Prof. Ravindran on The IIT Madras Playbook for Building AI Leaders | BP2B S1 Ep. 6

1. 0:00 – 8:32

   Intro

   1. BRBalaraman Ravindran0:00

      So I was at this, uh, AI conference in, in Macau, right? And then one Chinese student who was sitting there looks at me, that- he gets up, bows to me, and said, "I learned ML from your videos," and gave me his seat. But AI is not going to take your job away, but somebody who knows how to use AI to do your job is gonna take your job away. Then Director Professor Anand, he never believed in this artificial boundaries of, uh, departments and subjects and things like that. People were just, you know, teaching subjects, and students used to can, you know, put together their curriculum based on what they are interested in learning. [upbeat music]

   2. SPSpeaker0:50

      Hi, my name is Amrit. We've heard that IIT Madras is the best place to build. [upbeat music] So we've come down to the Sudha and Shankar Innovation Hub. We want to meet some people. These are builders. We want to talk to them about their work, and also ask them, "What makes IIT Madras the best place to build?" [upbeat music] Hello, and welcome to the Best Place to Build Podcast. Today, we are sitting with Professor Ravindran, who is the head of the Wadhwani School of Data Science and AI at IIT Madras. Professor, welcome to the podcast. Uh, we know each other for a while now. We traveled together to Bangalore to meet JEE aspirants earlier this year. That was a great experience. What I remembered from it, though, is that both of us drink black coffee now because we can't digest milk very well.

   3. BRBalaraman Ravindran1:45

      Sure.

   4. SPSpeaker1:45

      Um, and, uh, and also I remember that I had given you the Best Place to Build sticker that day, and you immediately put it on your, uh, laptop. Yeah.

   5. BRBalaraman Ravindran1:54

      Yes.

   6. SPSpeaker1:55

      That was amazing.

   7. BRBalaraman Ravindran1:55

      Because I strongly believe that this is the best place to build.

   8. SPSpeaker1:58

      Yes, sir. Um, and I, I, I also... It's amazing how popular you are among the student community. Your classes are taken by, uh, hundreds, 200, 300 of students. And, um, I want to start with the question that: Was it always like this? In the sense that, was AI this popular?

   9. BRBalaraman Ravindran2:20

      Sure. So, um, I have been with IIT Madras since 2004, right? And in 2004, I started teaching the course on, uh, reinforcement learning, right? At that time, people had not heard of reinforcement learning, so whenever I say reinforcement learning... And my office used to be in the Building Science block, so people thought I worked in concrete.

   10. SPSpeaker2:39

       [chuckles]

   11. BRBalaraman Ravindran2:40

       Right? So, uh-

   12. SPSpeaker2:41

       Because reinforced concrete.

   13. BRBalaraman Ravindran2:42

       [chuckles] So, uh, at the first time I taught this class, there were about six students in the class.

   14. SPSpeaker2:48

       That's right.

   15. BRBalaraman Ravindran2:49

       And ironically, in the early days, there used to be more non-computer science students were taking my reinforcement learning class than there were computer scientists. Because, uh, uh, reinforcement learning, it's such a wonderful subject. It has motivations from control theory, it has motivations from, uh, uh, neuroscience, uh, and cognitive psychology, and so on and so forth. So I used to have people doing PhD in neuroscience in my class, and so on and so forth, but the class used to be so small, five, six students. I think up to, up until about 2012, the largest reinforcement learning class I had was about 16. Now, the smallest class is about 130, right? So year on year now, we get about 180 students who take it, and then the first assignments and the first quiz rolls around, and then I'm left with about 110, 120 students. About 50 drop the class after that, but it's still a very popular class. It's not always been like this, but I'm happy that we are.

   16. SPSpeaker3:43

       Yeah, uh, and, uh, you, uh, not just the courses you take, uh, for machine learning and reinforced learn- reinforcement learning, which are very popular, you're also involved in the, uh, data science, uh, online course, which I think how many people take it?

   17. BRBalaraman Ravindran4:00

       Oh, yeah. So that was about, uh, I- uh, the number keeps changing almost weekly. I think we are thirty-five thousand plus now.

   18. SPSpeaker4:08

       Thirty-five thousand plus are enrolled in the-

   19. BRBalaraman Ravindran4:10

       Online BS, uh, in Data Science and AI, right? So there's an amazing team that runs this. So I basically teach reinforcement learning in that program, but then there's a lot more that is designed. It's a full degree program, and a lot of people put in the effort to make it, uh, as rigorous... In fact, in some cases, I would say even more rigorous than classroom-taught, uh, courses. So I think it's, it's, it's been an amazing experience both for the teachers and the students.

   20. SPSpeaker4:37

       In fact, I read today morning that a lot of students who, um, for whatever reasons, haven't managed to get into IIT or who are in some other college, end up taking the BS, uh, in Data Science and sort of augmenting their skill set. That is amazing.

   21. BRBalaraman Ravindran4:51

       In, in fact, it's not just people who didn't get in IIT. There are people who are enrolled in IITs, uh, who are also doing this as an additional degree, because their primary degree could be in something very different. Uh, they're also doing this as an additional degree. And in fact, some of the courses are so good, right? So we are allowing even our regular BTech students to take them for credit.

   22. SPSpeaker5:12

       So that's where we're gonna... And I, I know that, um, s- so I graduated in 2008, and I entered IIT in 2003, so a lot of my friends sort of missed the machine learning, AI, reinforcement learning wave. So a lot of them have, uh, subsequently learnt it. I didn't, but subsequently they have sort of learnt it, and a lot of them come back and tell me that we actually started with Professor Ravindran's NPTEL lectures. So your NPTEL lectures are also very popular, so-

   23. BRBalaraman Ravindran5:40

       Yes, they are. In fact, they still remain, a, a useful resource. So I think we hit a sweet spot, right? So most of the online content, uh, are pretty light on the formalism, and then they don't introduce the math. They don't- they just directly jump into what I call a, uh, kind of a laundry list way of, uh, teaching, uh, machine learning. They just go through the algorithms, right? Instead of talking about the fundamental principles, and so on and so forth. And I think we hit the sweet spot because it was not really designed as an MOOC.... right? If you look at my in- machine learning and my reinforcement learning NPTEL lectures, they were actually taped in my classrooms. So this is designed as classroom, uh, uh, lectures, and, uh, the interaction that, uh, that's going on with the students adds a lot of value to these videos, and that is, you know, kind of gives a literally an immersive feeling to the people watching the videos because they feel like they're part of the class, they're watching along-

   24. SPSpeaker6:33

       Yeah

   25. BRBalaraman Ravindran6:33

       ... with the students. I think that was actually very helpful, I think.

   26. SPSpeaker6:36

       Yeah. But, uh, uh, on this point, so we are tracing back from 2003, 2000, 2004, 20 years back, we have six students. 2014, we have, I don't know, about 200 students.

   27. BRBalaraman Ravindran6:47

       Yeah.

   28. SPSpeaker6:48

       And now it's, like, 400 in classroom, 35,000 in online degree, and another, I don't know, 100,000 in, uh, in NPTEL. It's amazing.

   29. BRBalaraman Ravindran6:57

       And so the NPTEL machine learning course is, is, well, fairly popular. I think about 8,000 people write the exams, and, and so a lot more take it, right? I mean, many people start doing it, but they don't come to write the exam. Four- 8,000 people write the exam every time it is offered, and it has been offered, oof, 30, 35 plus times now. So a lot of, a lot, uh, yeah, a lot of people have looked at it.

   30. SPSpeaker7:19

       It's safe to say that a lot of people, a lot of students in India have been in your lecture in some way. It's amazing.
2. 8:32 – 11:51

   Machine Learning and Reinforcement Learning explained

   1. SPSpeaker8:32

      come back to this? Uh, so the course we were talking about is reinforcement learning and machine learning. I understand that your area of interest, your area of research, started from reinforcement learning. So can you tell us a little bit about what that is and what you were researching? I know from the friends and my research team that you are a very well-respected, um, uh, researcher in that field. So maybe if you can give us an overview.

   2. BRBalaraman Ravindran8:57

      Sure. So, um, normally, when people talk about machine learning, right? So how it operates is that you have a set of, quote, unquote, "labeled examples." What do I mean by that? So that is a specific input. It could be an image, it could be, you know, a set of numbers that you give, and you expect the machine to produce a certain output, right? That could be, "Okay, if this is an image, I want you to say it is an achate, or if this is an image, I want you to say it is a deer," right? So that's basically what you would wanting this to learn. And mathematically, if you look at it, what that machine is doing is learning a function that takes the input and maps it to a specific output. But there are a lot of problems which do not naturally fall in this paradigm, right? The usual example I give is a thought of learning to cycle. Suppose you're learning to cycle, nobody can sit down and say, "Okay, if your cycle is going at this speed, and it is tilting at this angle, then you should push down with your right leg, with so much, you know, pounds of pressure." With these kinds of, you know, labeling is just doesn't make sense, right? When you are to trying to learn to cycle. The typical feedback you'll get is, if you are a, if you are a kid learning, and that there's a parent watching you or some- or an elder watching you, they'll say, "Oh, no, please be careful, don't fall," right? So things like that, right? So that's all the kind of in- instruction you're going to get. Not saying what's the exact output you should be producing, right? So these kinds of problems, where you're getting, you know, very minimal feedback. Somebody's telling you, you are doing something wrong. When they say, "Hey, careful, careful," that means they're telling you, you are doing something wrong, but they don't tell you what is right, right? So such, uh, you know, kinds of problems, where the feedback is more, uh, more sparse, right? Where you're not... Somebody is not telling you what is the right thing to do, but is evaluating your output.

   3. SPSpeaker10:42

      Hmm.

   4. BRBalaraman Ravindran10:43

      Right?

   5. SPSpeaker10:43

      As you are saying-

   6. BRBalaraman Ravindran10:44

      Not as reinforcement learning.

   7. SPSpeaker10:45

      A- as you're saying this, I'm reminded of my, um, swimming coaching experience in IIT Madras, where we were taken on top of a diving board, uh, the Olympic-size swimming pool, and the coach just push. I, I died. I mean, I didn't die obviously, but I had to be rescued. Okay, um, so are you saying that-

   8. BRBalaraman Ravindran11:05

      That's, that's very [chuckles] extreme, but suppose you're swimming, right? I mean, think of it, right? So in fact, a lot of things that we do, not just cycling, right? Swimming, walking, even talking as baby, right? So all of these was learned through this kind of a trial and error model, right? You try something, and somebody tells you whether it's working or not, or you yourself realize. If, if you are a baby, you get up, you try to walk, you fall down, right? So you realize that you're doing- not doing it right, and then you keep trying it again. If you're pushed into the water, right, and then you're trying to swim, somebody says: "No, no, no, don't do that. Keep your head down." Right? But that doesn't really tell you that, uh, you know, what you should be doing to stay afloat, right? So, so, so very kind of minimal feedback that is coming from understood. That is what reinforcement... I mean, the mathematical formalism that allows you to learn is what reinforcement learning is. Understood.

