Microsoft Research Podcast - 049 - Celebrating 20 Years of MSR in Asia with Dr. Hsiao-Wuen Hon

Episode Date: November 7, 2018

In 1998, Microsoft Research opened a small lab in Beijing to expand its research efforts and draw on the immense high-tech talent pool in China. No one expected that only twenty years later, MSR Asia would become the dynamic organization it is today, with innovative research contributing to nearly every part of Microsoft’s business. Dr. Hsiao-Wuen Hon has watched it grow from the beginning and this year, celebrates the lab’s 20th anniversary as Managing Director, Corporate Vice President and Chairman of Microsoft’s Asia-Pacific R&D Group. On today’s podcast, Dr. Hon gives us a brief history of MSR Asia, from its humble beginnings to its significant role in the AI boom today, talks about MSR Asia’s unique talent pipeline, shares his vision for the complementary roles of machine intelligence and human wisdom, and explains why, he believes, the more progress we make in AI, the better we understand ourselves.

We want the researcher to really think about how far they can push technology, how far they can push the state of the art. And so this is why we encourage people to take a risk and will even go as far to tell them. If you think too much about technology, most likely you will make too small step. And that's not good enough for research. Because if the most thing we do are incremental, then I think research will fail its purpose. 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 Huizenga. In 1998, Microsoft Research opened a small lab in Beijing to expand its research efforts and draw on the immense high-tech talent pool in China. No one expected that only 20 years later, MSR Asia would become the dynamic organization it is today, with innovative research contributing to nearly every part of Microsoft's business. Dr. Xiaowen Han has watched it grow from the beginning, and this year celebrates the lab's 20th anniversary as Managing Director, Corporate Vice President, and Chairman of Microsoft's Asia-Pacific R&D Group.

On today's podcast, Dr. Han gives us a brief history of MSR Asia, from its humble beginnings to its significant role in the AI boom today, talks about MSR Asia's unique talent pipeline, shares his vision for the complementary roles of machine intelligence and human wisdom, and explains why he believes the more progress we make in AI, the better we understand ourselves. That and much more on this episode of the Microsoft Research Podcast. Xiaowen Han, welcome to the podcast today. It's great to have you in person in the studio all the way from China. Thank you, Gretchen. It's nice to be here.

So you're the Corporate Vice President of Microsoft, the Chairman of Microsoft's Asia Pacific R&D Group, and the Managing Director of Microsoft Research Asia. What has inspired you to do the work you do and to take on this giant set of tasks? I actually think I have the best job in Microsoft in the world. The reason is every day when I wake up, I will explore the future possibility. I will do this experiment, that experiment, try this idea, try that idea. I do know not all of them will work.

Actually, many of them will fail. I mean, that's the nature of the game because you're really trying to explore the unknown. But it's really the process. And I understand people like to get a good result. They want to get reward. They want to really have impact and results even. We all for that. But I think the feeling of every day you wake up, you know you will explore something new.

And then someone even pay you to do those kind of work. And a lot of times they even encourage you to take more risk, to think bigger, to actually do more risky project. That process itself is so enjoyable. So you've been with Microsoft for a really long time. You started in 1995? Yes. And that's not long after Bill Gates started Microsoft Research in 1991. Were you with the research organization when you started?

Yes. What was the vision for a research division in general then? And how have you seen the enterprise writ large evolve over the years since you've been there? If you actually go back to 1991, Microsoft still, compared to today, is a tiny company. But even then, Microsoft already, in the young software industry, is already the undisputed leader. And I think it's really built foresight. We really need to continue to move the state of VR,

the technology, landscape, in general software area. So I think he set up the MSR. He wants the future computer can listen, can understand human's language, can actually recognize objects, understand the world, can do the conversation. And I think at that time, I mean, if you remember, there's a 2AI winter. That's really the first winter. And so I think it's really a very gutsy and the false side of Bill to actually start the research. I remember when I joined the MSR, we have less than 40 people.

And today we grow to the size of a thousand people. Of course, Microsoft also grows a lot. And I think the whole industry also grows a lot. So I think certainly it's a good journey. So then the research division was here? Yes. And you were here? Yes.

