Microsoft Research Podcast - 036r - A conversation with Microsoft CTO Kevin Scott

Episode Date: May 20, 2020

This episode originally aired in August, 2018. Kevin Scott has embraced many roles over the course of his illustrious career in technology: software developer, engineering executive, researcher, angel investor, philanthropist, and now, Chief Technology Officer of Microsoft. But perhaps no role suits him so well – or has so fundamentally shaped all the others – as his self-described role of “all-around geek.” Today, in a wide-ranging interview, Kevin shares his insights on both the history and the future of computing, talks about how his impulse to celebrate the extraordinary people “behind the tech” led to an eponymous non-profit organization and a podcast, and… reveals the superpower he got when he was in grad school.  

When I spoke with Kevin Scott in the summer of 2018, he talked about his path from small-town kid in rural America to Chief Technical Officer of Microsoft. Since then, he's taken on another role as Head of Microsoft AI and Research, and in April of this year, released a book called Reprogramming the American Dream, where he shares his vision for how AI can bring

the benefits of technology to all areas of the country, including those that have historically been left behind. Whether you heard my chat with Kevin in 2018, or you're just tuning in today, I know you'll enjoy episode 36 of the Microsoft Research Podcast, a conversation with Microsoft CTO Kevin Scott. It's a super exciting time, and it's certainly something that we are investing very heavily in right now at Microsoft in the particular sense of, like, how do we take the best of our development tools, the best of our platform technology, the best of our AI and the best of our cloud to let people build these solutions where it's not as hard as it is right now. 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. Kevin Scott has embraced many roles over the course of his illustrious career in technology. Software developer, engineering executive, researcher, angel investor, philanthropist, and now chief technology officer of Microsoft. But perhaps no role suits him so well or has so fundamentally shaped all the others as his self-described role of all-around geek. Today, in a wide-ranging interview, Kevin shares his insights on both the history and the future of computing, talks about how his impulse to celebrate the extraordinary people behind the tech,

led to an eponymous non-profit organization and a podcast, and reveals the superpower he got when he was in grad school. That and much more on this episode of the Microsoft Research Podcast. Kevin Scott, welcome to the podcast today. Well, thank you so much for having me. So you sit in a big chair. I think our listeners would like to know what it's like to be the chief technical officer of Microsoft. How do you envision your role here and what do you hope to accomplish in your time?

I.e., what are the big questions you're asking, the big problems you're working on? What gets you up in the morning? Well, there are tons of big problems. I guess the biggest and the one that excites me most and that prompted me to take the job in the first place is I think technology is playing an increasingly important role in how the future of the world unfolds and has an enormous impact in our day-to-day lives from the mundane to the profound. And I think having a responsible philosophy about how you build technology is like a very, very important thing for the technology industry to do. So in addition to solving all of these sort of complicated problems of the how, what technology

do we build and how do we build it, there's also an if and a why that we need to be addressing as well. Drill in a little there, the if and the why. Those are two questions I love. Talk to me about how you envision that. You know, I think one of the more furious debates that we all are increasingly having, and I think the debate itself and the intensity of the debate are good things as sort of around AI and what impact is AI going to have on our future and what's the right way to build it and what are a set of wrong ways to build it. And I think this is sort of a very important dialogue for us to be having because in general, I think AI will have a huge impact on our collective futures. I actually am a super optimistic person by nature, and I think the

impact that it's going to have is going to be absolutely astoundingly positive and beneficial for humanity. But there's also this other side of the debate. Well, I'm going to go there later. I'm going to ask you about that. So we'll talk a little bit about the dark side, but also, you know, I love the framework. I hear that over and over from researchers here at Microsoft Research that are optimistic and saying, and if there are issues, we want to get on the front end of them and start to drive and influence how those things could play out. Yeah, absolutely. There's a way to think about AI where it's mostly about building a set of automation technologies that are a direct substitute for human labor, and

you can use those tools and technologies to cause disruption. But AI probably is going to be more like the steam engine in the sense that the steam engine was also a direct substitute for human labor. And the people that benefited from it initially were those who had the capital to build them because they were incredibly expensive and who had the expertise to design them and to operate and maintain them. And eventually, the access to this technology is fully democratized, and AI will eventually become that. Our role as a technology company that is building things that empower individuals and businesses is to democratize access to the technology as quickly as possible and to do that in a safe, thoughtful, ethical way. Let's talk about you for a second. You've described yourself as an engineering executive,

