Microsoft Research Podcast - 103 - Innovating in India with Dr. Sriram Rajamani

Episode Date: January 22, 2020

Dr. Sriram Rajamani is a Distinguished Scientist and the Managing Director of the Microsoft Research lab in Bangalore. He’s dedicated his career to advancing globally applicable science in the testbed that is India. He is, by any measure, a world-class researcher and leader. He’s also, as you’ll find out shortly, a world-class storyteller! Today, Dr. Rajamani talks about the unique challenges and opportunities of leading MSR’s research efforts in India and what it takes to build a robust research ecosystem in a country of huge disparities. He also dispels some preconceptions about poor and marginalized populations and explains why ‘frugal innovation’ may be one key to solving societal scale problems. https://www.microsoft.com/research  

I think the number one thing that strikes you when you try to build technology in India is the resource constraints. You know, if you want to build technology that actually fits the lowest common denominator, that actually works everywhere, the resource constraints that you have to think about cost, bandwidth, the diversity of users, I think those are extreme in India. Because of that, if you build systems that somehow work in those constraints, you are innovating for the world. You're listening to the Microsoft Research Podcast, a show that brings you closer to the cutting edge of technology research and the scientists behind it. I'm your host, Gretchen Huizinga. Dr. Sriram Rajamani is a distinguished scientist and the managing director of the Microsoft Research Lab in Bangalore.

He's dedicated his career to advancing globally applicable science in the testbed that is India. He is, by any measure, a world-class researcher and leader. He's also, as you'll find out shortly, a world-class storyteller. Today, Dr. Rajamani talks about the unique challenges and opportunities of leading MSR's research efforts in India and what it takes to build a robust research ecosystem in a country of huge disparities.

He also dispels some preconceptions about poor and marginalized populations and explains why frugal innovation may be one key to solving societal scale problems. That and much more on this episode of the Microsoft Research Podcast. Sriram Rajamani, welcome to the podcast. Thank you. Excited to be here. You're a distinguished scientist and the managing director of Microsoft Research India, and that's a big job that encompasses a lot, not only your own research, but the research of the people that you supervise and guide and direct.

So tell us in broad strokes what you do for a living. What does a day in your life look like? What gets you up in the morning? Oh boy. So I'm a morning person, so I'm up quite early. I usually read in the morning. I have a big reading list. My colleagues, they send me a lot of reading material about work that they do.

And I usually have a week or two worth of reading material in advance. That's my reading queue. So my mornings are usually spent reading at home. And then most of the day is actually spent in small group discussions where we sort of take a research topic, get into it with real depth, and we ask many difficult questions are we doing the right thing are we investing this right should we change direction should we pivot that's the most fun part of my job i am usually home by five and with my family i do a group yoga class in the evening that's all of my my break in the evening and then late evening there's redmond calls so starting like 8 p.m 9 p.m for a couple of hours there's usuallymond calls. So starting like 8 p.m., 9 p.m. for a couple of hours,

there's usually conversations with either researchers in Redmond or product groups in Redmond. I do spend a lot of time traveling because as Microsoft Research, we engage with academia, not only in India, but throughout the world. And I come here three, four times a year. So that hopefully gives you a sense for what I do. Yeah.

So tell me about your personal passion and what drives you? What questions are you asking that you would really like to answer? What problems would you really like to solve? Tell me what your heart is for research. Yeah. So my personal research is in systems. My PhD was in formal verification. A lot of my personal research quest is actually understanding these extremely complicated computing systems that have really transformed everything around us and understanding what it takes to build them so that they are stable, they are robust, and they do what we intend them to do. So that's my personal passion. But these days, actually, a lot of my time is spent not only on my own work, but the

work of my colleagues, which ranges from mathematics, algorithms, to artificial intelligence, to machine learning, to systems, to human-computer interaction. A lot of my energy gets spent on understanding these various topics. I'm a research junkie, so actually I spend a lot of time learning. That's what I do. That's my main passion is learning. Microsoft Research has labs around the world, and each one brings something unique to the research endeavor. So give our listeners a brief history of MSR India. It's kind of fascinating.

