MIT's Anant Agarwal on AI in Education - Ep. 197

0:10Hello, and welcome to the NVIDIA AI podcast. I'm your host, Noah Kravitz. Our guest today is Anant Agarwal. Anant is the founder of edX, a massive open online course provider that has been working at the forefront of digital education for more than a decade now. Anant is not only an education visionary, but he is also an MIT professor and former director of MIT's Computer Science and Artificial Intelligence Laboratory. He now serves as chief platform officer at 2U, who acquired edX in 2021. Anant is here to talk about edX's new chat GPT plugins, which are designed to enhance the learning experience, making it more interactive, personalized, and engaging than ever before.

0:52More broadly, we'll also be exploring the role of Gen.AI in education and what education can and should look like as everything from the job market to the purpose of colleges and universities changes in the years to come. I'm really excited about this one, so let's get right into it. Anant, welcome, and thank you so much for joining the NVIDIA AI podcast. Thank you so much for having me, Noah. Really excited about this conversation. It's near and dear to my heart. As we were talking about before and the intro alluded to, I can only imagine given your background. So hopefully we'll have time to dig in a little bit to the things you were doing before edX and to you later in the episode.

1:30But I have a little bit of a background in education myself earlier in my career, so I want to dig in. Let's get right into it. Maybe we can start by you telling the audience a little bit about edX and to you now as well. And then we can get into the new AI products that were announced recently. Fantastic. EdX is an online learning platform. The organization was founded by Harvard and MIT in 2012. And as a platform company, edX brought together learners from all over the world who came to edX to learn from some of the top institutions in the world, like MIT and Harvard, of course, Berkeley, Stanford, and companies as well like IBM, Google, Linux Foundation, and others.

2:16It's a platform company in that we don't have any professors at such at edX. I'm probably one of the few professors here. but really we connect learners to great content. We have over 4 ,000 courses and programs from some of the top institutions. And we have about 70 million learner prospects who come to our platform looking to learn. And the learning experiences can range from individual courses that might be free to professional certificate programs, to executive education, bootcamps, or degrees. Right. I was familiar with edX before, but in leading up to this, I logged in and was checking out a data science course, actually.

3:00And so there are free options. There are paid options, some of which lead to certifications and other accreditations that you can use in your career in job hunting. What Will You? One of the things that struck me in researching for the conversation is how many adults, I think it was in the US, but I'm sure it's globally, have skills, even have some college credits that are applicable in today's job market, certainly, but haven't necessarily finished a degree. And so I'm wondering, kind of without getting out over our skis a little bit, I'm wondering how you've seen in the time that edX has been out, the role of kind of developing skills versus, you know, getting a degree, how employers and the job market kind of more broadly are looking at these things, and what that says to you about the changing face of post high school education in particular.

4:00You know, there are over 40 million people in the United States that have some college, but no degree. And for them, going back to school or college and getting a full degree in the usual way is near unthinkable. Many of them are doing gig jobs, driving Lyft or Uber. And so no way can they go and get a full degree on campus. And so these may be the people we call non-traditional learners that need to upskill. And there's a huge demand for upskilling as well The OECD, for example, has projected that over a billion jobs would be completely transformed within the next decade. And frankly, I think that it's an underestimate.

4:44I think it's even bigger. And so we have an entire planet that needs to be upskilled. You know, tens of millions of people in the United States and elsewhere don't have a degree. And so what edX brings to the table here are a variety of courses and options where people can upskill. let's say you mentioned data science. We have professional certificates in data science from IBM and Harvard. We have micro-masters in data science from MIT. And we have complete degrees in data science from Berkeley, for example, and UT Austin. And we have bootcamps in data science as well from places like Southern Methodist University and others.

5:24So we provide a range of options all the way from courses and programs like bootcamps and professional certificates, which are credentials that you can upskill, but not get a degree. And we have degrees as well. But I think an important point that people often miss is that it's the power of and. I call it and thinking, where it's not an either or. It's not about just getting a micro-credential or a degree, it's and. With the data science example you mentioned, sticking on the topic for a second, learners can complete a bootcamp in data science from SMU. And after they complete that bootcamp, they can actually start a degree and apply for and start a degree in data science from Southern Methodist University, also fully online in partnership with edX and 2U.