an angel investor, and an all-around geek. Tell us how you came by each of those meta tags. Yeah, the geek was the one that was sort of unavoidable. It felt to me all my life like I was a geek. I was this precociously curious child, not in the sense of, you know, like playing Liszt piano concertos when I'm five years old or anything. I was the irritating flavor of precocious where I'm sticking metal objects into electric sockets and taking apart everything that could be taken apart in my mom's house to try to figure out how things worked. And I've had just sort of weird, geeky, obsessive tastes in things my entire life. And I think a lot of everything else just sort of flows from me at some point fully embracing that geekiness and wanting, I mean, so like angel investing,

for instance, is me wanting to give back. It's like I benefited so much over the course of my career from folks investing in me when it wasn't a sure bet at all that that was going to be a good return on their time. But I've had mentors and people who just sort of looked at me and for reasons I don't fully understand, and I've just been super generous with their time and their wisdom. And angel investing is less about an investment strategy and more about me wanting to encourage that next generation of entrepreneurs to go out and make something and then trying to help them in whatever way that I can be successful and find the joy that there is in bringing completely new things into the world that are, you know, sort of non-obvious and complicated.

Speaking of complicated, one common theme I hear from tech researchers here on this podcast, at least the ones who've been around a while, is that things aren't as easy as they used to be. They're much more complex. And in fact, a person you just talked to, Anders Heilsberg, recently said, code is getting bigger and bigger, but our brains are not getting bigger. And this is largely a brain exercise. So you've been around a while. Talk about the increased complexity you've seen and how that's impacted the lives and work of computer scientists and researchers all around. I think interestingly enough, on the one hand, it is far more complicated now than it was, say, 25 years ago. But there's a flip side to that where we also have a situation where individual engineers or small teams have unprecedented amounts of power in the sense that

through open source software and cloud computing and the sophistication of the tools that they now use and the very high level of the abstractions that they have access to that they use to build systems and products, they can just do incredible things with far fewer resources and in far shorter spans of time than has ever been possible. It's almost this balancing act. Like, on the other hand, it's like, oh my God, the technology ecosystem, the amount of stuff that you have to understand if you are pushing on the state of the art on one particular dimension, which is what we're calling upon researchers to

do all the time, it's really just sort of a staggering amount of stuff. I think about how much reading I had to do when I was a PhD student, which seemed like a lot at the time. And I just sort of look at the volume of research that's being produced in each individual field right now, the reading burden for PhD students right now must be unbelievable. And it's sort of similar, you know, like if you're a beginning software engineer, like it's a lot of stuff. So it's this weird dichotomy. I think it's perhaps if anything, the right trade-off because if you want to go make something and you're comfortable navigating this complexity, the tools that you have are just incredibly good. I could have done the engineering work at my first startup with far, far, far fewer resources with less money in a shorter amount of time if I were building it now versus 2007. But I think that tension that you have as a researcher or an engineer, like this dissatisfaction that you have with complexity and

this impulse to simplicity, it's exactly the right thing. Because if you look at any scientific field, this is just how you make progress. Listen, I was just thinking when I was in my master's degree, I had to take a statistics class. And the guy who taught it was ancient. And he was mad that we didn't have to do the math because computer programs could already do it. And he's not wrong. It's like, what if your computer breaks? Can you do this? That's fascinating because we have this old fart computer scientists, engineers like me, have this, like we bemoan a similar sort of thing all the time, which is, these kids these days, they don't know what it was like to load their computer program into a machine from a punched paper tape, and they don't know what ferrite core memories are and what misery that we had to endure. It was fascinating and fun to learn all of that stuff. And I think you did get something out of

it. It gave you this certain resilience and sort of fearlessness against these abstraction boundaries. Like, you know, something breaks, like, you feel like you can go all the way down to the very lowest level and solve the problem. But it's not like you want to do that stuff. Like, all of that's a pain in the ass. You can do so much more now than you could then because, to use your you know statistic professors phrase because you don't have to do all of the math right your career in technology spans the spectrum including both academic research and engineering and leadership in industry. So talk about the value of having experience in both spheres

as it relates to your role now. You know, the interesting thing about the research that I did is I don't know that it ever had a huge impact. The biggest thing that I ever did was this work on dynamic binary translation. And the thing that I'm did was this work on dynamic binary translation. And the thing that I'm proudest of is like I wrote a bunch of software that people still use, you know, to this day to do research in this very arcane, dark alley of computer science. But what I do use all the time that is almost like a superpower that I think you get from being a researcher is being able to very quickly read and synthesize a bunch of super complicated technical information.