What's your particular guiding mission, and what were the particular challenges and opportunities for opening a lab in Bangalore? The opportunities in a place like India are many. First of all, it's a country with more than a billion people, a lot of them very young. So there's tremendous amount of potential for talent. And what we can do there, it's a growing economy. And to just give you a sense, when I graduated in 91, most of my class left to the US to study. But now, if you look at the people that are graduating, a lot of them are staying back because there's enough economic opportunities there.

There's a lot of interesting things actually happening in India. It is a very interesting testbed. To give you a sense, right, there is about 150 spoken languages, each with 100,000 to a million or more people speaking those languages, which you can see how different that is from a place like the U.S. And if you look at actually languages that are spoken by fewer people, still tens of thousands, there would be 1,500. Huge disparities in socioeconomic conditions.

You know, you will find extremely rich people, extremely poor people, everything in between. And wide infrastructure variants. You know, if you go to a city, it'll be just like in the U.S. And if you go to a village, there'll be like nothing. So in terms of actually why we went there, we went there because of talent and the opportunities. In terms of how we converged on what we work on there and what's the sort of the unique value that MSR India brings, we've always tried to strike a balance between globally applicable science and being inspired by India as a testbed. So, you know, as a talent, right, India has always had a really good mathematics talent.

That's one of the reasons why we work in algorithms. We have a very strong set of people that work in algorithms there. Over time, we have built expertise in systems and machine learning. And we also work on socioeconomic development, which is a very local thing in India. And we didn't plan these areas ahead of time. We sort of meandered around, and we have converged on these areas ahead of time. We sort of meandered around and we have converged on these areas for 15 years. We have sort of evolved over time. And actually,

MSR lets lab directors the flexibility to just evolve in that story, which is wonderful. As both a scientist yourself and a leader in technology, you're in a unique position to reflect on trends. And I would say both those that you observe and those you create. So what does it take in your mind, Sriram, to be a leader in technology today? And how are you executing towards that in an age of AI? Most of a scientist's job is to predict how the world is going to look like in five years from now, 10 years from now. And nobody has a perfect crystal ball, right? So a lot of it is actually based on your intuition,

the intuition of your colleagues, social conversations you have with people, and painting a picture of how the world is going to be five years from now, 10 years from now. Let me give some examples to sort of illustrate what I mean. So I was a grad student at Berkeley in the late 90s. And during that time, if you sort of think about security,

we always thought about security of data, much like physical security. Like you store your valuables in a locker and you lock them up, and then you do access control. You sort of decide who gets access to your house, and similarly, you decide who gets access to your data. You sort of decide who gets access to your house and similarly you decide who gets access to your data. So most of security was about access control. We have a very fine young researcher in our lab, his name is Saikat Guha. Around 2008, he started thinking, oh no, that's not the right way to think about security in the internet age. We have to actually think about not only who has access to data, but what they do with it, which is a real conceptual shift in how they think about security.

And when he started thinking about it in 2008, there were very few people that subscribed to that view, right? You know, he was like a lone ranger working on that for several years, and he built tools to actually codify those ideas. And many years later, when GDPR came, he was already ready with frameworks and so on. And he built a framework called Data Map over the past decade that was so influential in how Microsoft thinks about GDPR compliance. Another example I would give you is, and I know you've had Manik Verma as a wonderful podcast.

So he worked on a machine learning system called extreme classification. So let me refresh your listeners as to what that is about. Today in machine learning, people think about classifying objects or data into a small number of classes. You could take a picture and classify it as a cat or a dog. But Manik Verma thinks about how to classify things so that the number of categories could be in the order of millions. And when he first started doing that, people thought he was crazy. But today, there are many, many applications in advertisements, in recommendations, in ranking.