So that's like my organization today in China, Microsoft Research Asia. The reason we have Microsoft Research Asia is because we want to expand the company. We want to expand the MSR. So I remember when I joined, I think Rick, our big boss in MSR, and Bill had a conversation. And then we immediately know we cannot just have the research lab in the U.S. Yeah. Because we know we want to hire enough. And also, we don't believe all the smart and the most passionate people doing research

will all come to the U.S. to work for Microsoft. So we better go to where they are. This is why we start international expansion. And then in 1997, we did our Europe lab in Cambridge. And in 1998, we actually did the China lab in the Asia-Pacific area. And then the rest is history. I think the key is at that time, no one think we are even close or willing to invest to actually do those kind of AI technology development or research. So this is why I think we should really be very very thankful to Bill's foresight, also the willingness to actually invest. Put money behind it. Even though Bill knows it will take a long time. And I think Bill

probably will think it will take more than 20, 30 years. I don't think he will predict we really have the AI boom today. And then today, you just cannot escape a day without any AI news or activity, right? So I think people thought about that, but at that time, people think about just science fiction. And then the reason they call AI winter, it's very few people want to invest in it, including government.

Right. And interestingly too, I think most research institutions embedded in industry were applied research as opposed to let's let some scientific minds have some free reign here and really think out for that long time. is there to make money. So typically, most companies can only invest on something more short-term basis, which is actually the right thing. And I think university, government, funding agency can do very long-term research. That's how our society works. And I think Bill, I think he recognized

the industry and also company, when they are successful, they need to participate in building this so-called state-of-the-art, advanced state-of-the-art technology. Because, very simple, look at the high-tech world. I think it's because the fundamental technology got advanced. So the pie becomes bigger and bigger. People can build better applications.

So in return, they have all this positive feedback cycle, so every company will benefit. And I think this is why when we build the MSR, the three core mission was defined very clearly. The first one is advanced state of VR. And that's really just similar to academic research happening in university, you also encourage the researcher to actually publish all they have done and then share with the whole world. That's really a very important aspect because in all the science and engineering advancement in our history, we always talk about you stand on the shoulder of the giant,

everyone can do further. So I think it's very important for industry, for company, commercial company to participate in this state-of-the-art building. And then the second mission, of course, is called technology transfer because, of course, Microsoft wants to benefit from all the great research results we have done to improve our product so that we'll make our product more competitive. And then the third one is incubation, which is also extremely important in our industry

because the breakthrough really happened in a much, much faster pace than any other industry. So Microsoft, we want to make sure we always stay on top of that. And then a better way to do that is to invest and also to invent the future. And I think the incubation becomes a third part of our core mission. Let's talk about Microsoft Research Asia. It's celebrating a milestone anniversary this year, 20 years in China.

Let's talk a little bit about the expectations for MSR Asia 20 years ago and why you believe this particular lab of MSR has been good for Microsoft, good for China, and actually even good for the world. So 20 years ago, the reason we go to China is because we want to expand MSR. And then we know China has a huge number of raw talent. And on top of that, I think the STEM attract a lot of the young kids to study those areas. So 20 years ago, our first, I would say, milestone really comes from the mission we set up, like first mission, Advanced State of the Art. I think to a large extent, we contribute a big part

to the effort to bring China to the world-class research landscape. I think really the first significant paper published in the world-class journal or conference really come from our collaboration work with the university faculty and the students. But 20 years ago, really the first breakthrough, right from zero to the landscape, I think we contribute a big part to that effort.

And we certainly feel very proud of. And also not just that, not just really take to the landscape and particularly you mentioned how this is good for the world. I think in terms of the academic exchange and the interaction between China and the US by and large, because we are a U.S.-based company. And they get to develop their partnership with the university, faculty, and students. And now, again, China represents,

in terms of the graduate students in the U.S., it's the biggest source, right? And also the interaction between the top universities in the U.S., top universities in China, now it's almost on a day-to-day basis. And I think we also feel very honored to actually play a big role to make this at the norm today. And I think we all know, particularly in the science and the engineering world of innovation,

the more global collaboration and interaction are good things for everyone. And then talking about Microsoft, of course, we benefit from that, right? Almost all major Microsoft products over the year until today all have the MSRA contribution. And then on the incubation part, we also have several technology inside Microsoft

are totally started and are responsible by MSRA to actually start the effort. That's an amazing track record for 20 years. Anywhere on the planet, that's just amazing. And so much of it comes from the talent that you get. Talk about MSRA's particular approach to the talent pipeline. You talk about incubation, you talk about your internship,

and you talk about your training. Tell us about that. Yes. We always say in Microsoft, our biggest essay is people, right? And in research, it's actually more so. How do you generate the best research result? Hire the best talent. And in China, it's a little bit different. When we went to China 20 years ago, they don't really have established research in computer science for us to hire. Almost. So we need to really help to foster the future talent for computer science research. So this is why from day one, we know we need to engage with university to help to bring up the students to prepare them for the future computer science research. Even here today,

we have a very, very distinct internship program. Typically, when people think about intern is for people to work in just summer, three, maybe four months in the lab.

