#336 Professor Mausam: Why India Is Losing the AI Race and What It Will Take to Catch Up

0:00This is the second episode in my series on India. As I said earlier, India and China were roughly equivalent a couple of decades ago, particularly in their research ecosystems. China took off and India did not. As a result, India, one of the largest economies in the world, with one of the largest populations in the world, and with a chain of terrific technical institutes, the India Institutes of Technology, has fallen far behind the U.S. and China in its AI development. I wanted to know why, so I went to India and spoke to some of the senior researchers at the Indian Institute of Technologies.

0:45Today, I speak with Professor Mausum. His mother blessed him with a single name, so Professor Mausum at IIT Delhi. Professor Mausum is one of India's leading AI researchers and is particularly articulate in explaining why India has fallen behind and what it has to do to catch up. Together, we delve into the contrast between India and China's approaches to AI, the challenges facing Indian academia, the various government initiatives, and what's needed to foster true innovation and talent retention in India. India has the opportunity to become an AI leader. The question is, is it starting too late?

1:34With that, here's a word about our sponsors, and then we'll speak to Professor Malsam. Yeah, can you give us that background, and then we'll start talking about your research and about AI and India. I started out, I mean, as an undergrad student at IIT Delhi, so this is also my alma mater for my undergraduate institution. after completing the course I wanted to do a PhD and so applied you know to the top places in the US and UW was kind enough to take me in interestingly I didn't apply to be an AI student at the time I applied to be a theory student but as I did my first AI course in the first quarter at UW and I just loved it I loved it beyond belief and I was really enamored by it at the time.

2:24This is what year? This is 2001. So very early in the AI. So, I mean, we'll talk about that, but AI is a 70-year-old on paper. Phenomenon. There was AI before that, but AI as a term got coined, believe it or not, just a few days back, 70 years ago. So 31st August, I think. That's right. Ending 55. That's right. Right at the Dartmouth AI workshop, a large field. And maybe if it was more recently, I was going into as a grad student maybe I wouldn't have taken AI but at the time it enamored me because there was a lot of things to be to do the philosophy was very exciting to me I mean we can talk more about that but so after finishing that I was looking so I wanted to come back to India but I was also looking for one stint in the US before I come back to India and it's a pretty stroke of luck I would say that UW my own PhD institution decided to offer me a research faculty position in the same institution right as I finished my PhD no so it was almost like one day I was a PhD student and next day I was a faculty member in the same institution again something I wouldn't recommend to others because it takes a lot of time for you to start feeling like a faculty member yeah because nobody thinks you're a faculty member right everybody knows you as a student.

3:42But anyway, so I was there for another six years after finishing my PhD as a research faculty. Actually, that is the time my profile started becoming stronger because I had the opportunity to really work with really amazing grad students at UW. And these are like, they pushed me in different directions. And I also had my advisor and another senior professor as people I was co-advising these students with. They were my mentors in some ways. And I was learning how to mentor through this kind of a relationship. So when it was time for me to come back to India in 2013, I had been a faculty member for six years.

4:20I had graduated as a student or two and many other students were about to graduate with their PhDs under my supervision or joint supervision. And so that is the time I sort of went broadly in the country. But my family is from Delhi, so it was most natural for me to come to IIT Delhi. And so I've been at IIT Delhi from 2013. yeah just to tell you my sort of path from then on that was also the time when ai was coming into some kind of you know broad visibility right for want of a better word there was a lot of energy about deep learning before 2013 up to 2012 mostly most of ai was probabilistic bayesian right that kind of ai and around 2011 2012 2013 all these initial results started coming about word to VEC about how you can do some, AlexNet had just come out, you know, all of those things were happening.

5:14So when I moved back to India, suddenly not only did I have to establish a research career in a new place in India, and sort of the standard practices I would be able to follow in Seattle may not necessarily apply here, but also the field was changing on me, right? So we had to deal with all of that change and disruption, but then over time things stabilized and we picked up new stuff and almost every researcher had to undergo some kind of a transition. That really gave us a lot of value because now suddenly AI became a great world and people wanted to do PhD or do research in AI. We started attracting strong students.