And extreme classification is now a unique sub-area in machine learning that he started. Today, if you go to NIPS or ICML, there is actually a workshop in extreme classification. Right. Right. That's an example of, again, foresight into how the world would look in several years down the line. A lot of what you need to paint a picture of the future is to have a hypothesis, have self-confidence in it, and have a community that works with you to create that future. MSR India focuses on four key areas of technology research, and you've alluded to them already, but let's talk about them specifically.

Algorithms, systems, ML and AI, and technology for emerging markets, or TEM. Talk briefly about how your lab is contributing in each of these areas. We don't have to get granular, but give us an overview of the vision for each of these areas and why they kind of go together, overlap, and have their own space as well.

Algorithms is pretty much the foundations of computing, right? That's the math behind data science, the math behind cryptography, the math behind everything that we do in computing. And we are very fortunate to have amazing, incredible minds that actually work in this space. A lot of machine learning actually starts out as algorithms. Thinking that actually happens today will lead to machine learning algorithms maybe five years down the line, ten years down the line. And today, if you look at them, they'll be math equations written on a whiteboard. So our algorithms group does a lot of leading-edge work that is going to only see the light of the day five years down the line, 10 years down the line. But that said, things that they did 10 years ago are now seeing the light of the day. For example, we worked on, you know, things like topic modeling,

which are now incorporated into working tools that are used by many, many people inside the company today. That's an example. One other thing that people work on is, you know, you may have heard a lot about deep learning. And one of the things that is interesting about deep learning is that even though it works in many cases, we don't quite even understand why it works, what the limitations of that are, when it'll fail. And so, you know, people in the algorithms group try and dissect and understand why deep learning does what it does, what its limitations are, and understanding what algorithmic tweaks that we need to do to make it even better. And then moving on to machine learning and artificial intelligence, that's a very wide

spectrum. I already spoke a little bit about extreme classification, which talks about classifiers in the large. We also work on machine learning in the small. We work on, you know, Edge ML, which is actually machine learning running on very, very small devices, devices that you could buy for $2 or $5. And, you know, how do you make machine learning algorithms work on them? You know, another very interesting topic that we work on is something called approximate nearest neighbor. Let me say what that is. Today, the way search engines work is by using something called information retrieval. But that's yesterday. Going forward, what happens is that because of deep neural networks, the search is actually done in the higher dimensional space. And this requires entirely

new algorithmic thinking. And people in our lab, they span over all the way from the algorithms to machine learning. So there's new algorithms that we have been designing on how to do this nearest neighbor search, which have the potential to transform the way search engines are built. And then moving on to systems, systems is the foundation of the infrastructure on which everything else is built, including AI and ML.

So we work on the interaction between machine learning and systems. We sort of think about how machine learning can make systems better. How can we get the signals that actually come from our data centers, where the data centers are constantly running billions and billions of computation, and if something fails, we get those signals back. Crashes, we get those signals back.

How can we use machine learning to map those things back to actual code that people write, so that when something fails, we can point out, hey, this fails because this line of code is actually not working right. We try to use machine learning to figure out how to optimize COGS, which is cost of goods. We also try and build systems for better machine learning. How do you build better infrastructure so that we can utilize GPUs better and do better GPU training? And then the final area, which is technology for emerging markets, you know, we do things ranging from public health to education, to we study illiteracy, we study human rights,

how to build technology so that they are just and fair. So those are the kinds of things that we do. Let's talk for a minute about why India is the ideal place for disruptive technology and how constraints drive innovation. How are the realities of life in developing areas of the country turning some current assumptions about technology upside down? As I mentioned, right, India has wide, you know, socioeconomic disparity. In Bangalore, you could go to a mall that would look much like Balu Square. And on the other end, you could go to a rural area in which, you know, there might not even be electric power.

