AI and Online Learning

Ramachandran S: Hello everyone. My name is Ramachandran. I'm the lead for engineering and manufacturing domains in our thought leadership team, Infosys Knowledge Institute. Today we are very privileged to have with us Professor Anant Agarwal from MIT. He's a professor of computer science and electrical engineering. He has been the head for the Center for Artificial Intelligence at MIT. Today we're going to talk to him about the application of artificial intelligence for online learning platforms or what we call MOOCs. Professor Anant is also a successful senior entrepreneur. He was founder and CEO of the popular learning platform, edX, which Harvard University and MIT founded. We are very happy to have you with this professor. Thank you for joining.

Professor Anant Agarwal: Thank you very much for having me, Ram. I'm super excited about this conversation.

Ramachandran S: Professor Anant, you have been a pioneer in the rollout of global MOOCs platforms. Also in your current role as chief platform officer of 2U. You have written a lot about blended learning or your inverted pan model for post-pandemic online education. In the corporate world we talk about hybrid work, so can you talk us through what are the recent trends in blended or hybrid learning? What are some of the recent trends in your inverted pan model? Can you talk to us about that?

Professor Anant Agarwal: Online education has been impacted by two major trends in the past three or four years. The first of course, was when we all went to remote work and learning because of COVID, suddenly all of the learning went online all around the world, and as we came out of COVID, the blended model of learning became incredibly popular where whether it was a corporation or universities began to combine both in-person learning with online learning and tried to get the most of these worlds. So for example, even within a classroom, you will have students watch a video before they come to class and then in class they spend the time with discussion. So that was a big trend post-COVID with the blended learning, and I'm pretty confident that this is here to stay. With corporate learning it has become much more online where as employees work remotely, they're not at one location anymore and so learning has become virtually all online in many places in the corporate world.

The second big trend that creates a huge opportunity and a challenge, frankly, not just for MOOCs, but all of learning, is rapid, rapid advancements in AI, particularly generative AI because of the technologies created like ChatGPT by companies like OpenAI. And Google is creating Bard, Microsoft with its technologies. So AI is suddenly come of age and everybody's using AI in virtually everything that we do, and we've been applying AI in learning as well. At edX, we've launched a tool called edX Expert. Expert is a online learning tool that can help learners interact with AI. It will help learners find the right courses when they're searching for courses. It will give them career advice and help them with careers. Inside courses, Expert will help them as a teaching assistant or as a tutor, helping them learn better. To me, this brings a lot of excitement as while MOOCs created massive access, what AI can do is create personalization and so we could achieve mass personalization, which frankly is a huge tenet of the fourth industrial revolution.

Ramachandran S: Artificial intelligence is seeing much more wider adoption in the MOOCs platforms. What are some of the recent trends you see in how artificial intelligence is used for online learning as part of MOOC's platforms? How is it integrated with MOOC's based learning?

Professor Anant Agarwal: As I think about AI and applying AI in education, at times I am super excited and at times I'm super petrified thinking about what the future might hold. I think in the end, AI will be what we make of it. It's like a knife, you can use... Shakespeare said, "There's nothing good or bad. Only think it, make it so." A knife, for example, in the doctor's hands can do real good and in the hands of a criminal can do real bad. So AI is an incredible tool and at the end of the day, it will be what we make of AI. And so my excitement is about very, very rapidly getting energized and using AI for the good in learning. I can see AI being applied in virtually every single domain. I'll give you one single example. When a learner calls into a help center and asks for some help, the help center person doesn't know all the information and can help the learner in a small way. Recent studies are coming out that when AI augments the help center personnel, average health center personnel become as good as the best help center personnel. In other words, what AI is doing is that AI can democratize the quality of healthcare support people.

As another example within our learning itself, what I'm seeing is that AI is going to make personalized training available to every single learner. In the past, having a live tutor was very expensive. If you go back to the ages in India, in the Mahabharata and the Ramayana and so on, if you were the son or daughter of a king, you might be coached by someone like Dronacharya for example, but you had to be the child of a king to get Drona to be your tutor. But for the average human being, there were no tutors.