In our internship, we always have all year long internship program. One of the reasons we can do that is many of our senior researchers, including myself,

we take professorship in the top university in China. Because of that, we actually can take on the PhD student. That means after the student finishes the coursework,

they will spend two or even three years with us to actually do their PhD work, supervised by us. So we actually have intern work in our lab for a long time. And the number is also mind-boggling. We actually have more interns than the full-time employees.

Our ratio is about 1.5 to 1 in terms of intern to employee ratio. So the talent foster is always a big part of our operation. And I think, of course, that in turn generates a lot of good results. So it's about three years ago, we decided to do an MSI alumni network. So today we have an alumni network that has more than 5,000 people. Of course, a big part of our intern. And then because of the unique timing and also the way we do these things, it also totally coincides with the development of the China IT industry.

And then you look at the number. We have 10 plus alumni which act as the CXO, mostly a CTO for the big public trading IT company. And then on the academic front, we have 30 plus professors, now established professors in the Chinese university. We also have the 50 plus founders for the startup in China. And then five of them

already reached the unicorn status. So it's actually, for so many influential people coming out of one organization in China or even in the whole world, probably it's unheard of. So I think truly it's a win-win a broader view, not just for yourself, because we will only improve or make the true progress if the entire ecosystem, entire platform has the right mix and representative of the talent you want, whether it's diversity, whether it's the best research talent.

If you don't take this view, I don't think you will go very far. Let's talk about the unique focus of Microsoft Research Asia. So I guess I can answer this question in the sort of philosophical way. I know people want a good result. I think Microsoft is also a result-driven company. And also, it's easy to report when you actually have a good result and make a good story in the press, right? But the real day-to-day work in research doesn't really happen that way. If the researcher say, well, the reason I do Project A because I think Project A can generate a great result so I can get a reward, I would say this is probably already a wrong thing to think about.

I think the right thing to think about is we are an organization. We encourage people to focus most of their time on what do they think the future technology will be. And don't yet think about the application yet, right? Sometimes we call this speculative research. Once I make the progress, there always be a lot of people who can think about how to apply this technology to make money, to go for the real world impact. So we really want the people to change the status quo of technology. Not necessarily the status quo of the market share or number of users or applications. We have plenty of business people and product people to think about that. We want the researcher to really think about how far they can push technology, how far they can push the state of the art.

And so this is why we encourage people to take a risk and we even go as far to tell them. If you think too much about technology, most likely you will make too small step. And that's not good enough for research. Because if the most thing we do are incremental, then I think research will fail its purpose. So earlier we talked about the AI winter and the AI spring that some people are saying is a result of a number of factors in high tech that are all converging at the same time. Talk a little bit about this AI spring. So I think earlier we mentioned, even myself, I actually experienced two AI

winter, and then now everyone can notice now AI is a hot topic that no one can escape from. So this is why some people call right now it's AI spring. But there are also people, particularly some researchers or researchers who are worried

it's too much hype and too much expectation. And then we might not be able to deliver. We'll cause another AI winter. So some people say maybe now it's AI autumn. But I think the real happen, I think Microsoft is the right word called digital transformation. Even though the AI technology we have today still has some limitations, but this can do a lot

with so-called data-driven intelligence. And that will actually power a lot of digital transformation for just about everything we do in business, everything we do in life. So that transformation is real

and it's happening. It's already happening for a couple of years and will actually continue at least for foreseeable future. So that's certainly very exciting in business. But in terms of technology, most AI today are the black box. You can solve the what problem, but cannot solve the why problem. This today is still a very hard problem.

We actually barely spread the service. We haven't made even any significant progress yet, but we need to continue to push the technology as a researcher. I use the word AI plus HI. So it's always the AI to help the human intelligence, help human to solve problem together. So we build AI system to actually help us to solve, but the algorithm always come from us. So who is master? It's actually very clear.

And then, of course, we need to make sure all the master will use the AI in the right way. Not actually use the AI to do any harmful stuff. So that's really how I think about it. I like this idea that you just mentioned, the co-evolution of human and machine intelligence. So could you talk a little bit about why you think co-evolution of intelligence is an important thing to address today? Yeah, I do have a talk I gave, particularly in China, talking about co-evolution of AI and HR, human intelligence. Actually, this is based on my personal experience. When I started AI 20 years ago, I also think AI very differently than today. And the more discovery you make in AI, the more

progress you make in AI, then we at the same time also understand ourselves better, myself better, my intelligence better. And I think the human intelligence is actually built on creativity, not really necessarily building on memorizing a lot of data. We do see data. We need to inspire by looking at data, but we cannot look at lots of data or memorize a lot of data. So in a sense, we invent computers to help us do things we know how to do. So we are free us to thinking about harder problems, to actually come up with the algorithm for other harder problems. And we also realize our creativity, particularly the problem solving,

not to mention the wisdom. I mean, I don't think I'm qualified to define wisdom, right? I mean, give people the wiser advice. I think that's also what humans so uniquely possess. If you asked me this question 10 years ago, I probably would not articulate that way because I think the intelligence, we used to think that someone can memorize

or someone can compute very fast, very accurately. Think about it as one form of the unique intelligence. But today, no one's trying to compete with a calculator, not even talk about computer. Oh, memory, right? I already moved for three years. I still have not remembered my home phone number. So I think the way we think about intelligence really changes once we make more discoveries.