I think the number one thing that strikes you when you try to build technology in India is the resource constraints. If you want to build technology that actually fits the lowest common denominator, that actually works everywhere, the resource constraints that you have to think about cost, bandwidth, the diversity of users,

I already mentioned the number of spoken languages and so on, I think those are extreme in India. Because of that, right, if you build systems that somehow work in those constraints, you are innovating for the world. One saying I've heard is actually, if you make something work in India, it'll work anywhere.

That's actually something I've heard. And it's so true, right? If you sort of go to a rural area and open up your mobile and press download on something. It just spins forever. How do you build a system that supports those users as well as users in the city? That, I think, is a tremendous opportunity.

So in our lab, we actually have thought a lot about this. One of the terms that actually describe best what we do is called frugal innovation. Innovation that actually thinks about cost essentially as its core, because if something is not low cost, it's just not going to fly. And the thinking about technology as an amplifier of human ability, I think so technology should not replace people because there's no point in doing that. So the point is actually use technology to amplify human ability because the real scarcity is actually talent, right? And skill. So how can you amplify skill that a few

people have to serve more people? Thinking about poor underserved populations a lot more carefully, you know, distinguishing between their needs and wants. Most of us actually in the West think about, you know, when we work with poor people, we think about health, education, right? Those should be their needs, right? But in reality, if you study them, they have a lot of wants. You know, they want entertainment.

They want employment, right? But thinking about poor, not as just consumers of information, but producers of information, they have very many interesting things to say. Thinking about the lived-in experience of the two billion people that are not yet part of the digital economy, because of many, many reasons. Illiteracy, thinking about illiteracy as a cognitive deficit, and

thinking about you could give them the best smartphone, you could give them the best 3G, 4G connectivity, but if they don't have textual literacy, how are you going to connect them and include them? I think the final thing I would say is that when you design technologies, you know, to serve this kind of community, being completely honest to yourself that it actually works. You know, doing rigorous scientific evaluations to actually see whether it makes a difference or is this a shiny object that you just designed in a lab, you know, just because it is fun. Right. Right. You know, you're hearkening back to Ed Cutrell, who was on the show, and I know he did work in India.

Yes, he used to be with our TM group for many years. Yeah. And some of the stories he told on the podcast he was on, I encourage people to go listen to that one, because there's actual stories of things that they thought would work in particular scenarios that they just wildly didn't, but not for the reasons they thought they wouldn't. Right? It's like...

Right. I could tell you a story. Do, please. I love stories. Yeah. So we have a researcher. Her name is Indrani Meditis. She's one of the world's leading experts on illiteracy. So, you know, she and a bunch of others wanted to build a job website for low-income people, sort of like a monster.com or something for like cooks or drivers and low-income labor. And so they built it with only pictures, right? Because these people wouldn't be able to read text, so they built the whole interface using pictures.

And I remember there's a slum near our lab, so they wanted to do a pilot in the slum. So there was a lot of discussion in the lab about how to put a computer there so that the computer won't be stolen. And so that actually you can access it, but you can't walk away with it. And then everything, you could apply using pictures, you could actually look at the job listings, and they did all of that, right? And after that, they deployed it and the usage was zero. It was there and people were curious about it, but nobody used it. And what had occurred to Indrani was that the reason is actually they have no conception of what this thing would even do. And so what they did was, you know, they enacted something like a soap opera in the lab with actors from the lab. There's a woman who sort of is complaining to her husband that she needs domestic help.

And then the husband goes and registers the fact that they need domestic work in this site and then there's a woman who comes uh accesses his computer in the slum and she clicks on this and she gets introduced and they meet and she gets a job this is now being run as a screensaver in the computer and then the usage of this thing skyrocketed interesting right so ind? So Indrani's main conclusion, right, is that illiteracy is not just about textual literacy. It's about lack of context and awareness. And unless you actually put yourself in the shoes of a person who has never seen something like this before, you're not going to fix this by just pictures, right? So that's an example of things that you think that would work but wouldn't work.