3. 11:51 – 15:35

   A Pioneer's Journey into Artificial Intelligence

   1. SPSpeaker11:51

      I feel like today there's a large awareness of AI, and-

   2. BRBalaraman Ravindran11:54

      Mm-hmm

   3. SPSpeaker11:54

      ... maybe I can understand what you're saying, like, but 20 years back or 30 years back, when you were a student, uh, how did you get into this field? What interested you?

   4. BRBalaraman Ravindran12:02

      Um, 30 years ago, AI was hot, right? And then it cooled off, and then it became hot again. But I didn't get into AI because, just because it was hot, because I, I went to a small engineering college down south.... and, uh, so the, the people are not teaching AI or anything like that. But I got fascinated, uh, uh, you know, more from trying to understand, uh, how, you know, how humans think, right? This all got, uh, kindled by, uh, uh, the, the kind of frenzy that was around AI beating, uh, human beings in chess back then, right? So there was all this, uh, chess-playing programs, these are the coming-

   5. SPSpeaker12:41

      Kasparov matches.

   6. BRBalaraman Ravindran12:42

      Uh, Kasparov matches came a little later, but, I mean, the, the one where Deep Blue beat Kasparov came a little later, right? That's in late, like, the mid-'90s. This was early, early '90s is what I'm talking about. So there are still computers that are playing them, but then, uh, uh, the humans, it was competitive with humans. So there's a lot of- because, because of Viswanathan Anand, right? So chess was a lot more popular in, uh, Tamil Nadu than elsewhere. Uh, uh, so we used to follow what was happening, and kind of all this, uh, AI models, uh, uh, were starting to beat humans in, uh, chess. And so I started thinking about, "Okay, what are these models doing?" Right? How, how... Uh, the, the bigger fascination was these AI models were supposed to mimic how humans think, right? So I wanted to understand how these models work. So I picked up books on my own, right? And, and we started, I started reading about, uh, uh, uh, neural networks and, uh, AI in general. And then, uh, I got a little disappointed because, uh, AI by then, neural networks particularly, by then had become, uh, a mathematical tool, that they had moved far away from their biological origins of, uh, trying to explain how the brain functions, and so on and so forth. And, uh, so slowly I started reading up and started exploring stuff, and then I came upon reinforcement learning. Uh, this is back in '93, uh, where, uh, at that time, people were doing experiments with monkeys, right? They were sticking electrodes in monkey brains, uh, by... And then, and then, uh, looking at how the, uh, uh, uh, the neurotransmitters and signals were behaving in the brain as the monkey is learning stuff, right? And then they came up with reinforcement learning. They said, "Oh, yeah, this field called reinforcement learning seems to explain what is happening in the monkey's head." And that kind of became a motivation for people to, you know, look deeper into RL, right? And that kind of fascinated me, and then I started looking at reinforcement learning. Again, it turned out I had to learn a lot of optimization to follow reinforcement learning, but, uh, still the, the motivations were more closer to the biological origins.

   7. SPSpeaker14:40

      You, you're talking about a time when, uh, as a student, you were maybe anticipating that these algorithms or these computers will eventually beat humans at chess, and then they did.

   8. BRBalaraman Ravindran14:53

      That happened, yeah, in a few years after that. Yeah. Uh, yeah.

   9. SPSpeaker14:56

      Must have been exciting.

   10. BRBalaraman Ravindran14:57

       Oh, very exciting time for me, yeah. Must have been really exciting, yeah.

   11. SPSpeaker15:00

       And now they've... Obviously, humans are beating Go. I mean-

   12. BRBalaraman Ravindran15:03

       Yeah. So I, I have this, uh, uh, nice, uh, uh, headline that I use in my talks, right? Uh, so in 2022, there was a headline saying, "Man beats computer in Go." [chuckles]

   13. SPSpeaker15:15

       [chuckles] That's, uh, given that, uh, computers are smarter than us at chess and Go.

   14. BRBalaraman Ravindran15:20

       And Go.

   15. SPSpeaker15:21

       Yeah.

   16. BRBalaraman Ravindran15:22

       Now, "Man beats computer at Go" is basically love become news again. Mm, sure.

   17. SPSpeaker15:27

       I mean...

   18. BRBalaraman Ravindran15:27

       I don't want to use the word smart. They're better than us.

   19. SPSpeaker15:30

       They're better.

   20. BRBalaraman Ravindran15:31

       Playing-

   21. SPSpeaker15:31

       Better. Better than. Um, I, I was,

4. 15:35 – 18:55

   Understanding Multi-Armed Bandits

   1. SPSpeaker15:35

      uh... So in your research, I think at one point, uh, you talk about, um, is the word bandits?

   2. BRBalaraman Ravindran15:42

      Mm-hmm.

   3. SPSpeaker15:43

      Uh, can you explain that to it? It's very fascinating. I mean, I don't even know what it means, but it's very fascinating. [chuckles]

   4. BRBalaraman Ravindran15:47

      In fact, in fact, the, uh, the term bandits, right? It's actually- it's usually, uh, multi-armed bandits, that's what the term is. Uh, uh, uh, the whole problem setting itself, right, predates, uh, quote, unquote, "reinforcement learning," right? Uh, it's, it's, it, it was studied by statisticians for more than a century now. So the idea is something very simple, right? So you have a situation, right, where you have, like, N choices, right? You can do X, Y, Z. Let's just stick it with two choices. Stick, stick to two choices, right? So you can either go to a movie, right, or go to a restaurant. Okay, only these two choices, and now you ch- you choose which one you want to tonight. So sometimes you go to a movie, it's, it's great. The movie, uh, uh, um, it's a great movie and you get a fantastic reward. Other times you go to a movie and, uh, it really sucks, right? So, so it's... You don't know whether, what kind of enjoyment you're going to get by going to the movie unless you actually go to the movie and, and test it out. And likewise, food, right? You might go to a restaurant, you order some food, and you might fall sick, or you might order some food and it's, like, amazing-

   5. SPSpeaker16:53

      Yeah

   6. BRBalaraman Ravindran16:53

      ... amazingly good, right? So you don't know what is going to be the outcome of it. You have some notion of what could be the outcomes, but you have to actually try these things out before you know what's going to be the exact outcome, right? So there are many problems that are modeled like this, multi-choice problems, where you have to actually test it to know what the exact outcome going to be, right? So these kinds of problems are essentially modeled as these multi-armed bandits, and the reason for the term bandits is very interesting. And now we see what it is. It's nothing to do with bandits, right?

   7. SPSpeaker17:24

      Mm.

   8. BRBalaraman Ravindran17:24

      So it's, it's basically a multi-choice problem with the stochastic outcomes, and you are trying to optimize for it. But then if you think about the term, right? So it goes back to slot machines. So you know slot machines, uh, you put a coin, you pull a lever, and the coin- and the machine takes your coin, right? I mean, that's how it normally operates. Occasionally, they give you-

   9. SPSpeaker17:42

      Yeah

   10. BRBalaraman Ravindran17:42

       ... back a few coins to keep you engaged and keep you going, right? So you can think of it that way, right? So in the slot machine, you basically have only one choice that you can make.

   11. SPSpeaker17:50

       Yeah.

   12. BRBalaraman Ravindran17:50

       Pull the lever, right? Once you pull the lever, with some probability, you get some positive outcomes, otherwise, you get a negative outcome, right? So people kind of said, "Okay, this looks like a bandit problem." Um, okay, I will tell you what's bandit is. It looks like a slot machine problem.

   13. SPSpeaker18:05

       Mm.

   14. BRBalaraman Ravindran18:05

       But slot machines had a very colorful name, they were called one-armed bandits.

   15. SPSpeaker18:09

       Oh.

   16. BRBalaraman Ravindran18:10

       They have one arm-

   17. SPSpeaker18:11

       Yeah

   18. BRBalaraman Ravindran18:11

       ... you pull the arm, and it steals your money.

   19. SPSpeaker18:14

       That's why.

   20. BRBalaraman Ravindran18:14

       That's called a one-armed bandit. They said, "Oh, this looks like a slot machine," except that you have multiple slot machines, and you can pull- you can choose which arm to pull.... so instead of a one-arm bandit, you have a multi-arm bandit.

   21. SPSpeaker18:25

       That's where it comes from. That's very interesting, guys. So-

   22. BRBalaraman Ravindran18:27

       In fact, in fact, the, um, I forget, uh, that is a m- there's an airport in the US which actually has a slot machine museum. Uh, I, I, maybe Detroit, I, I don't remember which one. They actually have a slot machine that is dressed up like a bandit, and it has this one hand that you go and pull. It's really cool. I wanted to take a picture of that, I, I missed it.

   23. SPSpeaker18:47

       We'll, we'll pull up a picture for this podcast and show it at this point to the users.

   24. BRBalaraman Ravindran18:51

       Sure.

   25. SPSpeaker18:53

       Uh, to, uh... So, um, and the other

5. 18:55 – 23:26

   Exploration vs Exploitation dilemma

   1. SPSpeaker18:55

      thing that I found very interesting, the, the terms that I picked up, and, and I know you have a, you have a problem with the word terms, it, with the use of terms in AI. Um, the term I picked up, and it resonated with me, and I thought about it for a while, is: is there a trade-off between exploration and exploitation?

   2. BRBalaraman Ravindran19:13

      That is something that is at the heart of reinforcement learning.

   3. SPSpeaker19:17

      Okay.

   4. BRBalaraman Ravindran19:17

      Okay? So, uh, let's go back to the simple example that we had with the bandits, right? So, uh, there is a movie and, and, and, and, and that, that is food, right? So it turns out that I'm, let's say, I'm, I'm a latent foodie. When you look at me, you know that I am. But let's say that I'm a latent foodie, I haven't realized that yet, and I would actually love to eat food more than go to movie, right?

   5. SPSpeaker19:37

      Yeah.

   6. BRBalaraman Ravindran19:37

      But then let's say I, I, I try going to a movie, and let's say I end up in a, in a great, great, uh, masterpiece by chance, right? So now I'm going to say, "Okay, that's great. That's, that's movies are, are great, but let me also try food once, because I don't know if maybe I'll get a better experience with food." I eat food, and then maybe I eat something that I'm allergic to, and I, I, I vomit. And so I say, "Hey, okay, going out to restaurant is a bad bet, let's just stay home and eat. Let's always go to movies," right? But it turns out, you know, on an average, you know, movies suck in general nowadays, right? So there are very few good movies that are being made. So you keep going to movies, and you kind of settle down, okay. Yeah, blah, okay, movies are fine, but, oh, they're way better than vomiting.

   7. SPSpeaker20:18

      Okay.