6:12And the credits that they earn in the bootcamp can be applied towards a degree. So it's not an either or. I see many of these credentials as pathways to complete degrees. Should people want complete degrees? In AI, the topic of this podcast, We have a large number of micro-credentials in AI. We have a professional certificate in AI from Google. We have intro to chat GPT courses. And then we have degrees in AI. We just launched a very disruptively priced degree in artificial intelligence and machine learning from UT Austin for$10 ,000. And when we announced the degree three months ago, we had 2 ,000 expressions of interest from learners within 24 hours of launch.

6:56And today, we have over 10 ,000 learners who have expressed interest in that degree. And this is just a few months. So there's a huge demand for all the positive learning. Yeah. So as you said, we're talking about AI on this show. So let's dig in a little bit to, or a lot, I should say, to the role of AI in teaching and learning. And specifically, I mentioned at the top, there are two new plugins, OpenAI rolled out plugins to a wider audience, ChatGPT plugins in the recent weeks. And there are two edX plugins, I believe, and there's also the edX expert product on the platform. So maybe you can talk a little bit about those.

7:36Sure. So edX had a history of innovating when we launched edX out of Harvard and MIT 12 years ago. There we brought in digital technology applied to learning and helped revolutionize learning both on campus and beyond with these micro-credentials, upskilling, and also creating much more engaging learning experience that enabled flexibility for the learners. So today, no matter where you are, we can give you upskilling at a very free of a very low cost and on your own pace and your own time and completely enabled by digital learning. And so digital technology completely transformed education. AI is another really, really big thing.

8:16You know, I think of the AI revolution as big or if not bigger than, you know, the steam engine revolution or the entire wireless telecommunications and so on. And this is the internet. This is big, if not bigger. And it's going to have widespread impact on everything that we do at work or at home. And certainly learning is no exception. You know, I strongly believe that education can benefit incredibly from artificial intelligence. You know, much as anything else, when I was at MIT in 2007 and 2008, my group at MIT was experimenting with using AI and machine learning to make computer architectures more energy efficient.

8:59We're using AI to analyze how codes were running and use machine learning to change how resources were acquired in order to make computation more energy efficient. Today, we are applying AI at edX to learning. and we've released two products. One is we've released a ChatGPT plugin. The ChatGPT plugin is available in the ChatGPT plugin store. And at this point, it has been rolled out to ChatGPT plus users. So you can go to ChatGPT, become a plus user and then go to the plugin store and you can install the edX plugin. That will enable you to have a great experience with discovering new learning experiences and it can also help you with learning in a way that goes beyond what ChatGPT can do in a generic manner.

9:49So that plugin takes content from within edX courses, for example, from my circuits course out of MIT and learners can ask it, if they have the plugin installed, they can ask it circuits questions and the circuits course will answer those questions through the ChatGPT interface. So that is one product. The second product is called Expert. Expert, X-P-E-R-T, is the set of technologies that we have released on the edX platform that we will continue evolving over time. The expert technologies will enable learners to learn better. So just as one example, we've released expert summaries. So when a learner is within an edX course and people can go try it out in my circuits course on edX, if you watch a video or any other piece of content, we have the expert summary button.

10:43You click on the button and you see a brief summary. And this has been powered by ChatGPT. So we have an enterprise integration into ChatGPT and we provide a summary. And so just summary as an example can help learners quickly refresh on materials. But that's just one example. We are working on expert tutor, which will be an online learning tutor for people. We are working on an expert help center and learning assistant that will help learners learn better as well as get doubts cleared, adaptive learning. We're also working on things like better authoring for faculty to enable, imagine. So there's many, many roles that AI can play.

11:23And we're just scratching the surface in terms of what we can do. When I think about AI and digital and education, digital learning, I always go back to the Neil Stevenson book, Diamond Age. And there's something called the primer, which is essentially a hyper personalized, very active learning from everything sort of tutor that it goes beyond what we think of as, you know, education and school and in all facets of life, but kind of this hyper personalized tutor that stays with you, has a memory, learns from you. knows about your learning style, all that kind of stuff. Is that the path you see things headed on when it comes to AI enhancing the learning experience?