You and your colleagues are tackling some what you call societal scale issues. Healthcare, education, agriculture, employment, connectivity, transparency. We've talked about quite a few of these already. Give us some more context for the research projects that your teams are working on that might give us cause for hope for some societal scale solutions. Yeah, so I could tell you a few stories. I already told you the literacy story. Keep going. So one project where we have made a lot of traction is a project called 99 Dots,

which is around technologies for tuberculosis medication adherence. This is a project that was initiated by Bill Theis and Andrew Cross. So the context for this is that TB is a curable disease, but you have to take medication for six months. And if someone falls out of medication regimen, then they get something called drug-resistant TB, which is both contagious and fail. So the only way to cure TB is to make sure that a healthcare worker meets with the patient every day for six months to ensure that

they've taken medication. And you can imagine how cumbersome it is for both the healthcare worker and the patient, right? So suppose we could use technology to get that information. Then the healthcare worker could spend all their time on people that are actually not taking medication. So they designed a sensing system, which they iterated many, many times. But what finally works is actually they work with pill manufacturers

and they designed a new paper strip so that actually when they dispense the pill, it reveals a phone number to which the patient is actually counseled to give a free call. On the other side, a computer picks up the call and it records that, oh, this person is now taking a drug. So the computer knows when the calls are coming and when the calls don't come, there's like a red bar saying this person hasn't called. And then the counselor spends time on that patient. And this was started out as a research project in our lab. And then we spun it off into a nonprofit because, you know, the government wanted to adopt it.

You know, the Gates Foundation, USAID, they wanted to fund it. So we spun it off into a separate startup company called Everwell. And it's walking distance from our office. They employ about 20 people and they've enrolled more than 200,000 patients. Some other examples are, one of the things we work on these days is road safety. So traffic accidents are a huge killer of people in India. And so we have a research project called HAMS, where what we're doing is just using a smartphone, we can monitor both the behavior of the driver and the surroundings. So we can actually know whether a driver is sleepy,

whether they are wearing a seatbelt, whether they are talking on the phone when they are driving. You could imagine how this technology could be used to monitor fleets, how to make driving safe. And a very interesting application of this is an automated driver licensing. So today, if you go to Dehradun, as of two months, you should go do a driving test. There's no instructor. It's a phone. It's a phone. And then you drive and then an automatic printout gets printed out saying, actually, these are the things you did right. These are the things you

didn't do right. And you passed or failed. Oh, interesting. So let me add a few more, right? You know, I mentioned HAMS, you know, like, you know, BlendNet is another project, you know, because connectivity infrastructure is such a big issue. know if you go to you know rural areas right you can get text messages by but if you try and download a video you know you'll see the wheel spinning forever and you'll be never able to download something right so blendnet is a very interesting idea where most of the popular videos and other bulk things you want to download actually other people want them too odds are that somebody else will have it. So BlendNet is what is called a cloud-connected content distribution network

where if you want to download a Bollywood movie, what you do is actually you use the 2G, 3G only to actually say what you want. And the cloud has some metadata which actually stores who has what video. The actual video might come from you. I just connect to the cloud and say, I want this movie. But the actual movie comes by your phone some metadata which actually stores who has what video. The actual video might come from you. Right. I just connect to the cloud and say, I want this movie. But the actual movie comes by

your phone turning on your Bluetooth or your local Wi-Fi, my phone turning on. And peer-to-peer. And peer-to-peer, right? And using that. So fascinating. You know, just going back to your Bollywood movie, Download, those are four-hour productions that encompass every single human emotion and dancing. At the same time. Because you want your money's worth. Absolutely. Well, several trends in technology have actually broadened the scope of the problems that we could

solve today. You know, hyper-compute power, sophisticated algorithms, and massive amounts of data. But people in the field are starting to recognize that we need more than computer scientists to solve these problems. So give us your take on the trend towards interdisciplinary research, especially in the light of technology for emerging markets. Yeah, I think this is a very important question. Maybe I'll, again, actually, in the spirit of storytelling, let me actually give that as an example with a particular project, right? You know, I mentioned a few times HML, right?