   8. BRBalaraman Ravindran20:18

      Right? So now I've started exploiting going to the movies. But if I had explored a little bit more, you know, saw a few more movies and eat, and go, go out to restaurants a few more times, I might have decided, "Hey, restaurants is what I want to do," right? So that's where the exploration vs exploitation trade-off comes in. I very quickly started exploiting, based on a couple of outcomes, right, going to movies, because they seem to do well for me. Exploration is when I'm actually trying out multiple things several times before I, I decide what I should do, right? So b- because I was telling you, right, unless you try things, you don't know the outcome in, in, in both in reinforcement learning and in bandits, right? Unless you try things, you don't know the outcome, therefore, you have to explore. You have to try things. If you don't try things at all, you're not going to learn anything about the system, so you have to try things. Exploration is essential. But then, as soon as you find something that is reasonably good, you can start exploiting it, right? Uh, if you start exploiting too soon, you might, you might end up with something that is not the best solution for you, that is suboptimal. But if you continuously keep exploring, you are leaving out an opportunity for, you know, getting your enjoyment, right? So because even though you don't like going to movies, you keep going to movies just so you want to explore. So that's not a great idea. So at some point, you have to decide, "I'm going to switch from exploration to exploitation." You shouldn't do it too soon, shouldn't do it too late. That's the dilemma. So when do I switch? And that is at the crux of a lot of reinforcement learning problems, and it makes RL hard, uh, the solving this exploration-exploitation dilemma.

   9. SPSpeaker21:50

      It, it's very interesting that we are, we are talking about this in context of RL, but it's true about-

   10. BRBalaraman Ravindran21:55

       Yeah

   11. SPSpeaker21:55

       ... human life.

   12. BRBalaraman Ravindran21:56

       Human life. True, but all kinds of decision-making, where you have to try things to know what the payoff is really, right?

   13. SPSpeaker22:03

       Yeah.

   14. BRBalaraman Ravindran22:03

       There are... I mean, life is complicated. There are very few things in life that you can model exactly so that you know what the payoff is going to be beforehand. You only have a rough estimate of things. So you... Yeah, it's-- So in fact, and it's not even called a trade-off, it's called the explore, exploit dilemma.

   15. SPSpeaker22:19

       Dilemma.

   16. BRBalaraman Ravindran22:20

       Dilemma.

   17. SPSpeaker22:20

       Yeah. And, um, and that's a great phrasing. I think students who end up at an institute like IIT or IIT Madras, where there's actually a lot of choice-

   18. BRBalaraman Ravindran22:31

       Mm-hmm

   19. SPSpeaker22:33

       ... uh, including whether to take your machine learning class or not, uh-

   20. BRBalaraman Ravindran22:37

       The students are complaining there's no longer their choice, but in me. [chuckles]

   21. SPSpeaker22:42

       I think it definitely that's running in their heads.

   22. BRBalaraman Ravindran22:46

       It's interesting.

   23. SPSpeaker22:46

       So let's move on, um-

   24. BRBalaraman Ravindran22:50

       So sh- I'll just explain why I said it's no longer their choice.

   25. SPSpeaker22:53

       Okay.

   26. BRBalaraman Ravindran22:53

       Because they're getting allotted by lottery now. [chuckles]

   27. SPSpeaker22:55

       Is it?

   28. BRBalaraman Ravindran22:56

       Yeah, because they just apply, and then if they are in, so the first 100 people get to apply, then you get it.

   29. SPSpeaker23:01

       But that's where everybody is applying.

   30. BRBalaraman Ravindran23:02

       Everybody is applying. So it's not their choice whether they would get to do it or not, it's, it's more like a system choice, not that... Yeah.
6. 23:26 – 26:20

   "AI is the new CS"

   1. SPSpeaker23:26

      Professor, I, I want to say that I've heard you say that AI is the new CS, uh, in multiple forums. So I, I want to understand this a little better.

   2. BRBalaraman Ravindran23:37

      Ah, so if you think about how compute- computer science as a discipline evolved, right? So in the early days, computer science was mostly math, right? So mathematicians used to do computer science, and then slowly became something which got clapped together with electrical engineering. Even now, if you see a lot of departments worldwide, you know, they are called EECS, right? Electrical Engineering and Computing, uh, Computer Science, right? So the, the, the idea here is that so much about building a computer was about electronics and electrical engineering. So you needed to know how you'll design the chips and how the different components communicate with one another, and, and so on and so forth. There's a lot that you had to know about building the whole system before you can get into learning about computers.... But now, slowly in the '80s, what happened was, computer science as a discipline matured enough that there was enough to study about computing, right? About writing program, about structuring programs, and thinking about higher levels. They're thinking about algorithms, thinking about protocols, that you could move away from the actual details of how the machine itself was put together, and start thinking about higher level concepts, right? Then, computer science started pulling away from electrical engineering. That the reason it could do that was because computer science had a lot of influences coming in from other disciplines. It was never just a sub-discipline of electrical engineering. So there were a lot of things coming in from logic, things coming in from mathematics, right? And, and so on, so forth, right? So, so computer science could, could mature as a field. So that is ex- uh, so sometime in the early to mid '80s is when that started happening, right? And AI is now at the same situation that computer science was back then. So, so far we have been studying AI as a subfield of computer science, but it's not truly that. I mean, there is a lot of influence from psychology, right? So people have... In fact, some of the earliest, uh, AI, uh, researchers were psychologists, right? So a lot of influence coming in from psychology, a lot of in- influence coming in from control engineering, from economics, right? Apart from, you know, logic and algorithms and systems, and so on. So, so now we are at a point where we don't necessarily have to know everything about how computing is done to be able to do meaningful and strong, uh, AI, right? Uh, uh, meaningful and more fundamental AI research and AI, uh, solutioning. Therefore, it's time for AI to pull away from computer science and become a discipline in its own right. I think we have-- the field has matured enough that we can start thinking about it. So that's why I'd say AI is the new CS, because it's exactly where the CS was in the, uh, in the mid '80s, right? So now w- the AI is in the-

   3. SPSpeaker26:15

      It's interesting. So at IIT of Madras, of course, um, the,

7. 26:20 – 35:19

   Wadhwani School of Data Science and AI

   1. SPSpeaker26:20

      the, the Wadhwani School, uh, of Data Science and AI is like a new department. I think it's the sixteenth, eighteenth, eighteenth department at IIT Madras. And, um, so how did it come about? And also, can I call you the co-founder of the Wadhwani School of Data Science and AI?

   2. BRBalaraman Ravindran26:39

      Sure.

   3. SPSpeaker26:40

      Okay, done. [chuckles] So please tell us the origin story.

   4. BRBalaraman Ravindran26:45

      Sure. Uh, so, um, so in, in IITM, right, back in 2015 itself, so we had kind of started thinking about, uh, AI and data science as a strongly interdisciplinary field, you know? This is not just something, uh, you know, the computer science or the management school folks do. But we found that a lot of people were actually not just using AI as a tool, but were actually developing new algorithms so that it, it could work well with their, uh, uh, you know, disciplines, right? Whether it is, uh, chemical engineering or biosciences, people are actually coming up with new algorithms, uh, that could work well for their kind of applications, right? So we, we felt that we should synergize this, and this was facilitated by our, uh, director back then, uh, Professor Bhaskar, uh, who actually, you know, said that you should start an interdisciplinary group that looks at data science and AI. So we, we, we started working back in 2014. And over time, what has happened is there are multiple, you know, degree programs that we offer, uh, which, which, uh, uh, students from, uh, different departments... In fact, uh, we have this interdisciplinary dual degree program on data science, uh, which, uh, almost every department student gets into it, at ex- except CS, right? Because computer science students never opt for it, but all the other, uh, department students opt for it. And I think at one point of time, I had, uh, students from 12 different departments doing a, a, a, a, a dual degree in data science.

   5. SPSpeaker28:04

      This is the I triple D Interdip- Interdisciplinary-

   6. BRBalaraman Ravindran28:08

      Interdisciplinary

   7. SPSpeaker28:08

      -Dual Degree Program.

   8. BRBalaraman Ravindran28:09

      Yeah.

   9. SPSpeaker28:10

      There are 10 of them, and you're talking about the one in data science.

   10. BRBalaraman Ravindran28:13

       Actually, more. Yeah, they're more than 10.

   11. SPSpeaker28:14

       Okay.

   12. BRBalaraman Ravindran28:14

       Yeah, the one in data science was the one-- one of the first that was started, and still the most popular, and still, still has the largest, uh, student strength. You know, 75 students take it every year. The idea here is that you do a BTech in some other, uh, program, and but you do a master's in data science, right? You get a five-year, uh, do two degrees, so that's basically the idea. And, and a lot of students were taking it, right? So basically, there were many such interdisciplinary, uh, degree programs that we were offering, and a lot of interdisciplinary research centers that we were starting, right? So where there will be an AI, data science and AI center, but faculty from many, many different departments. Computer science was certainly one of the largest, but we had people from chemical engineering, management school, mathematics, civil engineering, every, every department, right, was contributing to, uh, research and activities in these interdisciplinary centers. So at a point, we said, "Look, there are so many activities that are going on, right? And we really don't have a proper mechanism for doing all of this. So, so let us do a Department of Data Science and AI." I mean, literally, we were doing all the jobs that the department was doing without the department, right? So we said that now it's about time we start a Department of Data Science and AI. We wanted to, you know, take a larger view of it, right? Because there were so many interdisciplinary centers already running, right? And we have, uh, had all these degree programs. So instead of just launching a department which will administer the degrees, we kind of set up this as a larger school of data science and AI, right? Which had the degree programs, as well as all of these centers that were operating under one umbrella. So you could say that the School of Data Science and AI kind of houses most of the primarily AI activities in the institute right now. And once we decided that we are going to go ahead with it, uh, so, uh, we approached, uh, Sunil Wadhwani, who is one of our distinguished alum, and he readily agreed to support the, uh, uh, school. So that's why it's the Wadhwani School of Data Science and AI, with, uh, supported, uh... Of course, it's, it's supported l- largely by MHRD, so we have to say that the Ministry of Education, uh, uh, is giving, uh, it's, it's, it's, it's a standard department like any other. Right? So you're getting the same budget that's from the Ministry of Education. But we're able to do a lot more, thanks to the support from, uh, Sunil Wadhwani. He actually gave us about 100 crores, uh, for supporting the department. Right.

   13. SPSpeaker30:22

       Professor, I'll just, uh, I'll just say what I heard. Correct me if I'm wrong. So three things coming together. One is that AI, as a field, grew, uh, large enough.

   14. BRBalaraman Ravindran30:32

       ... Um, second is that there were already lots of interdisciplinary things happening. They could coalesce together. And third is that we had this alumni who was very keen to support this initiative.

   15. SPSpeaker30:41

       Yep.

   16. BRBalaraman Ravindran30:41

       Uh, Sunil Wadhwani, right?

   17. SPSpeaker30:44

       Nice.

   18. BRBalaraman Ravindran30:44

       So I should say, even if Sunil hadn't given the money, we would've gone ahead and started the department, but that was the kicker.