12:05Are there sort of paradigms in your head that you and your team, I should say, are kind of moving towards? Or is it very much right now a matter of we're just scratching the surface, like you said, and it kind of remains to be seen how this evolves? We're certainly scratching the surface. And no matter what I say, I will be woefully off. But, you know, that shouldn't prevent us from dreaming a little bit here. So one can imagine three approaches. One approach is, and these are incrementally more, what shall I say, more heavily dependent on AI. Okay. The first one is fairly straightforward tools like a summarizer or a video translator or things like a discussion forum moderator, a very specific essay grader.

12:57These are examples of things that are specific tools. So that's one. The second one is where AI augments the human teacher. And think of this as co-pilot. When you program in Python or whatever, you can access a co-pilot in many of these programming environments that help you write better code or help you write code faster. Similarly, one can imagine kind of every student potentially having an avatar kind of being a co-pilot, helping them learn, or a professor having an avatar or a co-pilot helping them teach better. So those are examples where these avatars are augmenting the existing learning experience.

13:40So for instance, if I'm learning and I have trouble with something, I can say, hey, I didn't understand that. Can you explain that to me better? Or can you help me understand that or give me a hint, for instance? Or with our edX chat GPT plugin, for example, one of the questions you can ask our plugin is, hey, quiz me on this topic. And our chat GPT plugin will quiz you on that topic. And so that's one example where it becomes It's more like it augments your own experience or augments the instruction as a co-pilot. That's two. And the third one is where the whole teaching experience is, whether it's adaptive learning or a tutor, where you literally give yourself up to AI and say, hey, I want to learn this topic.

14:26Hey, help me. And so it will just take you through the whole learning flow. And if you will learn adaptively, it might first ask you, what do you know? And it might ask you, where do you want to go? One of the tools we are building is an AI-based career navigator and career helping tool where learners can navigate and explore careers and find learning paths. So in that third model, it is where AI is providing the entire teaching experience, whether by tutoring the learner. it's almost like in India, when you read some of the major master, the old masterpieces like the Mahabharata and Ramayana, you see all the kings and princes and princesses had amazing tutors like Drona who are personalized tutors.

15:14And if you were really rich or if you were king's offspring, you would learn from a great tutor. But imagine now that everybody in the world can have a personalized tutor. And so, you know, AI in that third category can completely democratize education where everybody can have a great tutor. So I see these kind of three approaches. So one of the things that strikes me, and you say that, and the word democratize, that, and correct me if I'm wrong, but edX seems like one of the, perhaps the founding principle, one of the things that strikes me about it is how it offers, kind of democratizes access to these very high quality, big name learning experiences, MIT and Stanford and courses, IBM, whoever it is.

15:56And so somebody can get access to these courses and these materials created by professors, employees at these big name companies without necessarily having to, you know, be admitted to Stanford and pay full tuition, that kind of thing. What happens to the role of the professor in an age where, you know, if we extrapolate out a little bit, where somebody can go on to edX or use the edX chat GPT plugin and get access to these materials and then, you know, perhaps have a, an AI powered personalized tutor. And so then the, the professor themselves is kind of further removed from the act of teaching.

16:38If that makes sense, Where do you see the role of the professor, the teacher themselves going as we move forward into this age where maybe, you know, the professor has an agent who's helping to create the materials and to teach them? And I, as a learner, have an agent, an AI agent who's kind of helping navigate me. And, you know, you know where I'm headed with this. What happens to the role of the expert educator? You know, it's a great question, Noah, and it's a very personal question. I've been a professor at MIT for 35 years, so it's very personal as well. But first, let me address your first point, which is about democratization.

17:17Through the history of mankind, great technologies have really democratized access in various ways. It's always the case that initially these technologies have been accessed by the people who already have it all, the rich and educated. But over time, the technology gets available to everybody. It's like cell phones. Initially, the big bulky cell phones were expensive and they were bolted to your car. And so you don't own the car. But now, as I travel all over the world, in India, for example, I've seen young kids driving a bullock cart with a cell phone glued to their ear. Yes. And so everybody has it.