I believe that it's less about IQ and it's more of the skill that you learn when you're a graduate student trying to get yourself ramped up to mastery in a particular area. It's just like, read, read, read, read, read. Yeah, I grew up in this relatively economically depressed part of rural central Virginia, a town of 250 people. Neither of my parents went to college. We were poor when I grew up and no one around me was in the computers. And like somehow or another, I got into the science and technology high school when I was a senior. And like, I decided that I really, really, really wanted to be a computer science professor after that first year. And so, I went into my undergraduate program with this goal in mind.

And so I would sit down with things like the Journal of the ACM in the library and convince, oh, well, like, obviously, computer science professors need to be able to read and understand this. And I would stare at papers in JACM, and I'm like, oh, my God, I'm never, ever going to be good enough. This is impossible. But I just kept at it. And, you know, it got easier by the time that I was finishing my undergraduate degree.

And by the time I was in my PhD program, I was very comfortably blasting through stacks of papers on a weekly basis. And then, you know, towards the end of my PhD program, you're on the program committees for these things. And like, not only are you blasting through stacks of papers, but you're able to blast through things and understand them well enough that you can provide useful feedback for people who've submitted these things for publication. That is an awesome, awesome, like super valuable skill to have when you're an engineering manager, if you're a CTO or you're anybody who's trying to think about where the future of technology is going. So every person who is working on their PhD or their master's degree right now, and this is part of their training, don't bemoan that you're

having to do it. You're doing the computer science equivalent of learning how to play that Liszt piano concerto. You know, you're getting your 10,000 hours in and like, it's going to be a great thing to have in your arsenal. Anymore, especially in a digitally distracted age, being able to pay attention to dense academic papers and or, you know, anything for a long period of time is a superpower. It is. It really is. You aren't going to accomplish anything great by, you know, integrating information in these little two-minute chunks. I think pushing against the state of the art, like, you know, creating something new, making something really valuable requires an intense amount of concentration over long periods of time.

So you came to Microsoft after working at a few other companies, AdMob, Google, LinkedIn. Given your line of sight into the work that both Microsoft and other tech giants are doing, what kind of perspective do you have on Microsoft's direction, both on the product and research side, and specifically in terms of strategy and the big bets that this company's making? I think the big tech companies in particular in this really interesting position, because you have both the opportunity and the responsibility to really push the frontier forward the opportunity in the sense that you like already have a huge amount of scale to build on top of and the responsibility that knowing that some of the new technologies are just going to require large amounts of resources

and sort of patience you know like one example that we're working on here at Microsoft is we, the industry, have been worried about the end of Moore's law for a very long time now. And it looks like for sort of general purpose flavors of compute, we are pretty close to the wall right now. And so there are two things that we're doing at Microsoft right now that are trying to mitigate part of that. So like one is quantum computing, which is a completely new way to try to build a computer and to write software. And we've made a ton of progress over the past several years. And our particular approach to building a quantum computer is really exciting. And it's like this beautiful collaboration between mathematicians and physicists and quantum information theory folks and systems and

programming language folks trained in computer science. But when exactly this is going to be like a commercially viable technology, I don't know. But another thing that we're pushing on related to this Moore's Wall barrier is doing machine learning where you've got large data sets that you're fitting models to where the underlying optimization algorithms that you're using for DNNs or all the way back to more prosaic things like logistic regression boil down to a bunch of linear algebra. We are increasingly finding ways to solve these optimization problems in these embarrassingly parallel ways where you can use like special flavors of compute and so like there's just a bunch of super interesting work that everybody's doing with this stuff right now like from Doug

Berger's project brainwave stuff and here at Microsoft to some super exciting time, I think, to be a computer architect again, where the magnitude and the potential payoffs of some of these problems was just astronomically high. And like, it takes me back to the 80s and 90s, which were sort of the, maybe the halcyon days of high-performance computing and these like big monolithic supercomputers that we were building at the time. And it feels a lot like that right now,

where there's just this palpable excitement about the progress that we're making. Funny enough, I was having breakfast this morning with a friend of mine. And both of us were saying, man, this is just a fantastic time in computing. You know, like on almost weekly basis, I encounter something where I'm like, man, this would be so fun to go do a PhD on. Yeah. And that's a funny sentence right there. Yeah, it's a funny sentence.