   19. SPSpeaker30:49

       Okay.

   20. BRBalaraman Ravindran30:50

       Really, that was really the last, last thing that tipped the scale. Yeah.

   21. SPSpeaker30:54

       Nice. Um, I, I, uh, I mean, it's interesting that, um, um, today, of course, a lot of professors are, uh, founding companies with their students or with their, um- with each other, and so on, and the entrepreneurship scene in IITM Madras is blowing. Um, but I was in-- I was interested to see that you have been at the start of, um, this AI department, also of, um, other AI initiatives like CeRAI, uh, the CoE in AI, the Robert Bosch Center. Um, how is it... I mean, if you could tell us a little bit about these things. And also, what I'm interested to know is: How is it different or how is it similar to co-found a department versus to build a company, or build a department versus to build a new?

   22. BRBalaraman Ravindran31:42

       See, the one thing that is common between both is that the team is important.

   23. SPSpeaker31:48

       Uh.

   24. BRBalaraman Ravindran31:49

       Whether you are fou- founding a company... I mean, I've heard VCs tell me that, "I don't really listen to what the pitch is, I look at what the dynamics is between the co-founders before I decide whether the company will succeed or not." Something similar, right? So if you don't have sort of like-minded, motivated people coming together, any of these large initiatives that we are doing, the departments or centers that we have started, would, would fail, right? So that is the thing that is very similar. But then the whole bunch of other things are different. What is our-- What, what are we looking to in terms of delivering and things like that? All of those would be very different, right? And we are not essentially shooting at a single product, right? Like, like a startup would be, right. So it's not-- you don't have this-- need to have this one single, unique product that we are going to take to market. Uh, uh, and we have, uh, you know, our, our product is very different. It's more like, uh, uh, uh, uh, us enabling, uh, a culture of research and innovation, right? That's a very lot more vaguely defined, uh, product. And so we have to worry about what kind of KPIs that we will look at, and, and those things are different, right? So how you are evaluated and how you raise funding, all of this is fairly different. But then you have to raise funding. That is something common to... I mean, as a prof, right, I just sit back, go teach, and I don't have to worry about raising money because the government gives us money. But if we want to do something different or want to start a big center, so you'll have to do fundraising. And you have to make sure that, uh, you know, you, you improve your visibility or making-- you have KPIs, and you are achieving that. So in some sense, having this kind of a quasi-corporate culture in how we organize our research and how we do our a- academic activity, really helps us stay focused, as opposed to just doing the usual, you know, work-as-usual, uh, routine that normally sets into department. So the similarity is that, yeah, so you have to, you know, build a great team, you have to do some amount of fundraising, and then you have to keep your eyes on your deliverables. Of course, there's one other, uh, thing which is very common to startups and academia, and so I have a hundred percent guaranteed attrition from my workforce. [chuckles]

   25. SPSpeaker33:55

       Uh-huh.

   26. BRBalaraman Ravindran33:55

       So, [chuckles]

   27. SPSpeaker33:56

       Because your researchers will eventually-

   28. BRBalaraman Ravindran33:58

       Yeah, they have to-

   29. SPSpeaker33:58

       -graduate and leave.

   30. BRBalaraman Ravindran33:59

       Yeah, so one of their performance metrics is how soon they leave, right? [chuckles] So, so it's not that I can retain people for a long as, as I turn, turn them into faculty.
8. 35:19 – 39:05

   Robert Bosch Centre for Data Science and Artificial Intelligence

   1. BRBalaraman Ravindran35:19

      right? Then Bosch, the company, came to us and said, "Hey, we really love the way you are looking at AI as an engineering discipline, so let us-- we will come with you, right? So we'll fund you. So let's start this, uh, Robert Bosch Center for Data Science and AI." And that's where we did. So we had about two years of, uh, uh, the Interdisciplinary Lab of Data Science that kind of morphed into, uh, the Robert Bosch Center for Data Science and AI. And that was in two thousand and seventeen, uh, August, and since then, that has been in operation and has grown in size, right? We started with, uh, something like thirteen faculty members, a few of whom actually left the institute after that. They were-- they retired or they moved on to other things. So I think when the Robert Bosch Center, uh, started, we had about, uh, fifteen people, uh, from different people, and now it has grown to thirty-eight faculty members, uh, who come from fourteen different departments. We started with six departments being represented in the Robert Bosch Center. Now there are fourteen departments that are being represented. We have run, like, close to a hundred projects from the center. There are nearly two hundred and fifty papers that have been published from the center, and at least a couple of startups that we have enabled from the center. Uh, all of this has happened, so a lot of activity happening. Several hundred students have gone, gone through the center, they've graduated, and, uh-... it's been this fairly well-established now. Uh, and, uh, in fact, we have had people from other countries come and when they visit IIT Madras, "Oh, we have heard about this Robert Bosch Centre. Can we go and visit?" Uh, right. So it used to be CFI, now, but only IIT Madras is coming to Robert Bosch Centre as well.

   2. SPSpeaker36:48

      [chuckles]

   3. BRBalaraman Ravindran36:49

      Okay. We should talk about CFI.

   4. SPSpeaker36:50

      We'll talk about CFI-

   5. BRBalaraman Ravindran36:51

      All right

   6. SPSpeaker36:52

      ... but not now. [chuckles]

   7. BRBalaraman Ravindran36:52

      That, yeah, that is, that is the Robert Bosch Centre, and it's, it's kind of still continuing even though Bosch has decided to shut down their research operations, uh, in India and move back to Germany. So they're not funding that initiative anymore, but we are still, uh, active, right? And then the- as we also had this other center on the, uh, uh, systems biology, right? It's called iBSC. Uh, so iBSC looks at, uh, uh, you know, AI applied to biological systems, right? It could be gene networks, it could be proteins, so all kinds of 'omics studies. And then we are now looking at things like drug discovery and, and iBSC, and then, uh, drug design and discovery. And we also look at, uh, uh, microbiome research, AI applied to microbiome. In fact, some of our teams have even studied the, the microbiome in the space station, right? In the International Space Station, so they're part of the team that did that. Uh, and so that, that's another active, uh... The thing that roughly started around the same time as RBC Design, they're still going very strong. So we have people, uh, and that we are using that also to launch our AI CoE in healthcare now. Uh, that's going to come in. So a lot of people who are originally part of the iBSC initiative are going to be part of this healthcare initiative as well. And then, as we kept going, there are multiple other centers that were coming that I was not actively involved. See, once the Robert Bosch Centre started, right, it set a template for others to say that, "Hey, this is how we should build, you know, interdisciplinary research centers," right? In fact, a lot of the centers in the campus use the exact same governing structure, exact same, uh, yeah, you know, uh, uh, process for operation, uh, that, uh, the Robert Bosch Centre uses, right? Almost all the interdisciplinary centers on campus use that. Not just on campus. When IIT Bombay and IIT Delhi wanted to set up their interdisciplinary centers on AI, they came to us as our template: "Hey, we see what you have done with the Robert Bosch Centre. Here, we want to copy this," you know, and, uh, and in fact, IIT Delhi said, "Oh, we want to copy this. Please tell us all the mistakes you made so that we don't make them."

   8. SPSpeaker38:48

      I have a feeling that, uh, now that we have a department in AI-

   9. BRBalaraman Ravindran38:51

      Yeah. They are going to follow suit.

   10. SPSpeaker38:52

       Yeah.

   11. BRBalaraman Ravindran38:53

       Yeah, yeah, they've already talking to me about it. They're trying to get me some- get some inputs and things like that. So it's, it's, it's, it's been an amazing influence, right? Not just within IITM, I, I mean, across the country-

   12. SPSpeaker39:03

       Yeah

   13. BRBalaraman Ravindran39:03

       ... the, the Robert Bosch Centre has had an influence.

9. 39:05 – 42:15

   Centre for Responsible AI

   1. BRBalaraman Ravindran39:05

      So after that, in the last year or so, I have been, uh, more active in, uh, uh, setting up this, uh, Center for Responsible AI, right? So the reason is, uh, you know, AI is, is still a technology that is developing, right? We can do a lot of cool things with AI, but doing a proof of concept and then going into a, an actual field and deploying a product, there's a lot of difference. I mean, so people who have, uh, been through this journey understand that. Uh, so-

   2. SPSpeaker39:35

      Yeah, yeah, I, I remember that Google Gemini once suggested that, "Put glue on pizza to make it more cheesy."

   3. BRBalaraman Ravindran39:42

      Yeah. Uh, no, no, no. So, so somebody said, "Oh, my pizza-- cheese keeps sliding off my pizza. What can I do?" And then Gemini said, "Put glue on it so that it fixes it down." And the reason it did that is actually very interesting. It was not hallucinating it, it was not making it up. It actually pulled a search result and rephrased it, and somebody had written a prank article on Reddit about how to fix, uh, cheese to pizza, and Google just retrieved that prank art- the prank article and then thought it was truth-

   4. SPSpeaker40:13

      I mean-

   5. BRBalaraman Ravindran40:13

      And rephrased.

   6. SPSpeaker40:14

      Can it tell whether it's truth or not?

   7. BRBalaraman Ravindran40:16

      Um, good point. It can try to detect sarcasm, but sarcasm detection is one of the hardest problems to solve in-- automatically, right? So, in fact, I know many humans who don't get sarcasm. Forget about AI getting sarcasm, right? [chuckles]

   8. SPSpeaker40:31

      Yeah.

   9. BRBalaraman Ravindran40:31

      So, uh, so that's a challenge. Okay, that's, that's, that's an aside. But then that is a huge problem, right? So what I was finding, to a little bit to my worry, is that a lot of governments in India, a lot of state governments and even the central government in India, uh, were trying to push AI into, you know, public services, right? Looking at things like law enforcement and, and also other, other, uh, you know, uh, uh, public-facing services. And they were trying to use these AI models without even understanding what are the pitfalls, whether the AI is truly ready for solving the... right? So I got a little worried. I mean, we, we'll put ourselves into a major hole if we, we go that way, right? So I started pitching for the Center for Responsible AI, where the whole idea was to understand how to use AI in a correct fashion, right, a responsible fashion. As-- We should have called it Center for Responsible Use of AI. Doesn't quite have the same ring, right? [chuckles] So we, we just went with Center for Responsible AI. So that's something that I'm spending a lot of time with. And, uh, we have actually taken the interdisciplinary concept to the next level with Center for Responsible AI, because, uh, when you're talking about, uh, any technology interacting with humans, right, it's not just enough for the technologist to think about it anymore. You need, you know, a sociologist, you need economists, you need lawyers, you need a whole bunch of other disciplines, right? Who needs to think about it. I mean, so usually, uh, the complaint from other disciplines is, technologists look down, down their noses upon other non-technical disciplines. Uh, but then in the last year and a half, I have come to understand that everybody brings a very, very valuable, uh, you know, viewpoint to this discussion. So we are trying to expand the scope of CeRAI to get in, uh, uh, non-technical people into the conversation as well.

