getting much insight into how they're doing in the course. As another example, when you give homework, you have no idea how they're doing because they're doing the homework in their dorm rooms on paper and pencil, so you have no idea where did people have difficulty, you have no idea that people need help. So unless someone physically reaches out to you or unless you do a survey or unless you give an exam, you don't get much feedback. So it's very coarse-grained feedback. So using AI in a physical setting is certainly much harder in my mind than in an online setting where you can get real-time access to extremely rich data that can be used to personalize AI for you.

Look, I don't know about legacy and so on. Those are very grandiose terms and I don't think any of us on this earth is particularly important. But you know, I've seen many examples of how the process of building and teaching has had impacts down the road. On edX, for instance, as we gather a lot of data even well before AI, as we gather this data, there are a lot of people doing research on the data. We do A/B testing on the platform all the time, where many of the professors are doing A/B testing to see what works and what doesn't. And so in that sense, edX is a large experimental platform. When we launched edX, 14 years ago, I talked about how I would like to see edX and the Open edX ecosystem as a particle accelerator for learning. Just like the CERN accelerator was accelerating physics and understanding of nature, I would hope that edX would enable and accelerate learning. One of the key tenets of that is gathering real-time detailed data and sharing that data with people doing A/B testing on the platform so that researchers can use the data to understand how learning works and how we can teach and learn much better than we do. So in that sense, my hope is that through the work that the entire Open edX community is doing, through the work that edX is doing, through all the instructors and learners, we are learning more about learning and simply learning how to teach and learn better.

You know, that's a great question. There are cases where I can say that I'm in more contact with students than I was in a classroom, but I think the converse is true where students feel that they're in much closer touch with the teacher online than in person. I'll give you an example. I teach three electronics and circuits courses out of MIT on edX. These were the first courses on edX. And this course on edX drew 155,000 students in its very first offering in early 2012. And that number, 155,000, is greater than the total number of alumni out of MIT in its entire 150-year history. So we certainly had the scale. Now what some of the students were saying on the discussion forums was, 'My goodness, it seems like Professor Agarwal is sitting next to me and teaching me one-on-one.' The style of the videos that I created were what I call Khan-style videos, inspired by Sal Khan's videos, where it's all handwritten on a tablet. So it's almost as if a student is sitting next to me and I have a piece of paper and I'm writing on it, and the student is watching the same piece of paper and we're learning together. It's like how one-on-one tutoring works. Many students responded on the discussion forums saying they felt a very personal connection with the professor, almost as if the professor has been tutoring them one-on-one. And they can pause the videos, rewind, answer questions at their own pace, get multiple chances at questions. So everything runs at their own pace, which is very personalized. In a large classroom, it's one-size-fits-all where some students are asleep, some are bored because I'm going too slow. So it's really not the right speed for anybody. But with the one-on-one style of teaching with tablet capture, it tends to be very personalized to the student's pace. So in that sense, I think there's a closer connection between student and teacher than you might see in a 100- or 200- or 300-person classroom as I might teach at MIT.

No, I think that with social networks, everybody feels deeply connected to the idol that they can watch on a YouTube video or a TikTok video. If I listen to C.I.A., you know, and I watch him on my smartphone, it's almost like there's a personal connection as you watch these videos. So what is wrong with having education leverage the same social technologies to create deeper connections with students and learners? I think it's only increased today because students and learners of the younger generation are very used to communicating and listening to YouTube or TikTok videos. And frankly, my complaint is that they're spending much less time with each other and they're watching each other also on a screen. I've seen examples of students sitting next to each other and talking to each other by looking at their screens. And that part is sad. We're losing that connection. But in a world where people are increasingly interacting with their devices, what's wrong with education leveraging the same technology and creating connections with students in a way that students are so used to today, rather than dragging them to a classroom with 300 students at 8 in the morning when none of them want to be there.

Look, there are few things in life where the answer is one extreme or the other. I have a lot of gray hair.

And, you know, I've seen in life sometimes we may make rhetorical arguments at the opposite bookends to make a point. But generally, virtually all answers are somewhere in the middle. Similarly with education, there will be a lot of online learning, a lot of AI tutoring and AI content production, but at the same time we need to get the human touch in there as well. Just as with social, I think going to one bookend where humans stop interacting face-to-face but focus on a screen is not good. I think humans need to talk to each other, sit together in a group, attend a conference, or do a project together. The bandwidth of interaction is huge. On a small screen, I might look at the face of one or two people, and that's it. But in a small group, I can have seven, eight, nine, ten people sitting together, watching every interaction, hand movements, gestures, excitement, tone of voice—it's a whole environment. So the bandwidth of communication is incredible. It's very hard to replicate that with a screen or even with virtual reality. So I think the physical connection is important. Either extreme—only physical or only online—is not a good thing, whether for education or social or anything. I think something in the middle is good, where you spend a significant amount of time virtually but also some amount in real, physical connection with other human beings.

I think people sitting at a table interacting by phone is pretty extreme. I can imagine much worse extremes. At least here, people are sitting next to each other, and some social pressure brings them together. A completely worse interaction would be where everybody retreats into their cave and interacts with others only through a device, with the whole window into the world being a small device. That is a much worse situation. So my hope is that, particularly with the younger generation, we create much more physical interactions. The good thing is sports. I know people spend more time with video games than real sports, but I hope activities like sports and projects will bring people together to engage physically.