And then the other part I also mentioned, humans' intelligence is mostly not to do with big data, but with small data. Think about Einstein, 100 plus years ago. He has no equipment to observe. He just hypothesized. We don't know how he hypothesized. And then 100 years later, we use the modern equipment. We can barely observe those waves. You cannot argue Einstein certainly has no big data,

probably has zero data, right? So just like creativity, right? A lot of time, you just come up with idea. People ask you, where you come from? You cannot explain. On the other hand, today's AI is all based on huge amount of data. We cannot consume so much data, but we can come up with algorithm, let the computer to look through the data.

Then a lot of time, even that, still not enough. You still need to do your prediction and guess and the creativity. All combined, you actually can make a big progress. So I think the co-evolution, there's another aspect. We should also feel lucky, not feel threatened. We are the first generation of the humankind. We live together with the AI, something we created, right? So we can use that as a tool to help us to solve problems,

to inspire us to think about problems in a new way, also inspire us to think about ourselves differently. The reason I'm so optimistic is we are the master, right? We invented, since they are not real life form, I don't think there's a moral issue to say we are the master. Well, it's fascinating because, you know, you listen to the popular press, you listen to some of the big sort of naysayer voices out there that are, we're doomed. We've created our demise. And yet maybe they're looking at it the wrong way and thinking about it too largely.

And it's more like they say here, augment, not replace. The machine has no life. We have the life, right? So this is why if there's a machine will do all these harmful thing, I guarantee there's a life form designer behind that. There's got to be a human behind that. Machines have no life and no DNA, right?

We have the DNA to tell us we need to survive. We need to fight for our gene to continue. So this is why we love our kids. We need to continue our gene. Machine, come on. Machines got programmed by human. It's not there's a bio gene turning, they need to fight for

survival, fight for resource. This is why I think philosophically I'm not saying there's no bad machine, but the source of bad machine got to be a human behind that. A bad human designer. We better regulate the human, but not worry about the machine.

Yeah. And maybe it's not a bad designer, but somebody who designed for a good purpose and it got taken on by somebody with bad intentions. But still people. Absolutely. Still only good or bad, or at the end, all attribute to people. It's human. Right. You are, to many of our listeners, an inspiration, especially when it comes to your career in high tech. Among your titles is Corporate Vice President,

Chairman, Distinguished Scientist, Managing Director, IEEE Fellow. But you didn't start there. Much like Microsoft Research Asia, you started small. What was your path to where you are today? Yeah, I think when I grew up, for whatever reason, I just loved

math. But at the same time, I also liked application. Or you can say I like to see real-world impact. So this is why, looking back, it's actually not a surprise I end up here. So I got a double

E undergraduate degree, and I really like abstract math so I go to study computer science at CMU, got my PhD. After PhD, typically when you got the PhD you can go to academic. I like the academic but I also think I want to see application, I want to see the impact and the impact impact, at that time, I decided, well, impact. Doing computer science, if I can see the technology can benefit, can be used by millions. At that time, I think about millions. Today, I also think about billions, right? Billions of people.

So this is why I ended up joining Microsoft. And also, going back to your first question, Microsoft started research. That's such an exciting thing. I mean, not just product. I don't think I would be happy just doing product and not actually push the technology envelope to explore all these future possibilities, right? So

the combination of research and then product impact, not directly, indirect, because I'm not responsible for product, but if I see my technology, whether it's a computer vision or speech or system technology being used by millions, billions of people, I think that I feel I really do something tangible for humankind. So if I really look back, I think two things. One is, it's not a surprise to end up here. But second, I also say

I need to be very thankful. Microsoft gives me such a wonderful opportunity for me to fulfill not only my goal, also give me a job I actually find fun to do every day and get up every morning knowing I would explore this and that and would try this and try that. And then the process itself, a lot of time, it's enjoyable already, so enjoyable already, not to mention if I can have a good result along the way. Sounds like a wonderful life.

Xiaowen Han, thank you so much for joining us today. It's been inspiring. Thank you. To learn more about Dr. Xiaowen Han and the latest innovations from MSR Asia, visit Microsoft.com slash research.