Similarly, I think that with online learning today we have massive open online courses. We have courses on edX, like my own circuits course for example, on edX so far has educated close to a million students all over the world. And today we can achieve scale, but that personalization is still missing. What I see as one of the biggest potentials of AI is creating the personalization by student where we can still have the scale, teach tens of thousands and hundreds of thousands of students, but everybody can be getting a specialized experience. So for example, if I'm a learner and I have specific questions, today on edX you go to the discussion forum and you ask a question and other learners can pitch in and help you out. And sometimes a professor might be able to answer one of the many, many questions. With AI, a learner can get access to such assistance 7/24 at any time of the day, and it can be upleveled, escalated to faculty and teaching assistants if AI is not able to handle, let's say the first level, or if other students are not able to come in and help.

Ramachandran S: Talking about education, there is a lot of hype about large language models, about the use of ChatGPT in education. Some universities have even banned the usage of ChatGPT. What is your take on ChatGPT? How do you see it getting integrated in education?

Professor Anant Agarwal: Ram, that's an incredibly important question. Where ChatGPT and large language models really are what have caused this incredible move towards AI where everybody's talking about AI today, and large language models have been able to train huge neural nets, deep neural nets where virtually all of the world's data, and this large language model is really what is causing this amazing jump in quality of these conversational models where they can actually create new content. The knee-jerk reaction that I've seen from many schools and colleges and frankly some countries is that they're saying, okay, this is going to be banned. I think that is a big mistake. It is like saying, when I first started, I went to college, I joined IIT Madras from 1977 to 1982, and I still remember the calculators had just come out and we were switching from slide-rules to the calculator. So in IIT, some classes had actually told us we cannot use a calculator in the classroom. That was a big mistake. Everybody uses calculators today, use calculators at work, they're everywhere, and so banning calculators was very short-sighted. Instead, what you should do is create some kind of governance around the usage of calculators.

So at MIT, for example, in my circuits class, I always allow the usage of calculators. However, I say that, "Look, you can use calculators, but for certain problems I might tell you you cannot use the graphing capabilities of your calculator because I do want you to get an insight into sketching how waveforms look." And so you have to create some guardrails, but by and large allow the usage. Similarly, for AI, I'm a very, very strong advocate of learning how to use AI, just like in the classroom. So just like the calculator democratized numeracy. Before calculators in India, we all had to memorize mathematical tables, and I remember memorizing tables until about 12. I could not go past 12. But what the calculators did was it democratized numeracy. Now everybody could do quick math, and so they were a great tool.

Similarly, with AI and ChatGPT in particular, ChatGPT is democratizing literacy, particularly writing literacy, where people can now very quickly get drafts of writing and improve upon those drafts. You can be thinking critically and thinking deeply and negotiating and come up with the ideas, and then you use prompt engineering to provide a prompt to ChatGPT and it can give you a first draft of an article or an argument or a letter. And so that's a new way of using AI and a new way of writing. And we need to train people and particularly corporate executives into doing that because they can be extremely, extremely more efficient if they can use ChatGPT to create drafts of letters and things like that.

And so we've created several courses on edX in prompt engineering, and particularly for corporations we've created programs in ChatGPT for healthcare, ChatGPT for business, ChatGPT for education, where I believe every field needs to learn how to use ChatGPT. At the same time, we have to be careful of how to use ChatGPT and create guardrails. And so what we have done is edX has created a set of guidelines for our own principled use and thoughtful use of AI. So some of these guardrails say that while we use AI, we will use AI with these principles. One of these principles is that we will use AI for good, we will help learners and it must help faculty.

Second example is that as we train these large language models, we will be sure to protect learner privacy. We will be very careful not to train these models with learner private data so it doesn't go and become public. So these are just two examples of some of the guardrails we've put in place for our own responsible use of AI. And we've made these guardrails public so we can foster a dialogue around the world in terms of what are these guardrails for ethical and responsible use of AI.

Ramachandran S: Professor, on the corporate side, we have seen a shift towards skill-based employment. You have been a pioneer of microcredits. How do you see microcredits and their acceptance in the corporate world today in recent times? Can you talk to us about that?

Professor Anant Agarwal: Although the edX platform supports a whole range of courses, we have over 4,000 programs on edX ranging from individual free MOOCs and courses all the way to about 200 degrees. We have bootcamps, we have executive education for execs. It's a whole range of programs that we have. But what we've seen is that after the pandemic upskilling and re-skilling the workplace has become even more important than before because the future of work is completely being transformed and we are seeing a much, much higher uptake, both for corporate learning in our own enterprise business and also in our B2C consumer business. More and more uptake of these micro-credentials like professional certificates, bootcamps or executive education, micro-masters and micro-bachelors where micro-bachelors and micro-masters are backed by credit. So a lot of learners are looking for these short-form micro-credentials that they can complete in a few months. We've even launched micro-bootcamps. Our bootcamps used to be six months long. Now the micro-bootcamps, these are very hands-on with live interaction. A micro-bootcamp, we just launched one in artificial intelligence last week. It's a AI micro-bootcamp where you can upskill in 10 weeks. So if you have some basic coding experience in Python, for example, and you have a basic math background, you can take our micro-bootcamp in AI and upskill very quickly in terms of how to apply AI in your job.