Aside from your day job, you're doing some interesting work in the nonprofit space, particularly with an organization called Behind the Tech. Tell our listeners about that. What do you want to accomplish? What inspired you to go in that direction? Yeah, a couple of years ago, I was just looking around at all of the people that I work with who were doing truly amazing things. And I started thinking about how important role models are for both kids who are trying to imagine a future for themselves, as well as professionals, like people who are trying to imagine a future for themselves as well as professionals,

like people who are already in the discipline who are trying to imagine what their next step ought to be. And it's always nice to be able to put yourself in the shoes of someone you admire and say like, oh, I can imagine doing this. I can see myself in this career. And I was like, we just do a poorer job, I think, than we should on showing the faces and telling the stories of the people who have made these major contributions to the technology that powers our lives. And so that was sort of the impetus have made these major contributions to the technology that powers our lives. And so that was sort of the impetus with BehindTheTech.org. So I'm an amateur photographer. I started doing these portrait sessions with the people I know in computing who I knew had done impressive things. And then I hired someone to help, you know, sort of interview them and write a slice of their story so that, you know, if you wanted to go somewhere and get inspired about people who are making tech, you know, BehindTheTech.org is the place for you.

So you also have a brand new podcast yourself called Behind the Tech. And you say that you look at the tech heroes who've made our modern world possible. I've only heard one, and I was super impressed. It's really good. I encourage our listeners to go find Behind the Tech podcast. Tell us why a podcast on these tech heroes that are unsung, perhaps. I have this impulse in general to try to celebrate the engineer. I'm just so fascinated with the work that people are doing or have done. tech fellow at Microsoft, and who's been building programming languages and development tools for his entire 35-year career. Earlier in his career, he wrote this programming language and compiler called Turbo Pascal. Yeah, I wrote my first real programs using the tools that Anders built. And he's gone on from Turbo Pascal to building Delphi, which was one of the first really nice integrated development environments

for graphical user interfaces. And then at Microsoft, he was the chief architect of the C Sharp programming language. And now he's building this programming language based on JavaScript called TypeScript that tries to solve some of the development at scale problems that JavaScript has. And that to me is like just fascinating. How did he start on this journey? Like, how has he been able to build these tools that so many people love? What drives him? Like, how has he been able to build these tools that so many people love?

What drives him? Like, I'm just intensely curious about that. And I just want to help share their story with the rest of the world. Do you have other guests that you've already recorded with or other guests lined up? Yeah, we've got Alice Steinglass, who is the president of Code.org, who is doing really brilliant things trying to help K-12 students learn computer science. And we're going to talk with Andrew Ng in a few weeks, who is one of the titans of deep neural networks, machine learning, and AI. We're going to talk with Judy Estrin, who is former CTO of Cisco,

serial entrepreneur, board director at Disney and FedEx for a long time, and just one of the OGs of Silicon Valley. Yeah, so it's going to be a really good mix of folks. Yeah. Well, it's impressive. All those fascinating stories. Yeah. And just having listened to the first one, I was, I mean, it was pretty geeky. I'll be honest. There's a lot of, I was like listening to the mechanics talking about car engines and I know nothing, but it was- Yeah, right? Like-

But it was fun. That's great. And like, you know, I hadn't even thought about it before, but like, if I could be like the sort of computer science and engineering version of car talk, like that would be awesome. You won first place at the William Campbell high school talent show in 1982 by appearing as a hologram downloaded from the future.

Okay. Maybe not for real, but an animated version of you did explain the idea of the intelligent edge to a group of animated high school hecklers. Assuming you won't get heckled by our podcast audience, tell us how you feel like AI and machine learning research are informing and enabling the development of edge computing. Yeah, I think this is one of the more interesting emergent trends right now in computing. So there are basically three things that are coming together at the same time.