10. 42:15 – 46:47

    The world is interdisciplinary

    1. SPSpeaker42:15

       The interdisciplinary nature of modern life, um, is actually quite hard for a lot of people to comprehend, and we interact a lot at Ask IITM. We interact a lot with parents of students who are applying to, uh, IITs or to other colleges, and they have to make these choices between, well, chemical, meta-... CS, AI, CS at IIT Bombay versus CS at IIT Madras, um, or AI at IIT Madras. So in these, making these choices, they, they sort of struggle with the idea that the choice is not that strong a choice. Would you say that's true? Like, I think it, because of how interdisciplinary the world has become.

    2. BRBalaraman Ravindran43:02

       So the, the undergraduate discipline that you end up doing, right, influences the choices that you get later in life. Influences, doesn't determine, right? There's not something that is predestined. "Okay, oh, I didn't get into computer science in IIT Bombay. That's it, I cannot do anything in computers in my life anymore," right? That's not true, right? So people actually diversify quite a bit, yeah, right? I know I have a cousin who did, uh, chemical engineering, but now works with aerospace engines, right? So like, out to where, right? So, so you can actually... You know, the way that the, you know, the fields are moving now, the kind of resources that is available to you, you can move around, right? But the discipline that you get into does certainly prepare you for certain kinds of jobs. Yeah, right? So that is important. So if you wanna get a head start, right, to, to then to think about what is it that you want to learn, what is it that you want to do, and, and go for that, right? I would say that the student interest and the focus on what they want to learn, what they feel excited about, is more important in choosing a branch than thinking about, "Hey, what will be my, you know, career prospects 20 years down the line?" Because there are just so many imponderables that, uh, you can't say that for sure at this point, right? And in fact, the other day I was having a conversation with, uh, one of my school friends, and then suddenly he just woke up, "Hey, you did your master's in AI, right? And you're still working in AI. You're the only guy in my friend circle I know who's doing that." Okay, wow! [chuckles] You're actually doing what you wanted to, wanted-

    3. SPSpeaker44:38

       Yeah.

    4. BRBalaraman Ravindran44:38

       So that's how the field, that's how the life is, right? So people are, you know, learning one thing, they're moving on to doing other things, so we have to think about lifelong learning. So yes, so the choice itself, right, is not strongly in one way. It's just like, "You have to do this, otherwise you are, uh, you're, you know, you're gonna be a second-class citizen all your life," or something like that. No, that's not the case anymore.

    5. SPSpeaker44:59

       But a lot of, a lot of students do... I mean, you mentioned it earlier, uh, in this conversation, a lot of students actually pick up AI and then drop it because they feel like that's not their cup of tea.

    6. BRBalaraman Ravindran45:08

       Well, I said they pick up AI and drop it because the exams are hard. I didn't say they drop it because [chuckles] it's not their cup of tea. Uh, but, uh, sure, uh, uh, the people do change, right? So, I mean, in fact, I've, I've known people who have taken computer science, right? I mean, so I've, I've been a computer science, like, uh, professor, longer than I've been, uh, in the AI department. I, I've been, I've... Yeah, so longer than I've been in the AI department. I've seen people take computer science and then switch off in their second year because they decided that computer science is no longer for them, and they would rather do something else. I don't know, come, uh, build something with their hand and, or, or, or, or get into, you know, psychology or something else, right? Not necessarily into computer science. So sometimes this realization comes a little later in life. Some people know it early enough that that's what they want to do. So these things do change, right? But yeah, so, uh, but right now, AI is very hot, right?

    7. SPSpeaker45:55

       Yes.

    8. BRBalaraman Ravindran45:55

       AI is super exciting-

    9. SPSpeaker45:56

       Mm

    10. BRBalaraman Ravindran45:56

        ... right? And, um, so I have to, I mean, I will be remiss if I don't say this, right? I think, I think, uh, at this point, right, uh, uh, artificial intelligence is a transformative technology-

    11. SPSpeaker46:08

        Yeah

    12. BRBalaraman Ravindran46:08

        ... transformative field in some sense. It's not a single technology. It's not like steam engine, right? So artificial intelligence is something that is going to- that brings a suit of new ways of solving problems. It gives you n- new, new directions to look at, right? New ways of formulating problems, things that we couldn't even imagine a few decades back, that can be solved using automation are now, uh, being enabled, right? So it's an exciting time. So it'll be remiss if you don't learn about AI right now, right? Because that will certainly put you in a back foot, right? I mean, not that you need to learn how to, how to do AI, but you should learn how to use AI at this point. I think it's very important.

    13. SPSpeaker46:46

        Yeah, and, um,

11. 46:47 – 52:16

    AI and the Nobel Prize

    1. SPSpeaker46:47

       in our prep call, we were talking about the, the Nobel in physics and chemistry going joint to AI and traditional researchers in that field. That's a big nod, to say that it's become so interdisciplinary and there's so high a contribution of AI to, uh, those fields. I feel like I touched it all on.

    2. BRBalaraman Ravindran47:11

       Well, yeah, I mean, I, I mean, I'm super thrilled-

    3. SPSpeaker47:15

       Mm

    4. BRBalaraman Ravindran47:15

       ... that, uh, AI is getting the nod, uh, uh, for, uh, uh, for the Nobel, right? So as, uh, somebody was jokingly saying, this is, uh, a distinguished alum of IIT Madras, uh, Professor Subra Rao Kambampati, he was actually joking about this: "Oh, man, earlier, we only used to tell our students, uh, to tell our children to get the Turing Award if you're studying computer science. Now, if you're studying computer science, you tell them to get the Nobel also." [chuckles] So, so much more pressure on the Indian kids. So, so, [chuckles] so that, that is, that is there. So, but still, uh, uh, uh, but it's, it's, it's an amazing, amazing point, uh, in time, right? So, and, uh, this is not the first time, uh, uh, uh, uh, somebody known for his AI contribution is getting a Nobel. But, uh, certainly for the first time, somebody's getting a Nobel for contributions in AI itself. So Herb Simon has a Nobel, uh, which was- so even though he was one of the kind of founding fathers of, uh, uh, AI back in, uh, uh, back in the '60s, right? Uh, so '50s and '60s, his, his contribution- his Nobel was for something else, right? Uh, uh, not for his contributions to AI. But in this instance, both Hopfield and, uh, and Hinton, the physics, uh, Nobel laureates, have gotten their, uh, Nobel for their contributions primarily that came from AI. And, uh, the, the, the, the, the kind of the fig leaf the Nobel Committee offered to the physics community is that, "You know, don't worry, it's not just AI, because all of the things that these guys built in AI, we are using s- them in physics as well, and this actually caused us to do huge discoveries in physics."

    5. SPSpeaker48:51

       Yeah.

    6. BRBalaraman Ravindran48:52

       The way they thought about formulating the problem for-... solving using neural networks allowed us to, you know, solve other harder problems in physics, right? So that's something that they talked about. And, uh, the Chemistry Nobel is a little bit more straightforward, right? So it was, so the Physics Nobel was not actually shared. It, both, both the people who got it were AI folks, right? I mean, Hopfield is by training a physicist, but-

    7. SPSpeaker49:15

       He got it for his work

    8. BRBalaraman Ravindran49:16

       ... work in AI, and, and Hinton is, is, uh, got it for his work in AI as well. Uh, in, uh, the Chemistry Nobel, so Baker is a, is a chemist. Like, uh, the, the guy has been working on protein folding, protein structure prediction his entire life, right? And so he's not a computer science, computer scientist. While the, the other half of the Nobel went to people from DeepMind, who worked on this, uh, pro, uh, AI called AlphaFold, right? That used reinforcement learning to predict, uh, the structure of proteins, among other things. In fact, most, almost everything that DeepMind does has some reinforcement learning in it, right? So it's, uh, it's a reinforcement learning company, really. And so they, they worked on, uh, um, uh, this AlphaFold, and which actually made a significant contribution to how, uh, people looked at protein folding, right? So protein folding is a problem that people have been looking at forever, and it has a lot of different implications, right? It's not just, uh, a kind of an academic exercise, trying to figure out what the 3D structure of protein is. It allows you to use it, uh, yeah, and, uh, design drugs, understand various kinds of diseases, understand various kinds of biological processes, and so on and so forth. And, uh, and so it's a very, very important problem to solve.

    9. SPSpeaker50:27

       And, and AlphaFold has, I mean, transformed the field of protein folding-

    10. BRBalaraman Ravindran50:33

        Yeah

    11. SPSpeaker50:33

        ... entirely.

    12. BRBalaraman Ravindran50:34

        Yeah. It has, it has. So, uh, the thing is, when AlphaFold 2, right? Uh, AlphaFold 1 was fine, AlphaFold 2 was the one that really broke the back, right? When AlphaFold 2 came out, uh, people are actually saying, "Hey, has the protein folding problem been solved," right? And there are, uh, you know, uh, biologists and k- and, um, biochemists who are actually working on this problem their entire life, who said, "Oh, I'm going to retire now, there's nothing else left for me to do." And there are all kinds of downer articles written in these ma- uh, these journals, and so on and so forth. But it turns out it's not solved, right? So there are certain kinds of structures which AlphaFold still struggles with, right? Certain kinds of proteins AlphaFold still struggle with. So the, the, the focus of the field has changed-

    13. SPSpeaker51:17

        Yeah

    14. BRBalaraman Ravindran51:17

        ... right? But the biggest difference is the following, right? Earlier, we used to have databases of protein sequences. You don't know the structure, you just know the sequence of amino acids that form the protein. So people had published database of these protein sequences, which kind of kickstarted, uh, research in one direction. Now, with AlphaFold, we're able to publish a database of protein structures.

    15. SPSpeaker51:39

        Mm.

    16. BRBalaraman Ravindran51:40

        Right? So which completely, you know, changes the-

    17. SPSpeaker51:44

        Yeah

    18. BRBalaraman Ravindran51:44

        ... uh, set of problems that people can look at, and so on so. It did, it changed the, uh, field completely.

    19. SPSpeaker51:50

        We were talking about the interdisciplinary nature of, uh, AI, and in general, modern engineering and science, but definitely in AI, where, uh, AI, uh, people get into AI from all fields, uh, and AI gets into-

    20. BRBalaraman Ravindran52:04

        All fields

    21. SPSpeaker52:05

        ... all fields. Yeah. It's, it's interesting. It's almost like never before been so interdisciplinary.

    22. BRBalaraman Ravindran52:12

        Automatic.

    23. SPSpeaker52:15

        Yeah. So I,

12. 52:16 – 54:52

    Teaching the founder of Perplexity

    1. SPSpeaker52:16

       I want to, um, talk about... You were talking about, uh, how student paths are different, and, uh, it's- it'll be an amiss to miss the fact that one of the largest companies in the, in the AI space, Perplexity, was co-founder-- one of the co-founders is one of your students. And he talks about it. He talks about how he was an electrical engineering student, and landed up in the same machine learning, reinforcement learning courses. Um, it's quite exciting. How was he like in class?