HML is about running machine learning on very small devices. This is actually the dream that, you know, today, there are these very small devices, and they are primarily used as sensors, and their capability is to just transmit information to the cloud. And the assumption that they will work only when there is cloud connectivity. But the HTML project's hypothesis is that what if you could actually do machine learning there?

But now, it's a very difficult question. First of all, you have to start with the math to figure out, can you actually do it? That's where the algorithms people come in. And then after the algorithms people figure out that actually you could do it, then you need the machine learning people to actually design those algorithms. We need systems people and compilers people to compile those algorithms to run on those small devices. And then you actually need HCI people to think about what this might really solve. Now, when you imagine the future, right, you have to think about what the algorithms are, what the systems are going to look like, and actually how people are going to interact with it.

All right. Let's talk about talent. You alluded to that at the beginning. You're what we call in the United States a 4A high school. You've got a lot of kids to choose for your football team. So with billions, one of your problems might not just be that you have a lot of people to choose from, but you have a competitive environment for getting the best talent to come to work with you. So what's MSR India's value proposition to get the best and brightest AI talent these days? So India is a pretty interesting place from a talent perspective. You know, we have a really strong undergraduate population. But our graduate program still lacks critical mass with the number of PhDs that come out of India.

So one of the things we do is that our PhD recruiting is very global. And that's our hiring opportunity, right? So we recruit globally, you know, from people perhaps of Indian origin. And there are people like Bill Thees and Andrew Cross who are, you know, not of Indian origin, but they want to come live there because of India as a testbed. So, I think one of the things we have done very cleverly, if I may say so, is to think about recruiting

very, very globally, particularly at the PhD level. And even if a small fraction of Indians living worldwide want to come back, right? A small fraction of a billion is still a very large number, right? So every year, right, even if 10 people want to return, you just, you know, pick the best of those 10 and hire them. And then actually, undergrads, we actually work with undergrads in India. We have a program called the Research Fellow Program,

where it's sort of a pre-doctoral program, where we take undergrads and they spend one to two years with us as research apprentices. And then they go off to do grad school in the West, you know, typically in Europe or in the US. And, you know, in the 15 years we've been running this program, I think we would have graduated maybe 500 such research fellows. Many of them have now finished PhDs and they've come back. So, you know, we spend a lot of time nurturing young talent,

because we play the long game. That's the way we work with undergrads. And in terms of value proposition, right, there are people like me who want to do honest-to-God good science, right, and they want to live in India. Here's an environment where you could do research like anybody else in the world if you choose to live in India. And then, actually actually you combine that with locally relevant work, like technologies for emerging markets, where you connect with the community, think about India as a testbed, and you put those both together, and then you get a different kind of energy. And that's what MSR India is. Collaboration seems to be a big trend in an era of AI and ML research. So first, I want you to tell us why collaboration

is really important in your world particularly, and then tell us about some of the collaborations you're involved in and how they're bearing fruit. Yeah, so I already mentioned about interdisciplinary collaboration in the lab, and I think that's very central to what we do. But in India, the other thing that's very important for our lab is collaboration with our ecosystem,

which is the academic ecosystem. It's quite important because the graduate program is still not quite strong. So many of our staff are adjunct faculty in Indian universities. So many of us co-teach courses, we co-supervise PhD students, and that's a very integral part of what we do.

And I think that has actually built real trust and credibility with the academic ecosystem. Now, you mentioned Manik Verma. Manik Verma recently was awarded the SSB Prize. It's one of the most prestigious awards in interdisciplinary science. Also, there's an Indian National Academy of Engineering. So we have three fellows from INAE in our lab. I'm one of them, right? I know we have a MacArthur Prize winner, we have a Knuth Prize winner, and so on. So all of these, right, are not just bragging about our staff. I think these are really awards to collaborations that these people had with the community. And these recognitions come not because these guys sit in a lab and work, but they share

the work and bring the energy of the academic community. So that's actually super important in a place like India. Talk a little bit about the research ecosystem there and some of the work that you're doing to build community and train people. You've alluded to the research fellows program, but there are other things you're doing, sort of broader spectrum. Talk a little bit about that. So one of the things that we're doing is to, for example, bring conferences into India. Travel grants for Indian academics are very, very hard to get. We are actually privileged to be in a place like MSR where we could travel and go to conferences.