5:51So ever since the profile of AI researchers in India has sort of grown up, you know, just by market forces. And at some point, pretty much every university, a top university decided that, look, AI can develop within the computer science and electrical engineering departments, but we need to establish sort of separate units so that AI can really flourish. And so various institutes, including IIT Kharagpur, IIT Madras, IIT Bombay, and also IIT Delhi, started units, academic units within AI. So IIT Delhi started a school of artificial intelligence, later became Yardi School of AI because Yardi became a primary donor.

6:30and I was asked to lead the first as the founding head. Was that the first in India or the first in Delhi? I would say that this is around the time that everybody else who is our peer was doing similar things. So I don't know whether it was a few months earlier, a few months later, but I wouldn't take credit for being the first. And what year was this? This is 2020. I see. I would say that IIT Madras was a little bit of a leader there because they had established a center of excellence, I believe, in the Robert Bosch Center of Excellence in Data Science and AI. So they, I think, did a year and a half earlier than most of us.

7:08Everybody else had followed around the same time. And there was also IIT Hyderabad, I think. They also did something first. And that's about it. So that's been my journey so far. Last year, I was on a sabbatical. I spent a year in New York. I worked for Bloomberg. where I wanted to understand how in the modern world, AI is getting productionized because in the research field, in academia, we don't always think about products. We don't always think about exactly how it shows up in the end user, right? And so Bloomberg gave me that kind of an exposure. So I've just come back from the article and great to be talking to you.

7:44Yeah, and one of the things I'm interested in is, I mean, that's interesting that you came back right after Alex Nutt. So that was the first neural network supervised learning sprint. And then by 2020, when you were forming that school, I think were GANs out by then? GANs were much earlier. It was transforming to generative. Yes. So you've been here teaching through all of that. Yes. As I said, I spent a lot of my adult life in China, and I watched China go from nothing to being a competitor of the United States in AI. And in China, I always wondered, where's India? Because, turns out, engineers at a pace comparable to India.

8:39But India has some basis in English language. It has a much, until President Trump, a much easier relationship with the United States. Yeah. And so that's one of the reasons I'm going around to the IITs talking to people is why the delay. Now it seems that there is investment and a push. And the other question is, how quickly do you think India will catch up with the US and China? So that's a very complex question. So we'll have to deconstruct it. Pardon if I take a little bit more of the time, but I think this is a very important question. Let me first say, and maybe even correct to a certain extent what you said.

9:25You said China is a competitor. I'm going to say China has pretty much in lots of metrics overtaken the United States. The latest of those is the AAA conference. AAA is one of the largest conferences in the field of AI. and we just heard from the program chairs that there were 29 ,000 submissions in AAA. Now, that alone is mind-boggling. I actually was the program chair of the same conference four years ago and we did not have anywhere close to 20 ,000. We had, I think, 15 ,000, 16 ,000, which was a large number for us at the time. But even more importantly, and so 29 ,000 is just the denominator, so it's fine, 20 ,000 of them are submissions from China.

10:09Yeah. So this is not the final accepted papers, but I would not be surprised if the acceptance rate of China and acceptance rate of US are pretty much similar at this point. A few years ago, when I was the program chair, China and US were closed. China was a little ahead, but now it is just ahead by even leaps and bounds because 9 ,000 is the rest of the world in terms of submissions and 20 ,000 is China. So that's a measure of research. That's a measure of amount of research happening in a place. The number of accepted papers would be a measure of amount of high-quality research above and above happening at a place.

10:51And again, I am pretty sure that China would be higher than everybody else by a large margin. And this is not just AAA, because we were just at ACL in Vienna last month. And again, China had many more papers at the conference, accepted papers, than the U.S. any place else in the world. So at this point, as I'm concerned, China is not a competitor, right? China is sort of, in terms of quantity of research, much stronger than any country in the world. What of that 9 ,000 with the rest of the world, how much of that was from the U.S.? How much was from India? So program chairs didn't say that, but my a-priority guess would be that about 6 ,000 to 7 ,000 would be from U.S.

11:32But we should double-check because that's a guess. Sure, no. I wouldn't think that India would actually measure up in this comparison. I remember that for AAA21, because those are the numbers I had worked out myself, US and China were 650 and 450-ish each in terms of number of papers accepted. I'm not talking about submitted at that point. India was at 32. So we are not even talking a little less. We're talking like a couple of orders of magnitude less. And now, of course, the research in India in the last four years has increased significantly, but so has the research everywhere else. Maybe India's speed is higher.

