The Vergecast - Microsoft CTO Kevin Scott on AI for rural America

Episode Date: April 7, 2020

Verge editor-in-chief Nilay Patel talks to Microsoft CTO Kevin Scott about his new book Reprogramming the American Dream: From Rural America to Silicon Valley―Making AI Serve Us All. Learn more about your ad choices. Visit podcastchoices.com/adchoices

Support for the show comes from Retool. Too many companies run critical operations on duct taped spreadsheets, Slack workflows, and whatever else they could cobble together. Not because they want to, but because building internal tools means weeks of waiting on someone else's backlog. That's where Retool comes in. Build custom internal tools just by describing what you need. Prompts something like,

Build Me a Revenue Dashboard on our Salesforce data. And Retool actually builds it on your company's data, in your cloud with enterprise security built in. Go to retool.com slash Verchcast. We all need to retool how we build software. What's up, y'all. I'm Skyler Diggins, seven-time WMBA All-Star, Olympic gold medalist, and mom. And I'm Cassidy Hubbard, host and reporter for nearly 20 years covering the biggest names and stories in sports and mom. And this is Am Mom, a community for athletes, game changers, and moms of all kinds.

dropping May 14th. Tap in with us. Hey everybody, it's the United from the Vergecast. On this week's interview episode, I talked to Microsoft CTO Kevin Scott. Kevin's just written a book called Reprogramming the American Dream. It just came out this week. It's all about AI, how it's inevitable, and how it's going to remake the American economy, particularly in rural America, which is where Kevin's from.

He's at a really interesting path. He was at Google for a while. He left. He joined AdMob. He came back to Google. He left. He joined LinkedIn. LinkedIn got acquired by Microsoft.

and he's ended up as Microsoft's CTO, where he thinks on a longer time scale than Microsoft's other business units, which have to ship products and make money. Kevin's job is to think big, to think about where the company's investment should be. We had a really great conversation about AI where it's headed, the bets he's made, the wrong way to think about training our models, the right way to think about those models, and really how it's going to remake economy. This was a really interesting conversation, especially in the context of the coronavirus pandemic. Both of us are obviously talking to each other from home.

There's a moment, I think, where we're going to reset a lot of our assumptions about how society might work after this. It's a good opportunity to rethink some of those core assumptions. And AI is going to be right at the center of it. So a really interesting conversation. Hope you like it. Check it out. It's Microsoft CTO Kevin Scott.

Kevin Scott, you're the CTO of Microsoft. Welcome. Thank you very much for having me. So I just want to start with some background. You're fairly new to Microsoft, and you've had kind of this winding road. You've just written a book about a future of AI in America. There's a lot of stuff to talk about it. But give people a sense of your background and how you've arrived at this position at Microsoft.

Yeah, it's sort of a meandering path, really. And I write about a lot of it in the book. So I grew up in rural Central Virginia in the 70s and 80s. Like I had the good fortune to get interested in personal computers at exactly the moment where personal computers were becoming a thing. And I, for the first half of my adult life, like, I thought it was going to be an academic. So I went to study computer science at university and went to graduate school and thought I was going to be a professor.

And I met my wife, who is a historian. She also thought she was going to be a professor. And we moved to Germany together. And, you know, I had this epiphany that that really wasn't what we wanted to. to do, and we move from Germany to the United States, and then I joined Google in 2003 before the IPO and have had this really wild and interesting ride since them. So you were at Google, and then you left to go to AdMob? Yes.

And then you were like reacquired by Google. Yes. And then you went to LinkedIn, and then you were acquired by Microsoft. Is that about right? That is about right. Very good. I read, so I read the book. You did.

I'd pay attention. I'm so happy. The book, by the way, is called Reprogramming the American Dream. It's about how AI might change society, particularly rural society, which I think is worth talking about. But I want to talk about Microsoft for one more second. What does the CTO of Microsoft do? This is like a new role for the company and the way that I think your conception of it is

being sort of managed, right? Yeah. Microsoft had a, had a CTO a long time. time ago, a guy named Nathan Mirvald, who, like, I really admire a whole bunch of things that Nathan does. Like, he's a polymath and Renaissance man, but he left Microsoft a long while ago. And, like, there have been people who played a CTO-ish role there, so Bill did it for a while at Ray Ozzie and Chi Liu, who was running a big part of the Microsoft engineering team, played that role for Satya for a while.

And the way that I came into the role was Microsoft had just acquired LinkedIn and she had almost exactly the same time decided to retire. And Microsoft, like Sati in particular, was looking for someone to help him do some of this work in technology that cuts all the way across all of the engineering groups because it was falling disproportionately on him to sort of spot repetition of work. or holes in strategy or, and like, you know, we had a bunch of things like AI in particular. Like I've spent probably 75, 80% of my time over the past three years that I've been in this role, helping to make sure that the company's AI strategy makes sense, that we're building the right infrastructure, that like we are hiring the right people, that we are making the appropriate investments given where the field is going. And so I do a whole bunch of things like that.

So I run our AI supercomputing efforts that like span the entire company. So all the engineering groups use the outputs of this stuff. I'm helping us to make sure that we have the right strategy on edge computing and silicon and like a whole bunch of fun stuff. And I also now run Microsoft research and a bunch of our technology incubation teams, which is a ton of fun. And ironic in a way because when I was a PhD student, I was an intern at Microsoft Research, and I could not have imagined in a million years back then that I would one day be responsible for this unbelievably awesome research organization. Yeah.