But many students in India, they just don't have the ability to go to conferences. So if they can't go to a conference, you try and bring the conference to India. So that's something that we try and do. So we participate in a lot of those kinds of activities. We also organize workshops. Years ago, I started a series called Mysore Park Series where we get high quality peer interaction. People get to a community in a small group

and discuss topics for like four days, five days, because you have to actually get people to talk to each other. And we spend a lot of time and energy creating those kinds of conversations, nurturing those kinds of conversations. And the community is very welcoming of us doing that. That's one of the reasons why people join our lab. Because when they join our lab, they are not in a bubble. They are actually connected to an environment and connected to the ecosystem around us.

We've talked about what gets you up in the morning, Sriram. But this is the part of the podcast where I ask what could possibly go wrong. So given the power of AI and its potential for both great good and great harm, is there anything that keeps you up at night? And if so, what are you doing to mitigate it? There's some people who believe that AI will become like the Schwarzenegger Terminator and come back and kill all of us. I, for one, don't believe that. I don't believe that.

We are very far away from that. But what worries me more is not the fact that AI will be all-powerful and conquer us, but I'm a software reliability person. I'm a systems person. I actually want systems to work well. My worry is more that in our enthusiasm as technologists, we overestimate what AI can do and deploy it before it is ready. I think that worries me more than AI conquering us. AI is, of course, trained by data.

And if the data is not representative, it's going to cause huge amounts of bias. And it's going to take decisions that systematically amplify human biases that people have. People are aware of this. And that keeps me up at night, because you have to really think about whether the AI is actually really helping people.

Not only in terms of research, but I also think about it in terms of investment. Now that I'm in a lab director position, to give you a sense, right, one of the biggest promises of AI is natural language processing, because you can now talk to a computer. And if you're an illiterate, right,

that is going to open doors. Now, if you can't read and write, but if you can speak and the computer can understand you, it's going to bring you into the part of the digital economy but look at the investments in nlp they are all in english in german you know those are the markets where the money is right and that's where actually people are investing more and more to make you know your speech assistants understand uh you know these kinds of languages but what about the 150 languages? What about the 1,500 languages? What about the tribal languages

that are spoken by 10,000 people? And all of them are illiterate, right? So are we doing enough investment to include them in this AI-driven economy? And so that disparity, I think, is something that I worry about. I think it's extremely important to think about entrepreneurship, right? Because, you know, marginalized people, poor people, they want to live better and they

have a lot of energy in them, right? I think creating entrepreneurship opportunities for them so that they can generate economic value, so that you don't just donate money to them, but you sort of enable them to be successful and creating businesses and then creating economic value, which will then lead to an ROI. Right. But the real difficulty in these kinds of things, right, even if you do them, they are all going to be in the knee of the hockey stick. It's going to be many years of investment before you see the exponential that has come up, right? So I think the biggest challenge is actually in persevering through this exponential.

It's a very difficult thing to do. It's story time, and I would love to hear yours. So even though we've been telling stories pretty well the whole podcast, let's get a personal story in here. Tell us a little bit about your history and where you've studied, where you've worked, what got you started along your path, and how you ended up at MSR in your leadership position today? So I did my undergrad in India and like most of my colleagues, I came here for graduate school. I first did my master's at UVA at the University of Virginia and I thought

that maybe I won't be a researcher so I went and became a programmer. I wrote software in the Silicon Valley for a few years and I wrote hundreds of thousands of lines of code. And then after a few years, I decided I really wanted to do research. So I went back to the PhD program at UC Berkeley, and I did my PhD in formal verification. And after I did my PhD, I used to work on formal verification for hardware circuits. And around the time I graduated, I met Jim Larris and Amitabh Srivastava. You know, Amitabh was running this place called Programmer Productivity Research Center. And they recruited me to see whether these kinds of formal methods for hardware, can it be used for software?