12:13We can double check, but it will not be higher to the level that it can start matching up to any of these large countries. Large, not in terms of area population, but large in terms of the amount of research that goes on in those areas. So I think the real question to ask is, you know, why? You know, because that's the question, right? And we all have our theories and I can share mine. and these are, of course, theories. So the first question we have to ask is how did China do it from our point of view, right? I mean, I don't know the... And particularly, yeah, just sorry to interrupt, but China destroyed its higher education in the 60s.

12:51I see. Which makes it even more remarkable. There are a few things that really worked in their favor. And one was the fact that early on, and when I say early on, this is pre-AI revolution. maybe very early 2000s, like 90s, early on, they started incentivizing really top researchers in the US who are of Chinese origin to come back to China and establish research labs with almost, what is the right word, a lot of resources, basically. They being the sort of supreme in establishing that group. Now, that was, I think, very visionally, if you think about it. Because if you don't have a culture of research, see, research is almost the edge of the knowledge that we understand as a human race.

13:43So you cannot leapfrog into that place in a little bit of time or a little bit of effort. So people who have that excellence established, if you suddenly start paying them a lot of money and say, look, you should go come back to our country. of course you are Chinese so there is incentive in the nationalistic ideology is present in all of us we would like to go back and support our nation and not only will we give you we will give you a lot of salary we will give you a lot of resources you will have all the labs you can you know do this you can do that you have a lot of power a lot of authority a lot of initiative can be taken by you to establish a research credible research group a credible research group in this place I think that is something every researcher would yearn for because we become researchers for impact that we can give and impact comes in very many shapes and forms and sometimes impact comes just in the form of how many research papers I wrote and how much you know citations I got but more often than not impact comes in the form of how multiplicative my impact was with regards to you know education with regards to number of lives touched did my research go into a into some product did people get benefited from it, you know, anything like that.

15:00We all work hard to achieve all these kinds of impacts, right? And in the U.S., anybody who is not American is always on the fringe, right? You know, as an Indian, I would hang out with Indian community. As Chinese, Chinese will hang out with Chinese community. So we never feel that this is our country. Now, I'm not talking about second generation and third generation Americans. But as first generation, when we go into the country from first, after having grown up in a different country, it doesn't feel your country. It feels like a country that supports you. It feels like a country that gives you opportunity.

15:33It feels like a country that, you know, handles you. You know, you are happy there. I'm not saying you are unhappy. But if you can go back to the own country that you grew up in, you know, that will be a great kind of impact. So I think Chinese government and also there is always this sort of unfortunate thing that my mother used to say, I don't agree, but my mother used to say that, you know, a strong dictator with the right heart, with the right head, whatever, is much better than a democracy. Right, right. And, you know, she uses different words. I'm sort of paraphrasing. And so if the Chinese dictatorial government or a government which has a lot of authority, let's not use the word dictatorial.

16:14They also have some level of democracy. But if the head of the country decides one day that they want to establish China as a research for runners, then the amount of authority that they can exercise, the amount of things that they can do, how they can use the money very fast, you know, that doesn't easily come into democracies. There are many checks and balances, but the checks and balances sometimes slow you down. In conjunction to the fact that China had already by then established itself as a country of manufacturing. And that was bringing in a lot of money into the Chinese government. Pretty much everything sold anywhere in the world with a very high likelihood is manufactured in China.

16:57So all the markets are actually in some ways not controlled but sort of China has a role to play and China benefits from it. So I think there were three things that happened at the time. They had enough money, surplus money. The government had power and the government decided to do something about research and they gave, incentivized, used all the tricks of the trade and established mechanisms so that these amazing people can come back to the country. That doesn't create research culture. It starts research culture. Now, they kept improving their research for the next, I would say, whatever, 8-10 years.

17:37But till 2011, 2012, 2013, they were still not anywhere close to being a forerunner in AI or in any of such fields. But then something interesting happened. and what happened is AI changed on us. AlexNet happened. Now this is the time when every researcher has to reinvent themselves. There were only three or four researchers in the world like the Jeff Henton and Jan Lekoon and Joshua Benjio and Andrew, sorry, Andrewing and, you know, the other couple of other Stanford guys like Chris Manning and some people in Europe, Jordan Schmidt-Huber, few people in the world who had been saying to the rest of the world that look, you should be working on this kind of AI.