If you are a coder or if you are in business or if you are in data science, everybody's using AI now in what they do, and so you can quickly upskill. In fact, I argue that much like basic coding was an important skill, that computational thinking which was important for everybody, I believe going forward, understanding the basics of AI will be a basic skill that will be required for everybody. And we are creating many such short free courses like Prompt Engineering or Introduction to ChatGPT where you can upskill and learn about ChatGPT in about two hours. And then in terms of how you apply, how you apply AI in coding and so on, we have a number of machine learning and AI programs. We've even launched a one-year master's degree with UT Austin in machine learning and AI for just $10,000. So we are rapidly coming out with this program's short skills-based micro-credentials all the way to full degrees so that we can very, very quickly serve the huge need for AI upskilling and re-skilling in corporations and in colleges.

Ramachandran S: Again, on the corporate side, there is a lot of push towards work from office today for healthy collaboration, for networking, for relationship between colleagues. Do you see that happening in the education sector too? You have spoken about unbundling of education into its multiple components. Do you see unbundling taking a step backwards or is it still continued? Can you talk to us about that?

Professor Anant Agarwal: So unbundling relates to creating a system where you can get a piece of something. So let's use my own example. When I went to IIT, I joined it in 1977, IIT Madras and I graduated in 1992, in 1982, 5 years. I didn't take 15 years to graduate. In those days, and even now for that matter, in most universities, if you go to a university and let's say you spend one year in college and then you say, look, I want to go and do a startup company. I've learned enough of coding and math that I want to go do a startup and go do a startup. And many of our students at MIT that we have are doing this. Today we call them a dropout. If you leave after spending one year or two years in college, you're called a dropout. In India, if I had done that, I could not go and show my face in my house. My parents would've kicked me back and send me back and say, "You go back now."

But what unbundling does is it celebrates smaller pieces of learning. So the unbundling argument I make is that why should everybody learn everything from one university? Why should I spend four years in college and learn everything from there? Why can't we build an education system that is like Lego blocks where each course or a set of courses is like a Lego block and I can combine these Lego blocks to create a full degree where some pieces can come from other places. So I'll give you an example. So we are working with several universities in India, for example, DY Patil University in Mumbai and in Pune. So there students, let's say they're getting an MBA or a degree from DY Patil, and we have a partnership with them. Think of it as not B2B, but B2U, business to university.

So we're working with many universities all over the world where DY Patil subscribes to our university catalog of programs. And as students are going through DY Patil's education, they take a micro-credential, a Lego block from one of our edX partners. It could be from a micro-bachelor from London School of Economics, for example. Or it could be a micro-master's from MIT. They take a program as a small unbundled Lego block from edX so that when they graduate from DY Patil, they not only get a MBA or a degree from there, but they also have a very skills focus credential that has come from somewhere else. And that when they put on their resume or on their LinkedIn profile can really help them with the job market. And so this is one example of unbundling where in your education you get one piece or many pieces from different places so that you can synthesize the skills that you need even if your university doesn't offer it. And this is becoming particularly true for AI and machine learning where many universities and colleges and corporations for that matter, don't have existing training in AI and machine learning, so people need to rapidly upskill. And so here, getting such a subscription license and having your students take these programs from elsewhere can help you very quickly upskill your students and make them job ready on day one.

Ramachandran S: Thank you, professor. A lot of interesting points you have brought up. We can keep on talking forever, but we really appreciate you taking your valuable time and sharing these insights with us. Thank you so much. You've given us a lot of food for thought to go back and see where we are, and I'm sure our audience will also find this very interesting. We really appreciate the time you have taken to spend this time with us on this call. Thank you so much.

Professor Anant Agarwal: Thank you very much, Ram. It was a real pleasure joining you.

Ramachandran S: For all of you watching, if you like this, please check out our earlier conversations on the video section of Infosys Knowledge Institute page, Infosys.com/IKI. Until next time, keep learning. Keep sharing. Thank you very much.