One thing is the growth of IoT and just embedded computing in general. You can look at any number of estimates of where we're likely to be, but we're going to go from about 11 or 12 billion devices connected to the internet to about 20 billion over the next year and a half. But you think about these connected devices, and this is sort of the second trend, like they all are becoming much, much more capable. So like they're coming online and and the silicon and compute power available in all of these devices is just growing at a very fast clip. And going back to this whole Moore's law thing that we were talking about, if you look at $2 and $3 microprocessor and microcontrollers,

most of those things right now are built on two or three generations older process technology. So they are going to increase in power significantly over the coming years, particularly this flavor of power that you need to run AI models, which is sort of the third trend. So you've got a huge number of devices being connected with more and more compute power, and the compute power is going to enable more and more intelligent software to be written using the sensor data that these devices are processing. And so those three things together, we're calling the intelligent edge. And we're entering this world where you'll step into a room and there are going to be dozens and dozens of computing devices in the room, and you'll interface with them by voice and gesture and a bunch of other intangible factors where you won't even be aware of them anymore. And so that implies a huge set of changes in the way that we write software.

How do you build a user experience for these things? How do you deal with information security and data privacy in these environments? Just even programming these things is going to be fundamentally different. It's a super exciting time, and it's certainly something that we are investing very heavily in right now at Microsoft

in the particular sense of, like, how do we take the best of our development tools, the best of our platform technology, the best of our AI and the best of our cloud to let people build these solutions where it's not as hard as it is right now? Well, you know, everything you've said leads me into the question that I wanted to circle back on from the beginning of the interview, which is that the current focus on AI, machine learning, cloud computing, all of the things that are just like the hot core of Microsoft Research's center, they have amazing potential to both benefit our society and also change the way we interact with things. Is there anything about what you're seeing and what you've been describing that keeps you up at night? I mean, without putting too dark a cloud on it, what are your thoughts on that? The number one thing is I'm worried that we are actually underappreciating

the positive benefit that some of these technologies can have and are not investing as much as we could be holistically to make sure that they get into the hands of consumers in a way that benefits society more quickly. And so just to give you an example of what I mean, we have healthcare costs right now that are growing faster than our gross domestic product. And I think the only way in the limit that you bend the shape of that healthcare cost growth curve is through the intervention of some sort of technology. And like week after week over the past 18 months, I've seen one technology after another that is AI-based where you sort of combine medical data or personal sensor data with this new regime of deep neural networks. And you're able to solve these medical diagnostic problems at unbelievably low costs that are

able to very early detect fairly serious conditions that people have when the conditions are cheaper and easier to treat and where the benefit to the patient, like they're healthier in the limit. And so I sort of see technology after technology in this vein that is really going to bring higher quality medical care to everyone for cheaper and help us get ahead of these significant diseases that folks have. There's a similar trend in precision agriculture where in terms of crop yields and minimizing environmental impacts, particularly in the developing world where you still have large portions of the world's population sort of trapped in this agricultural subsistence dynamic.

AI could fundamentally change the way that we're all living our lives, all the way from all of us getting sort of cheaper, better, locally grown organic produce with smaller environmental impact to, you know, like how does a subsistence farmer in India dramatically increase their crop yields so that they can elevate the economic status of their entire family and community? So as we wrap up, Kevin, what advice would you give to emerging researchers or budding technologists in our audience as many of them are contemplating what they're going to do next? Well, I think congratulations is in order to most folks, because this is just about as good a time, I think, as has ever been for someone to pursue a

career in computer science research or to become an engineer. I mean, the advice that I would give to folks is just look for ways to maximize the impact of what you're doing. And so I think with research, it's sort of the same advice that I would give to folks starting a company or engineers thinking about the next thing that they should go off and build in the context of a company, find a trend that is really a fast growth driver, like the amount of available AI training compute or the amount of data being produced by the world in general or by some particular subcomponent of our digital world. Just pick a growth driver like that and try to attempt something that is either buoyed by that growth driver or that is directly in the growth loop. Because I think those are

the opportunities that tend to have both the most headroom in terms of, you know, like if there are lots of people working on a particular problem, it's great if the space that you're working in, the problem itself has a gigantic potential upside. Those things will usually like accommodate lots and lots and lots of

sort of simultaneous activity on them and not be a winner takes all or a winner takes most dynamic. You know, and they're also sort of the interesting problems as well. You know, it's sort of thrilling to be on a rocket ship in general. Kevin Scott, thanks for taking time out of your super busy life to chat with us. You are very welcome. Thank you so much for having me on. It was a pleasure. To learn more about Kevin Scott and Microsoft's vision for the future of computing, visit Microsoft.com slash research.