18:22But they were not. Nobody was. In fact, all these papers are getting rejected from NeurIPS, from AAAI. They have a lot of bad blood from those times, right? So this community, the research community on neural networks was ostracized. NeurIPS, which has neural word in the first letter, was rejecting all the neural papers. It had become, people used to joke that it should become BIPs, not NIPs, because for Bayesian, right? So modulo these six, seven people, research groups, pretty much everybody else was doing what everybody else was doing, which suddenly became less important. It was a time when there was almost like a reset button pressed in the field of AI.

19:02If you look at a paper that was published in 2012, and you look at a paper that is being published today, the amount of connection will be extremely loose. Because that style of research has now gone out, neural networks is overtaken, And, you know, that is the in thing. And that is the thing that is causing all this change and revolution. At that time, this fledgling research culture that China had got a boost because now it was suddenly the case that no group was any better than any other group. Yes, right. So this change, and because they were constantly trying to establish themselves, maybe it was even slightly easier for them than established research groups because the research groups have to unlearn something.

19:46They have to change something that has already been going on at a very high level and slowly move, right? On the other hand, if you are starting out relatively fresh, you are trying to establish yourself and suddenly this thing happens, you are much faster to jump into it and nobody had any benefit at the time. So I think these are the few things that really will have been China's favor. Yeah. And so then from that point onwards, China is on a path. And now they don't need to incentivize people to go back to China. In fact, another thing that works in China's favor, I'm sorry to say, is the fact that they are pretty bad at English.

20:21Yeah. You know, these things go both ways, right? So for China, it is really good. In fact, China doesn't even encourage English as much in their country because they want, I mean, I'm saying it from my vantage point, that they want, you know, people who are growing up in their country to contribute back to the country with a higher probability. And this would naturally happen if the English is weak. If the English is very good, it can still happen. This is a much easier path for almost everybody. So now let's look at India now, in contrast to all of this, right? So the first thing to realize is that in the 90s, India was bankrupt.

21:03When I say bankrupt, I mean that our reserves had dwindled to a large extent. And it is said that we didn't have money to pay our government servants. This is the time when the finance minister opened India up. So until then, at until that point, India was a really closed economy, the socialist economy to a large extent. It was never communist, but it was still socialist to some extent, maybe almost pro-left kind of ideology was more common if you look at it from a distance. And we were mostly living in austerity because money wasn't readily available. Now, in the 90s, Indian markets were open so that inflow of money can happen from everywhere else in the world so that we can pay our salaries.

21:50Yeah. Okay. So in a few years, Indian economy did stabilize. That's the good news. So far, so good. Then the other revolution that started happening in our country was around the early 2000s, where the IT service industry really started to develop. And this service industry really changed the face of the country. it changed the face of the country in fact there are all these terms and the time my job got bangalared started happening because a lot of the work that was happening in the rest of the world was actually now moving to India India really benefited from this no question about it but this was not yet the same level of benefit that the manufacturing industry probably gave China earlier than this so none of this led I mean it led to better quality of life in the country it led to people no longer feeling that we live in a poor nation right feeling that it's not like the third world or if it's the third world you're the best of the third world you know things like that yeah that sentiment also changes slowly perception changes slowly people started spending people were until then saving people were always in the saving mode and never seen any of my parents or their you know generation people unless they are from the money from the riches to be spending for luxury, to be spending for comfort.

23:07We grew up in a household where you would think carefully about various products, do pro-con analysis before buying any small, you know, like VCR or things like that. Anything little, there will be a lot of effort put in to do the right thing because we can't buy it, we can't throw it. If something gets broken, we go and get it fixed. We don't throw it. That's part of our culture, right? So with that mindset, we never really could take drastic steps like China did. So we always want to have the person, people who are the top researchers in the world come back to India. But we will never give them anything special.

23:47We will certainly not give them a million dollar salary. If they want to come back to India, they should work in the salary that everybody else is getting. They should work in the resources. Sure, you are more visible, you will probably attract more resources sometime down the line, but you will not have a leadership position from day one to establish a research group or a culture. So this has been part of the mindset at every step. We do not, we do think stop down also. I'm not saying we don't do things stop down, but there are lots of checks and balances. Whenever the government, the PM decides that they want to do something, there are five organizations who want to take it forward.