    2. BRBalaraman Ravindran52:42

       Uh, so I, I, I mean, I'm speaking the truth. So [chuckles] he was very quiet in class, right? And, uh, he almost always dread when he opens his mouth because it's always, almost always to point out something that was wrong in the lecture. Uh, uh, but he was very quiet, and he, he, he didn't ask too many questions in class, and things like that. And, of course, he was either the topper or among the top two in every class, and he's one of those guys who come into the class and skew the grading curve, so that the rest of the class hates them, you know? You know the kind-

    3. SPSpeaker53:14

       Yeah, [chuckles]

    4. BRBalaraman Ravindran53:14

       ... yeah, yeah. So he was one of those. He'll come in and skew the grading curve, uh, one way or the other. But it's amazing. So he started doing research projects in his third year, and, uh, so we, we have been working... I got a good two years to work with him because of that, right? So, and then, uh, he's amazingly, uh, uh, uh, uh, persistent researcher. And, uh, even then, uh, he kind of had a knack of picking problems that, uh, you know, people will be interested in hearing the solution to. Right. And, you know, the Perplexity was not the first product that he built, right? He built something else completely different. His first product was something to search through tweets. There was no, there was no way Google Search actually hit the tweets, right? There was a lot of information that was being-

    5. SPSpeaker53:56

       Yeah

    6. BRBalaraman Ravindran53:56

       ... generated in the Twitter sphere. And so he built this thing that will allow you to go through that, and then pull out, uh, interesting, uh, threads of discussions from Twitter. Then slowly, that kind of, you know, grew into Perplexity AI now, and it's doing amazing stuff. Yeah.

    7. SPSpeaker54:14

       That tool is, um... I mean, at least for us, it's changed the way we look at researching or look at-

    8. BRBalaraman Ravindran54:20

       Yeah

    9. SPSpeaker54:21

       ... collecting information.

    10. BRBalaraman Ravindran54:22

        Yeah. So, I mean, Jensen Hong said that's his favourite search tool. In fact, he said something very interesting. He said: "Even if I'm don't have a search query, I like to go and see what Perplexity will tell me about some things." So I- he just, he uses it at least three, four times a day, just, just out of curiosity as to what Perplexity would tell, even though he know, he knew what the answer he wanted was. That's really changing the way people consume information.

13. 54:52 – 1:02:37

    Centre for Innovation Faculty Advisor: Facilitating students to BUILD

    1. SPSpeaker54:52

       Let's, um, switch topics for a bit. Professor, you were also the faculty advisor for CFI, and we are sitting in CFI. Of course, this is the new CFI.

    2. BRBalaraman Ravindran54:59

       Yeah.

    3. SPSpeaker54:59

       Uh, and this building is about two years old, and you were a faculty advisor from-... 2015 to '18-

    4. BRBalaraman Ravindran55:05

       Yeah

    5. SPSpeaker55:05

       - or '14 to '17, I think.

    6. BRBalaraman Ravindran55:07

       '15 to '18.

    7. SPSpeaker55:07

       '15 to '18. Um, how, how was CFI then? Like, I mean, j- just for a clarification, when I graduated, CFI was not there.

    8. BRBalaraman Ravindran55:16

       Yeah.

    9. SPSpeaker55:16

       It started, like, two months after I graduated.

    10. BRBalaraman Ravindran55:19

        Mm. Okay, we just waited for you to leave, Krishnaji. [chuckles]

    11. SPSpeaker55:21

        Yeah. [chuckles] Thanks. Um, yeah.

    12. BRBalaraman Ravindran55:25

        No, in fact, we are sitting exactly in the old CFI.

    13. SPSpeaker55:29

        The location?

    14. BRBalaraman Ravindran55:30

        Yes.

    15. SPSpeaker55:30

        Building is different.

    16. BRBalaraman Ravindran55:30

        Right, the building has changed. This is exactly where CFI was, right? And, uh, and CFI has always been something which, you know, pulls in the students who wanted to do things, explore outside their, uh, subject area. So it is always an exciting place, right? And, uh, in fact, uh, uh, I think I mentioned this to you earlier, I was the first faculty member to be designated as advisor of CFI. Before that, the Co-Curricular Affairs advisor faculty used to take care of CFI as well, right? But then I came in and I said, "Look, CFI has grown so big," right? There's so many interesting things that are happening, there are so many projects that are going on, so many different clubs, and, and thing. And so it needs a dedicated advisor. You cannot have somebody who's taking care of all the co-curricular activities on the campus, also take care of CFI, uh, but it needs somebody who's going to do this. And then I said, "I can take up CFI, but I don't want to do the co-curricular affairs thing," right? So, so then, so we split that, right? So, uh, uh, so Rajan became, uh, the, uh, Co-Curricular Affairs advisor, and I became the CFI advisor. So that's, that's the point, right, literally the cusp when CFI was blowing up, so it was, uh, exciting times. And then is what we... When we decided to build this new building. So in fact, I was the CFI advisor when we designed this building-

    17. SPSpeaker56:46

        Mm

    18. BRBalaraman Ravindran56:46

        ... when this building was commissioned, right? Uh, and then I stepped down, and, uh, the next advisor took over.

    19. SPSpeaker56:52

        I'm curious, um, we, we talk of CFI, like, I, I mean, sometimes I feel like it's taken for granted, right? Like, it's there, everybody has access. And we're talking about all these entrepreneurship, and you've just mentioned, like, we started this center, we started that center, this center, and then it's just like, I sometimes sit back and wonder, um, 20, 30 years back, maybe, maybe it was like this, uh, at a smaller scale. But today it feels like there's this massive build culture, uh, across, uh, IIT Madras, different departments, different student clubs, and everybody's talking about it, and if they're not building something, they're supporting someone else who's building something. B- what is, uh... And what changed? What is in the water?

    20. BRBalaraman Ravindran57:40

        The, uh... I think it's, it's something in the DNA of IIT Madras, right? Remember, we started, uh, with the collaboration of the Germans, right? And Germans love to build, right? Germans are this-- guys are so focused on engineering and building things, right? I think that somewhere got stuck in our DNA, that we also, you know, are, are building this entire ecosystem and this culture that encourages people to actually become true engineers and true builders, as opposed to just studying something, you know, and from the books and then going away, right? And as you rightly said, most of these things wouldn't have been possible if the, the support has not been coming from top down, right? So for some of the initial activities that we did, including starting CFI, we needed the support of the then director, Professor Ananth. He backed it big time, right? So he said, "Okay, yes, go ahead. Let's do this." Right? And in fact, Ananth was envisaging, uh, uh, uh, uh, ecosystem, where if you just go into, you know, a cafeteria, you should find that, like, uh, 10 different tables discussing 10 different exciting projects, and you should, you know, get motivated by that, go join one group and be able to do stuff, right? He never believed in this artificial boundaries of, uh, departments and subjects, and things like that. He-- So he was just... He had this vision about, you know, uh, uh, uh, an IIT, where there are no departments, and people were just, you know, teaching subjects, and the students used to can, you know, put together their curriculum based on what they are interested in learning. And research would happen in a very, very fluid, interdisciplinary manner, without worrying about, I don't know, departmental boundaries, administrative boundaries, and things like that. So we have had-- We've been lucky to have a series of these kind of visionary leaders, and Bhaskar was amazing, right? So when Professor Bhaskar was the director, he again pushed so many, so many different, uh, directions, right? And then he prepared the institute for what happened when Kama came out, right? When Professor Kamakoti took over as director, he created two new and completely interdisciplinary departments, right? So Medical Sciences and Technology department, again, pulls in a lot of people together. And the goal is not just to, you know, do, uh, uh, um, you know, teach people, right? His goal is to actually train guys so that they can go make a difference in the field, like, who design new, new equipment, new, new ways of, uh, you know, doing, uh, uh, health tech, right? So that's the whole idea behind that, right? Likewise, the Data Science and AI, uh, department was again enabled, uh, so that, uh, the focus from the department, right? In fact, if you look at our curriculum also, is a lot of hands-on components have been inserted into the curriculum, as opposed to what would be a traditional, uh, you know, just run a few code, you know, programs and things like that. But we're actually having a lot more industry engagement in the curriculum itself. So all of these things have been possible because there's been this culture of innovation and building in IIT Madras.

    21. SPSpeaker1:00:21

        And maybe we can trace it back to the German collaboration.

    22. BRBalaraman Ravindran1:00:23

        Maybe. Maybe. That's, uh, that's speculation, but, uh, it certainly has been true for a long time now.

    23. SPSpeaker1:00:29

        I, I feel like you're being humble, because obviously, professors like you and directors like, um, Professor Ananth and Professor Bhaskar, and Professor Kama, have had a lot to contribute.

    24. BRBalaraman Ravindran1:00:39

        Yes, yes, yes, they did. That's what I'm saying, right? The, the drive comes from the big- from the top, right? So I'm, I'm, yeah, so I'll give a lot of credit to the admin.

    25. SPSpeaker1:00:47

        Nice.

    26. BRBalaraman Ravindran1:00:48

        And of course, I, I would be completely remiss if I don't mention the Research Park.

    27. SPSpeaker1:00:52

        Yeah.

    28. BRBalaraman Ravindran1:00:52

        So which was started during Professor Ananth's time, right? So he really wanted what, uh, what happens in IIT Madras to have a real life impact, right? And the way to do that is to connect closely with the industry. The Research Park has enabled that big time, right? So a lot of st- the, the reason we have more startups coming out of IIT Madras than any other IIT, is because of the access to the innovation ecosystem in the Research Park.

    29. SPSpeaker1:01:13

        Yeah.

    30. BRBalaraman Ravindran1:01:13

        Right. And of course, uh, uh, Professor Jigjasa has been, uh, running that for, uh, forever, uh-
14. 1:02:37 – 1:04:23

    Personal Reflections: Early years as a student

    1. SPSpeaker1:02:37

       Uh, so can we, um, go back to a few things that you had mentioned, uh, to me earlier, uh, about, um, how you were as a student and, uh, your journey to IIT? And I remember you mentioned, um, that, uh, y- you were at ISC for a while, and y- you were running a forum in the early '90s.

    2. BRBalaraman Ravindran1:03:01

       There's some page, yeah.

    3. SPSpeaker1:03:03

       What is it?

    4. BRBalaraman Ravindran1:03:04

       Okay, so I got fascinated by the web, right? So very early on, right, even though, uh, we didn't have a lot of internet access in India. So in fact, the first time I s- sent email was in ISC in the, uh, when I was in my master's already, right? Before that, we didn't even know something like email existed. And, uh, back then, if you wanted to get a copy of a paper, we write to the authors, like write, paste, mail, and then the authors will mail us a copy of the paper, right?