17:59And so great technologies have democratized access, certainly with education, digital technologies like with edX and so on are democratizing education where no matter where you are in the world, we have 48 million people learning and taking free courses on the edX platform. And so no matter where you are, you might be a kid in Kigali or in Bangalore or whatever, and you can be learning from the top professors and engineers, whether it's at Google or IBM or MIT or IIT Bombay. So I completely agree with the democratization point. So to your question about what will this do to professors, as the famed Yogi Berra said, this is like deja vu all over again.

18:51When we started edX in late 2011, 2012, I was asked the same question almost every day, saying, hey, Anand, what will this do to your job when we have all this content and courses online with digital technology? What does this do to the teacher when people can access great recorded videos and computation can grade essays in the cloud and all of that stuff? What is it going to do to the teacher? And so in some sense, AI is just more of the same here. To me, I see AI as making learning better for the learners in terms of access and outcomes. At the same time, I also see AI making authoring and creating courses and teaching better for professors and course authors.

19:39So I see AI as a rising tide that will up the game of learners as well as professors, certainly in the short to medium term. Where for learners, much like when calculators came out, it really helped everybody do computations faster. And similarly, when AI came out, as AI comes out now, as AI is coming out, I can see it helped learners learn better and faster. Adaptive learning, where we can create learning sequences that are not one size fits all, where learners can learn much more efficiently. If I'm faster, I can be shown more advanced concepts. And if not, I can be given more background materials.

20:23And so we are developing that technology with something called BlockStore within the Open edX platform. So I think adaptive learning is another is one big example of what I can do. Similarly, for authoring courses, frankly, as an author and as a professor, one of the most time consuming things is creating a great problem set. And so I can imagine looking at a piece of content or pointing AI to my book chapter in my circuits of electronics textbook and say, hey, you know, cons me or create for me a problem with kind of those features. And it can create 10 different problem types, you know, for me.

21:01We've begun to do that with ChatGPT with our plugin already. And so with structures like that, we'll be able to help professors create problems very quickly. And we also have to create mechanisms where instructors can edit the problems to make sure that it is correct and, you know, properly provides what we want to do. So I see that as a rising tide that will up the game of learners in learning and professors in teaching. In the very long term, it's hard to say where the world is going to go. I can see all kinds of scenarios. Some scenarios are exhilarating and some scenarios are petrifying. And I expect the answer to be somewhere in the middle.

21:42To that end, not to the petrifying extreme, but what are some of the potential pitfalls or warning signs that you're aware of and edX is working to avoid as you roll out the plugins and roll out expert and think about all the things you were just talking about, about AI helping both create the materials and enhance the learning experience? Noah, the timing for this question is perfect. I do what I can. Just yesterday, we rolled out our plugin. The plugin got rolled out to the ChatGPT plugin store in the inaugural launch of the plugin applications on ChatGPT about a week or so ago. Right. And so as part of that, it was very clear that as I've talked to professors and talked to students and technologists all over the world, there's always concerns about, hey, what happens to data privacy?

22:40What happens to learner data? What happens to our content? This content is IP that may belong to a professor in a university. What happens to this content as we train a generative AI model? Does the data go out into the wild? What's going to happen to all of this stuff? And so we thought through all of that pretty hard. And we came up with an initial set of what we call core principles for responsible use of AI. So it is a core principle for responsible use of AI. And we shared that with the partners to show them that, look, this is a great experimental age. We want us to experiment together, but we can put guardrails in place.

23:21So for example, one of the core principles is that, look, whatever we do must benefit learners and or faculty and institutions. That is one. Second example, and very critical, is guardrails, where as we work with the ChatGPT plugin or as we work with the enterprise APIs for our content, the guardrails should ensure that content, partner content, partner IP, which is in the courses and so on, does not get incorporated into public AI training sets. That IP should stay where it is and should not get released into the wild. Another example of one of these core principles is that we need to be careful about accuracy and bias.