24:23and five organizations will take it forward in their own ways. I am not going to say that any way is wrong and one way is right. I'm not making bad value judgment. Area times velocity is constant. So if the same amount of money gets dispersed at a larger in different directions, you know, in the smaller pockets, then the impact slows down, right? So I think those are the challenges that we have been facing in the country. So we have never been successful in bringing that professors. if you let me just complete. Secondly, we have good English and so it is much easier for us to go elsewhere and establish our personal lives, right?

25:02And to make matters worse, I'm sure this is similar in China, but at least today, the development in China is at a different level. So the quality of life in China is at a very different level than India. So India is also progressing but it is not progressing at the same pace just because we don't have the same amount of money and the same amount of resources. And also there is some corruption in every country and again I wouldn't go as far as saying that corruption is the reason we don't succeed but I would say that it is also a factor right I don't know how much of a factor it is in China so all of these things add up we assimilate really well in the US we become the CEOs they are the CEO of Google we are the CEO of ex-CEO of Twitter we are the CEO of Microsoft we are the CEO of pretty much every company if you start thinking about it right But these are people who went there and never came back.

25:47They will give back. But it doesn't change the issues that India is facing. Just money cannot cut it. Yes, money does help. But we need to also retain our talent. To retain our talent, we need to give them the environment so that they can be successful. If our environment slowly creates roadblocks on the path of success, then people will at some point get frustrated and say, to hell with it i'll just go back go elsewhere and just set up my shop there and i've seen i've talked to a couple of startups who are doing well in the west coast and so on i was in the u.s last year i traveled and so a few people told me one group specifically that look we tried for three years to establish a startup in our country because we cared for it but there were always little things that kept coming in our way and at some point we realized that if we really want to succeed you have to go elsewhere maybe we'll come back to india once we have succeeded but that's not the same.

26:40So I think we have been a little bit unsuccessful in creating an ecosystem of innovation, of research. And I would go out on a limb to say that all of this gets spurred by professors. Now, I am a professor, so you can almost feel that I'm tooting my own horn. But I really believe that information flows downwards. A person who understands the depth at the highest level will train the next teachers. A teacher will understand a depth to a level that they will train the next engineers. The engineers will have trained well so they will you know do work here and create the ecosystem. If you do not have the highest quality teacher professor you will not create the highest quality of teachers.

27:25If you don't have highest quality teachers you will not create a large body of strong engineers who are ready to take on the world. Even if you had them you don't have the ecosystem. So I think it's systemic change will be needed if we really want to, you know, get out of this local optima. Yeah. And that's something I wanted to ask. I'm here really because my wife is an education researcher and she was at a conference in Mumbai and we went to Kota to see the coaching schools. And, you know, Kota is sort of subsiding, but Coaching is spread all over the country. Right. And you have these kids that are working so hard to pass this JEE, main and advanced, in order to get into an IIT, but those are not focused at all on AI subjects.

28:25And then the IITs, you have these AI schools now, But the bulk of the people coming out of the IITs are not AI engineers. And from what I've heard, there's still, you go through this, you know, this tiny hole to get into the IIT. And then you come out the other side and there's still very high unemployment among IIT graduates. I don't know if that's what I've been told. I don't know if that's true. But I've seen some of the IETs are now offering degrees in AI for which you don't need to pass the JEE. And is that exam structure, has that been counterproductive in developing a cohort of AI engineers?

29:17Let me deconstruct your question step by step, because you actually raised a ton of issues in just framing the question. And I'll come back to whether it is productive or not productive for AI, but it's important to understand the system first. Why is everybody fighting to get into AI, to IIT? Now, in the US universities, let's always keep the example of US in mind. You have a lot of Ivy League universities, but you also have a lot of non-Ivy League, but really amazing universities, right? Nobody's going to say that, oh, come on, UC San Diego is not good, or UC Alpine is not good. These are amazing universities.

29:48They are just not a Harvard, and that's all right. maybe Harvard is a bad word in the current political system. Anyway, why is it that everybody is trying to get into IITs? And you should know that 20 years ago we had 7 IITs. Now we have 22. The reason is that for whatever reason, IITs and Indian Institute of Science and a couple of other such institutions have been able to maintain their premier status at a level that they can be at least for undergraduate education compared to education anywhere in the world. I can believe that IIT undergraduates are as good as Stanford undergraduates. Maybe they don't have as much startup ecosystem, but other than that, the quality of undergraduates, the quality of education that we provide them, they're all at pass.