    5. SPSpeaker1:03:32

       Okay.

    6. BRBalaraman Ravindran1:03:32

       So it'll take us about a month to get it, right?

    7. SPSpeaker1:03:34

       Okay.

    8. BRBalaraman Ravindran1:03:35

       Uh, and so that's how, how it used to be, right? And then this whole, whole web came in. It was a new concept back then, right? And not too many people in India actually had access to, uh, the internet. Even if you had access to email and things like this, there's the web going out and looking at the World Wide Web was not a thing. Uh, but I got fascinated by this, right? So, uh, it started off with me experimenting with this stuff and, uh, maintaining what I called a food page back then. It's like a, what in today's terminology would have been called a blog, but the word blog hadn't been invented back then. So I used to write about restaurants, uh, restaurants, review them, and write about recipes and things like that. And then I give- used to give an email address on my page. If people wanted to submit recipes, they could do that. I actually got recipes from people I never knew on how to cook certain things. And I thought, "Okay, this is a great way to connect

15. 1:04:23 – 1:09:51

    Going viral before it was even a thing

    1. BRBalaraman Ravindran1:04:23

       with people." So what we did, then a group of us who were very, very interested in, uh, film music, right? All, uh, Ilaiyaraaja fans, right?

    2. SPSpeaker1:04:30

       Such a different time, right? I have heard from someone that, uh, early '90s stories, that they saw a recipe for pani puri.

    3. BRBalaraman Ravindran1:04:38

       Uh-huh.

    4. SPSpeaker1:04:39

       Uh, and, uh, they made a puri, like a normal puri.

    5. BRBalaraman Ravindran1:04:41

       Yeah.

    6. SPSpeaker1:04:42

       And [chuckles] they stuffed it like how a pani puri would be, but it's totally different puri, right?

    7. BRBalaraman Ravindran1:04:46

       Yeah, yeah, yeah.

    8. SPSpeaker1:04:46

       And you didn't, you didn't know that because you grew up in the south, and-

    9. BRBalaraman Ravindran1:04:49

       Yeah

    10. SPSpeaker1:04:49

        ... that's a north dish. It's very interesting.

    11. BRBalaraman Ravindran1:04:52

        Yeah, so it's, it's, it's, it's, it's different times. So basically, we started this, uh, TFM page, Tamil Film Music page. And back then, we didn't have enough bandwidth to host it in India, so we used to do all the things. We made the, uh, you know, the HTML files, and then we sent it off to a friend who was in Oxford, and he used to host it in his, uh, machine in Oxford, even though it was a TFM page. Uh, but it used-

    12. SPSpeaker1:05:16

        But you were emailing and not posting it?

    13. BRBalaraman Ravindran1:05:18

        We were emailing it. Uh, email, email is fine, right? So email was happening then, right?

    14. SPSpeaker1:05:21

        Yeah.

    15. BRBalaraman Ravindran1:05:21

        It's like that's a custom. Uh, so we, we, we used to send him these, uh, uh, uh, things in email, and then he used to post it then, uh, in his, uh, lab webpage, until a point when the lab basically said, "Hey, you're, you are increasing the traffic to our, uh, web servers, uh, manifold," right? "So everybody's coming to see your music page, not, uh, the academic, uh, content on the web pages." So they asked us to take it down, and at that point, we started hosting it, uh, in a service in Singapore. But it was running strong till about two years back, so-

    16. SPSpeaker1:05:53

        Oh, really?

    17. BRBalaraman Ravindran1:05:53

        Yeah.

    18. SPSpeaker1:05:54

        Oh, okay.

    19. BRBalaraman Ravindran1:05:54

        So we were used- we used to get something like... I think at the peak, uh, we used to get something like fifteen thousand, twenty thousand visitors a day.

    20. SPSpeaker1:06:02

        Oh, wow!

    21. BRBalaraman Ravindran1:06:02

        To the page. It, it became very popular. In fact, uh, we used to partner with, uh, radio programs in Malaysia, and so they'd say, "Oh, go to this page and, you know, give us your, uh, songs, and we will play them," and here, and we used to have a lot of amazing articles that were written-

    22. SPSpeaker1:06:17

        Mm

    23. BRBalaraman Ravindran1:06:17

        ... uh, uh, uh, uh, by collaborations on this, right? So one of the things that I did, uh, as part of TFM page was create-- I mean, I wrote all the entire back end for creating a discussion forum. At, at that point, there were no, like, bulletin board services or all these, uh, chat services that you could just post, right? I actually wrote the entire, uh, back end and, and, and the front end. I mean, so all of that. And, uh, we started something called a discussion forum, right? Uh, and then, uh, in fact, it's a funny story. So I, I started the discussion forum, and then I came back to India to get married. So I, I had no access to this. This was after I had gone to UMass, right? So this was in, uh, like '96, '97, mid-middle, middle of '97, we started this, uh, discussion forum, and then I came back, I got married, and I go back, and I find that thing had just caught fire, right? There were like, like hundreds of users on that, uh, without any, I mean, without any publicity or anything. The people just discovered it, and then they just come there, and there's, like, hundreds of users, like, several, uh, topics that are being discussed. In fact, there are thousands of users, right? And there are a lot of topics that are being discussed. So then I had to go in there and do a lot of hacking at the back end to make sure the system can scale to this large, uh, user base, right? And that was so interesting. So we said, "Hey, look, so many people are interested in a forum to discuss Tamil music, let's start a forum-... like, just like a hub, right? So the word hub comes from, uh, Arthur C. Clarke. So he used the word hub to describe certain kinds of, uh, you know, space, uh, colonies that were getting formed and things like that in his- So the, the Arthur C. Clarke has his Rama series of books. You know, Rama I: Garden of Rama, Rama II, and things like this. Like, a, a, a, a splinter, uh, fraction of humanity gets launched into space, and then they build this whole culture in space and things like that. So he- the main thing there is called the hub. So I said, "Okay, the hub is where everybody's going to come and start discussing stuff," and so we called some- we built something called the Forum Hub, and it became more popular beyond my imagination. There was, like, so many different topics. We, we allowed people to, you know, self-form. It was really one of the earliest social networks out there, right? So people were coming in, they were forming, they were forming their own groups there, right? And, and then, uh, they, they used to discuss various topics, and there are multiple, for example, cooking, cooking sites that, that went professional later. They had their initial, uh, start there. So this guy started posting, he was called Hemant Trivedi. He started posting recipes on this, uh, forum hub, and then people just loved what he was doing, and then he went and joined- launched his own, uh, web page later, right? It became very popular. It's... I think it's still running. So I like that. Uh, we, we wrote, uh, we had some, some of the earliest collaborative writing, fiction writing, that was happening there. So people just got together and say, "Oh, let us try and write a book together. Let's the 10 of us write books together." And then they was doing this. And there was another, uh, person who was actually- we didn't figure out until much later that she was actually an IAS officer. Uh, [chuckles] but she started writing fiction on our forums, right? Became so popular, then she got encouraged by the responses she was getting on the forum, and went out and started publishing books on her own. And, and of course, there, there are people who got married [chuckles] because they got introduced on this forum and things like that. So it, it became super successful. We're very happy.

    24. SPSpeaker1:09:26

        I see what you mean by saying that it was like an early social network kind of thing. Um, yeah, I mean, how far we've come, right?

    25. BRBalaraman Ravindran1:09:32

        Yeah, after I got interested, more interested in RL, and so I stopped pushing that. I mean, if I had got... There are people who actually offered to buy the site-

    26. SPSpeaker1:09:39

        Oh-

    27. BRBalaraman Ravindran1:09:39

        ... we, we, we refused to take it up.

    28. SPSpeaker1:09:41

        So far we've come from then, where email was new, and forums were new, and RL was a-

    29. BRBalaraman Ravindran1:09:48

        Yeah, man, I'm old. Yeah.

16. 1:09:51 – 1:21:21

    How the AI-human relationship is evolving

    1. SPSpeaker1:09:51

       I want to ask you, [laughing] I want to ask you about where are we going in the next five, 10 years, because I think it's-

    2. BRBalaraman Ravindran1:09:58

       So we're taking this human connection, right, for granted, right? And also at the same time, we are evolving a new kind of kinship, right? So I, I used to think my son is not being very social, because he used to sit in his room all the time. And then I find that he has way more friends now than I did, and I mean friends-friends, not Facebook friends. I mean, the guys who know what he's doing, what he's thinking, guys he is sharing stuff with, and, and all of this interaction is happening online, right? So he had a roommate during his college days, right, and, uh, they met online, and then they managed to live together for their entire, uh, college time in the same, same apartment, sharing one bedroom, right? And they had no problems at all. And he didn't know him from Adam before he started taking classes, and they met complete... Uh, his first year was COVID, right? So nobody met each other, so all the students met only online, but they figured out how to work. So there are kind of kinship and other things that are building, right? So the way people build relationships, I think the boundary between online and offline relationships is kind of blurring. I mean, there are-- I have friends who I have never met-

    3. SPSpeaker1:11:12

       Yeah

    4. BRBalaraman Ravindran1:11:12

       ... who I've met, who I've, I've only online, but, but that thing is... But, but, so some of- one of my best friends now is somebody I interacted for five years only online.

    5. SPSpeaker1:11:23

       Right.

    6. BRBalaraman Ravindran1:11:23

       I knew him from '97, and met him for the first time in my life in 2003. Right? So and now we's, we're still friends, and we keep meeting each other. We-

    7. SPSpeaker1:11:35

       I get what you're saying.

    8. BRBalaraman Ravindran1:11:36

       That is a, that was a one-off thing, right? But now what is happening is that is kind of becoming the norm now, right? So the way the society is going to evolve, we're going to change significantly. Boundaries are going to become less critical, right? And of course, unfortunately, it has all these other side effects of people consuming everything from the internet and, uh, and, and all the fake narratives and other things that are being peddled. And the other major concern for me is the hyper-personalization that is happening.

    9. SPSpeaker1:12:05

       Yeah, I heard this term called, uh, echoverse.

    10. BRBalaraman Ravindran1:12:09

        Okay.

    11. SPSpeaker1:12:09

        Um, which sort of gives the idea that we live... Because the algorithms around us work like that, we live- we seem to be living in this sort of bubble of our own making.

    12. BRBalaraman Ravindran1:12:20

        Yeah. Yeah, that also could happen. But the bottom line, right? So societal structures, as we know it, are changing rapidly, changing around us, and some of us notice it, some of us don't, and, uh, so most of us are just going along with the flow, which is usually the way it is, right?

    13. SPSpeaker1:12:39

        Yeah. And, and, and what about, what do you, what do you think-

    14. BRBalaraman Ravindran1:12:42

        I mean, I mean, I can- on, on bad days, I can see the dystopian [chuckles] version of this. On good days, I, I, I'm much more hopeful for us becoming, uh, more of a global community and not...