I found it very intriguing. So I came here, you know, with that, you know, hook in mind. And I met Tom Ball, who's still a researcher here and he and I did in many years of collaboration where we sort of try to combine formal methods both from the hardware area together with theorem proving together with compiler style stuff that the software people do to really think about how to formally validate software mostly analysis work is what I did when I was here. And then I went back to India around 2005,

in a few months after our lab started. And my initial work was on design, software design. So there, actually, instead of finding bugs in a driver after the driver is written, think about how might you write it so that by construction, your software is actually better? So we designed a language called P, where you design software in a high-level language, and you analyze your design and make sure your design is robust, and then you generate

code from it, and that's what runs. And methodology from this is actually what is now being used to run your USB stack, right? Then I worked on security in MSR Cambridge to build cloud where you can actually guarantee that hackers can't have access to your data. So in my own story, I went to MSR India when I was in my mid-30s. And three years ago, I became a lab director. So I've had the fortune of being an individual contributor, a researcher, a group manager, and now a lab director. So I've had the fortune of being an individual contributor, a researcher, a group manager, and now a lab director.

So that's been my journey. What's one thing that people might not know about you that may have influenced you to be a researcher or a leader in tech? I think probably most people don't know that I'm a village boy. My dad used to be in the agricultural department in southern India. He worked for the government. So I was born and I grew up in villages with no electric power. I'm not that old, right? It's true. I'm looking at him. He's not that old. But I grew up in villages in which there was no electric power. There was no cooking gas. So my mom used to cook with charcoal and firewood. So I have that kind of upbringing.

And I think that influences me in many, many ways. I'm the first person from my family to ever leave my country. And it's sort of full circle for me to be from that environment, go study here and go back and be a considerable part of our lab, work on technologies that benefit rural people, people living in poor areas and so on. As we close, I want to give you the last word, Sri Ram. You've compared research to a marathon. Tell our listeners who may be just getting into the race what they have to look forward to and why, in the long run, they shouldn't be afraid of the long run. If you want to do science that changes the world, you need to give time.

It's extremely important to do that because to make any research mark, it's going to take many, many years. Because you've got to try. Many things will fail. Some things will work. And even if some things work, it has to actually gather critical mass. It has to attract attention from people. The right environment should be there for it to get deployed and so on. So things take a long time.

So one advice I would give is just be prepared for the long haul. It takes many, many years to make a mark. As a result, it's extremely important to pick problems that you like. Pick areas that you like so that you have fun. Otherwise, it's hard to actually sustain the energy to run the marathon. The other advice I would give is to not be lonely.

Not do it alone. Build a community of colleagues to collaborate with you. Pick people that have quite different skills from you to collaborate so that you can actually learn from others. You teach what you know, you actually learn from others. Research is very much a social process. That's another thing that I would encourage. And the other thing I would encourage is,

think about problems that many, many people care about. Real world problems that if you actually solve them, it will make a real difference. And work on those problems that are hard to solve, you know, rather than, you know, count the number of papers you publish. Right?

So I would say, I think it's far more satisfying to do a few things that change the way science progresses, change the way a field changes, you know, rather than have a laundry list of publications. Sriram Rajamani,

I thank you for coming all the way from Bangalore just to see me. Gretchen, you know, I'm so happy that you spent the time thinking about what a lab is and doing this podcast. And also thank you for the opportunity to share this story with your audience. The gratitude is mutual. To learn more about Dr. Sriram Rajamani and the latest innovations out of MSR's lab in India,

visit microsoft.com slash research.