24:07We need to take steps to ensure that we provide clean data inputs to avoid bias. And similarly, in all of the work that we do, we're including user mechanisms for user feedback. When the user interacts with AI, with expert, in each of these interactions, We provide them a mechanism for feedback where they can tell us how it's working. Another one is innovation and transparency, where we want to be continuing to innovate, but also provide transparency to the learners that says, look, this is being answered by AI. So their expectations can be set. And then finally, as with all that we do, we are open sourcing our learning and teaching technologies as open source to help the world.

24:49And so these are just some examples of core principles that we've come up with for responsible use of AI. And are the faculty, the institutions who you've talked to, are they receptive? They're extremely receptive and they were very appreciative that, you know, although we may not have the perfect answers, that we've been thoughtful about it, we've been responsible about it. And thinking about it in a way that balances innovation and speed with responsible use. You know, many technologists have said, let's move fast and there'll be breakage. You know, I think what we are saying is that let's move fast, but let's create guardrails so that we can be very thoughtful about how we use these technologies.

25:32I'm speaking with Anand Agrawal. Anand is the founder of edX, an open online course provider that's been at the forefront, really, of the digital learning revolution for more than a decade now. edX is now part of 2U, and we've been talking about their work broadly, but specifically about the role of Gen.AI in teaching and learning and the products we were just talking about, the ChatGPT plugin that rolled out recently, as well as the expert feature on edX's platform. Anant, I want to briefly, and this is a whole separate podcast that maybe we can have in the future, ask you about your time prior to edX when you were the director of the MIT computer science and AI lab.

26:17And I don't know the best way to ask you to tell the audience about your experiences there in, you know, five minutes or less. You know, you referenced the Yogi Berra quote, excuse me, about deja vu all over again. And as you were talking about the IP considerations, it reminded me of years ago, I was working at a small college in Brooklyn, New York, helping them design a distance learning program. And the professors had the exact same questions about the IP, the time they'd put into the course materials, and where does that all go, who owns it, all these kinds of things. So I'm wondering, maybe a good question to ask you is, as we sit here talking about Gen.AI and the future of how this is going to impact everything, education, jobs, all of this, do you have feelings of deja vu all over again from your time running the lab at MIT?

27:08And if so, are there particular things you think about as you're working on the future of education? I think the thing that comes most closely to mind, having been a professor and researcher at MIT for 30, 35 years, and certainly as the director of the lab in 2011, that as educators and researchers, It is incumbent upon us to experiment and create the latest technologies to advance humankind. And at the same time, for us to do it in a way that balances the ethics and other considerations to make sure that these technologies are used thoughtfully and carefully. So we need to experiment hard on the one hand, and these technologies need to benefit humankind.

27:59MIT's mission statement talks about how we need to impact the world and students all over the world and create new knowledge and so on. And similarly, our research and teaching should create new knowledge, should help the world. So benefiting the world is really important. And at the same time, we need to do it in a way that is very thoughtful. And we also need to be thinking about the ethics of it. So for instance, the AI master's degree that you launched in partnership with UT Austin, that is going to have a significant artificial intelligence, of course, technology and machine learning and so on, but also a significant ethics component.

28:37And so I think the ethics is one big part of it. I think the second thing that I found that is very interesting is that, as I've been talking to many of these companies, whether it's OpenAI or Google with their Bard work, what has been amazing is that Ultimately, our students are our ambassadors. And I was talking to a key technologist at Google. He told me that he got his degree at MIT in computer science and these technologies while he was at MIT in the late 80s. And so many of the folks working in the area, similar to OpenAI, one of the leaders we talked to has taught courses on edX before.

29:16So at the end of the day, what I've learned over the years is that one of our biggest products, if you will, is the students we educate. And that's how we transmit technology and knowledge around the world. And I think that may well be our single most valuable resource. Some of the best AI was being done at MIT Computer Science and AI and many universities and research labs around the world. My colleagues working on NLP and speech processing and a lot of the principles of vector computing and so on that formed the foundations of a lot of the chips today that Power AI came out of the work that was done at MIT, Stanford and Berkeley.