    15. SPSpeaker1:12:55

        What about how the future of work looks like with AI coming in?

    16. BRBalaraman Ravindran1:13:00

        So AI is going to affect every aspect of life, so it's, it's going to change how, uh, how we are going to do the job, right? I mean, there was some technology in the past that, uh, we, we are now saying that we can't do without, right? So like that, right? So I like to give this analogy of, uh, how, uh, we use, uh, office tools. All right? So everybody needs to know something about either a document processing tool, or a spreadsheet, or the slide maker, whatever be your, uh, choice, right? You could use Google, or you could use Microsoft, or whatever, but you, you need to know-

    17. SPSpeaker1:13:36

        Yeah

    18. BRBalaraman Ravindran1:13:36

        ... all of these tools.

    19. SPSpeaker1:13:37

        You, you have to excel at it.

    20. BRBalaraman Ravindran1:13:38

        You have to excel at it, right? So in fact, our, uh, most MBA programs start off by teaching you how to use Excel effectively.

    21. SPSpeaker1:13:45

        Yeah.

    22. BRBalaraman Ravindran1:13:45

        Anybody can go and start using a spreadsheet, fill in numbers, and use the rows and columns, right? But if you really want the true power of Excel, you need to go train.... on Excel. So AI is going to become like that, right? So for everybody to do their job, they need to know what is the AI tool that they have to use in their job, right? It's not like, uh, uh, it's gonna- AI is gonna com- take away your job or anything, unless it's a small number of jobs that will get lost, right? I mean, so who, who runs an STD booth anymore, right? So we, STD booths-

    23. SPSpeaker1:14:13

        Oh, my favorite example of jobs that have been lost in the march of technology, uh, in the 1700s and 1800s, there was a job for someone to go and light all the lampposts which were-

    24. BRBalaraman Ravindran1:14:24

        Yeah, yeah

    25. SPSpeaker1:14:25

        ... running on oil.

    26. BRBalaraman Ravindran1:14:26

        Yep.

    27. SPSpeaker1:14:27

        So obviously that, that's not a job, that's-

    28. BRBalaraman Ravindran1:14:29

        That's gone. But, uh, you don't have to go to 1700s, 1800s.

    29. SPSpeaker1:14:32

        Right.

    30. BRBalaraman Ravindran1:14:32

        I'm talking about telephone booths.
17. 1:21:21 – 1:23:14

    Artificial General Intelligence is a misnomer

    1. SPSpeaker1:21:21

       Professor, no conversation with, uh, somebody as involved in AI as you are, is complete without talking about AGI. Quickly, what are your thoughts about AGI? Is it going to like-

    2. BRBalaraman Ravindran1:21:30

       See, I think AGI is, is, is, is a misnomer, right? So there is nothing called general intelligence, right? So you could be great at, uh, you know, being a marketeer, but, uh, you could suck at, uh, chess.

    3. SPSpeaker1:21:43

       Which I do.

    4. BRBalaraman Ravindran1:21:45

       So that doesn't mean... So AGI, people seem to think that when they talk about AGI, it means there's one agent that solves, uh, everything, right? So that's not quite the way it is, right? So the, the target, at least of the- technically, the target that people shoot for when they talk about AGI, is saying that, uh, a machine that can learn to solve problems that it has not been exposed to and not been trained on before, not instances of problems, okay? But problems themselves, that they have not been trained on before. Instances of problems is fine, right? So, uh, uh, uh, so that is what they call as AGI. But even then, right, there are some things that, you know, a human would struggle to learn, right? I mean, so I might be good at learning something, but from the life of me, I know I can't learn music, right? So, uh, uh, uh, uh, so those kinds of, uh, uh, challenges are there, right? So in fact, there is nothing out there in nature that really is, uh, uh, a general intelligence, uh, uh, engine or tool or creature that we can hope to emulate when we are doing these, right? Of course, it's good to have an aspiration to be something better than anything biology has designed. Uh, but, uh, we are a very long way from achieving anything close to what people mean by AGI. And of course, AGI itself is a bad term, but some people could argue that AI itself is a mistake.

    5. SPSpeaker1:23:09

       Hmm.

    6. BRBalaraman Ravindran1:23:10

       AGI is certainly much more so.

    7. SPSpeaker1:23:12

       Hmm.

18. 1:23:14 – 1:30:50

    How they put together the curriculum for AI and Data Analytics

    1. SPSpeaker1:23:14

       I cannot let you go-

    2. BRBalaraman Ravindran1:23:15

       You can't?

    3. SPSpeaker1:23:16

       ... before I ask you this question.

    4. BRBalaraman Ravindran1:23:18

       Oh, okay, I know what's coming. Yeah.

    5. SPSpeaker1:23:21

       IIT Bombay CS [chuckles] versus... No, I'm asking you this question because, uh, this was the first year that the BTech in AI was available.

    6. BRBalaraman Ravindran1:23:30

       Yeah.

    7. SPSpeaker1:23:30

       And it's clear from the opening and closing ranks, this is one of the most sought out of- sought after courses. So what should a student who scores JEE and got a really good rank, really look at?

    8. BRBalaraman Ravindran1:23:44

       Let me just step back and tell you about how we put together the BTech in ADAR, right? AI Data Analytic, right. So there is so much that needs to be taught for you to do AI well, AI properly, right? There's a lot of math background that needs to come in, right? And even from the computing side, I mean, you need to look at data structures in a different way. You need to look at algorithms in a different way and things like that. So we just felt that you cannot take the computer science curricula, right, and then add a few electives or add a few more courses to it and make it into an AI curricula. Right? There will be still some things that you learn without the proper grounding, right? So we kind of designed the whole curriculum from scratch. So this is not a CS plus AI curriculum. This is a purely ground-up designed AI curriculum. So there are certain topics in computer science that is, well, like, standard in a computer science curriculum, which you would not learn in this AI curriculum. That is because you're spending a lot more time on doing AI correctly, right? So if you're going to look at people getting into AI, core, hardcore AI jobs in the future, right, data science and AI jobs in the future, I think the ADAR curricula will train you to be better AI scientists and AI engineers than a CS curriculum, where you are adding things on top. That is the goal that we have designed, right? And going forward, in the future, I'm not going to say that CS guys or, I mean, mechanical engineers or chemical engineers cannot do AI jobs. I mean, obviously, right? I mean, look at the IT company now, right? IT industry is, uh, has, a small fraction of it are computer scientists.

    9. SPSpeaker1:25:19

       I mean, look at Arvind, he's electrical engineering.

    10. BRBalaraman Ravindran1:25:22

        Close enough. So... Close enough. We can talk about other people, right? So who are not even computer scientists, or, like, not even in the circuit branches, right? People from mechanical engineering who are doing great, right? So things like that. So there are a lot of, uh, people from a very, very different background who are now at the top in computing and other places. So I think, I'm not saying that you cannot move, right? It's, it's ob- obviously, it's not to say that if you do CS, then you will not be fit for AI jobs, right? I'm saying you'll be able to do the AI jobs better off the bat. So that's, that's basically what we have done with the curriculum. And as the fields are moving, right, you're going to see that you're going to get more and more jobs that require you to be good at AI.

    11. SPSpeaker1:26:03

        Right.

    12. BRBalaraman Ravindran1:26:04

        And right, so then eventually, uh, this AI curriculum would be preparing you for that.... as supposed to a pure com- computer science curricula, right? I mean, there's always this argument to be made: "Okay, if I do electrical engineering, can't I still do computer science jobs?" Yes, you can. Like Arvind, who went on to do a PhD in computer science, right? But then maybe if you had done computer science, you are better prepared to... In fact, actually stepping back, if you do a pure computer science curriculum, you're probably not completely equipped to do research in AI, because computer science curriculum doesn't teach you the right math. They teach you discrete math, and they teach you, uh, s- other things, right? They don't teach you the probability theory, optimization, and other things that you need to do AI, uh, uh, better, right? So linear algebra and optimization that you need to do AI better. So, so we teach all of that. We, in fact, take... Uh, and if you squint your eye and look at it, we look 20% like an electrical engineering curriculum, [chuckles] and another 30% like a computer science curriculum. So we kind of pulled in all the subjects that you need from different things and created this.

    13. SPSpeaker1:27:05

        And you, of course, you were a CS professor for-

    14. BRBalaraman Ravindran1:27:08

        I didn't study computer science.

    15. SPSpeaker1:27:10

        But you were teaching-

    16. BRBalaraman Ravindran1:27:11

        I know

    17. SPSpeaker1:27:11

        ... in the CS department.

    18. BRBalaraman Ravindran1:27:12

        I'm just saying that I, I'm an example.

    19. SPSpeaker1:27:14

        Oh, right, right.

    20. BRBalaraman Ravindran1:27:14

        From outside computer science anyways. Yeah, anyway, go on.

    21. SPSpeaker1:27:18

        Damn cool. Um-

    22. BRBalaraman Ravindran1:27:20

        You, you, you're asking a question about me being a CS professor?

    23. SPSpeaker1:27:23

        No, I was saying that, um... I mean, it's confusing for parents and students because, um, on one hand, it's interdisciplinary, it's a little flexible, it's not hard code. Uh, on the other hand, the branch is, as you said earlier, it's a step up. It's like a directional, uh, idea, and if you want to go deep into it, you should move straight into it. So they have to sort of figure out whether they want to be, uh, 50% committed to CS, 50% committed to AI, or whether they want to commit at all, and commit later in life. They have to also balance it with what they like to do in class 12, and in class 12, everybody just likes to fly planes. I mean, think about aerodynamics and, uh, rockets.

    24. BRBalaraman Ravindran1:28:07

        Really? Okay. My son wanted to do physics, but-

    25. SPSpeaker1:28:09

        Is it?

    26. BRBalaraman Ravindran1:28:09

        Yeah.

    27. SPSpeaker1:28:10

        Okay. Interesting. I wanted to build robots.

    28. BRBalaraman Ravindran1:28:13

        Yeah.

    29. SPSpeaker1:28:13

        Um, because, you know, it's like, it's moving and stuff.

    30. BRBalaraman Ravindran1:28:16

        Yeah. Sure, sure. So le- let me put it this way, it's not like you take an AI curriculum, right? I mean, you do, do B.Tech in ADA, you become unemployable for computer companies, right? So you, you are doing... You, you are going to do a lot of programming. In fact, you are going to do a lot more heavy-duty programming than most branches. In... I would actually even go to the extent of saying, including computer science. Uh, you're gonna do a lot of coding, right? And you're gonna do coding using AI as well, right? Which is how the field is itself is evolving, right? So you would be r- ready for, like, 80% of computing jobs as well, right? It is not that you do B.Tech in AI, that you are committing to doing only AI, not, not CS. So at least for the next few years, when, when we see, uh, IT jobs are still going to be more than the AI jobs, right? But not... Well, by the time these guys come to placements, right, when they are graduating, right, I think the balance would be more in favor-

Episode duration: 1:30:50