So your focus here in the book is AI. But obviously your job at Microsoft spans a lot of other things. What does that balance look like for you? I mean, I think of Microsoft research, to this day, and this has not a lot of relevance anymore, but to this day, I think of, like, the courier video, which, like, posited this dream of a folding tablet that would just, like, go around with you. I think of the very first surface table video. Yep. Right?

Like, the idea was we're going to push the concepts of what we can do in the future. Is that a big part of your role? Yeah, it is. The way that I look at my job is I sort of do a similar set of things to my engineering colleagues. So Rajesh Jha, who runs the experiences and devices group and Scott Guthrie who runs the cloud and AI group and Phil Spencer who runs gaming. So they're all running these big engineering teams and they have to think about their technology strategy and how they're delivering all of this stuff. But we sort of spend our time just sort of in a different mix of the time horizons that we're paying attention to. So, like, they really have a lot of stuff that they have to deliver in a quarter and in a year.

And they spend a huge amount of their time and energy sort of focused on what their groups need to do to make sure that, like, all of this really complicated stuff gets done in that time horizon. And then they also are thinking about the future. I spend more of my time thinking about what it is that we need to be doing two to five years from now that we will very much regret if we're not investing in the moment. Like what are the technology trends? What's the competitive landscape look like? Where do we want to be pushing where like we believe that we've got a point of view about what the future ought to look like and where are the sort of inexorable trends and technology going that whether we like it or not we're going to have. have to react to. And I do spend some time thinking about, you know, like what we need to deliver this year. Like the AI supercomputing stuff is a good example. Like we, we have a very rapidly

growing investment that we're making in the infrastructure required to train these very, very big AI models that the world is building right now. And there's just a bunch of blocking and tackling that has to happen to go make sure that infrastructure gets built in the right way and delivered on time. But to your point about Microsoft research, the whole purpose of having a research organization like that is that you can take risk and make these explorations into what the future might look like or could look like that are very hard to make when like your accountabilities are all about, you know, the next 12 months. Yeah. So let's talk about AI. You just just said you've got to train the models the right way. My question is, what's the wrong way?

Well, the field is changing fairly dramatically. And it's a bunch of ways to answer that question. There's certainly all of the sort of fairness and bias and ethical use questions. And there's a bunch of wrong way for building things on that dimension. There is, and this is part of the reason that I wrote the book, there is, once you have the technology and it is fair and unbiased and being used ethically, there's still the questions around who has access to the technology and what is it getting used to do for the good of the whole public. And then there are just a bunch of technical things that are changing very, very quickly in the field that make it very difficult to make high conviction prognostications about what's going to happen. So like we've had this really

startling shift from supervised machine learning to unsupervised learning in the past two years, where, you know, two years ago people would have like laid down good money that supervised machine learning may be the one and only way to go build things because we'd had so much success with it, like all of the speech recognition and computer vision things that are happening or were all built on supervised deep learning. Kifu Lee wrote this really good book where he just drove a stake in the ground and said, you know, like, we are in the era of supervised deep learning. And, you know, and like that's, I mean, he didn't say that's all that there's ever going to be,

but like he was making a very big argument about like how that was the most important thing in the world. And then almost immediately after that happened, like this, these just remarkable things. started happening with reinforcement learning and unsupervised deep learning for natural language processing that have really revolutionized the entire field. And so it's like all three of those things. It's like the fairness bias ethics, there's the, you know, like are you building this set of technologies as a platform that's accessible to everyone and that can be used by anyone who needs to create technology? And then there's the like, are you really? really paying attention to what's happening with the fundamental trends with the technology

so that you can make sure that your investments are aligned with where the science is going. So when you think about where the science is going, in the book, there's this concept that you write about, which is there are these periods of frenetic activity in AI, and then there are sort of AI winters where things slow down, where obviously it feels like a period of frenetic activity. But is something like supervised learning? Does that just hit a maximum of what it can do and you've got to move on to the next model? Are they happening in parallel? How do you evaluate those kinds of choices?

Yeah, I think there's a whole bunch of things happening in parallel. It's not like supervised learning is over with. And like one of the things that we're even seeing with these big unsupervised models that we're building now is one of the really attractive things about them is that transfer learning is finally working. So you can train a model that's very general. and then either with no fine-tuning at all or with some fine-tuning to a particular task, and the fine-tuning is usually supervised, you can get them to do an incredible breadth of things that weren't possible before. So, yeah, I do think that there's more of a mixture of things being explored right now than

there were before. And, like, there are a whole bunch of people who, you know, if you look at what's happening with the unsupervised learning, the compute demand is going up by almost an order of magnitude a year. It's something like 8x every 12 months. And it doesn't take too many years on that curve before you are using every iota of available compute on the planet Earth to train models. And so, like, obviously, that won't be sustainable for long. And, like, you'll reach some point of, you know, diminishing marginal returns either on the amount of money, that you're using to pay the models,

or the models performance themselves will start to flatten out. And so there are a bunch of people doing like really interesting fundamental work on what are the alternatives to these approaches. You know, it's back propagation for, you know, in these deep neural networks, is like that the only way to do machine learning.

You know, there's a computer scientist, an entrepreneur Gary Marcus, who is a vocal critic of the current path that a large number of people are on for deep learning. And like, you know, his assertions are, yeah, and so its observations are actually correct, that, you know, we're building these incredibly sophisticated systems. They are brittle in ways that no human beings, intelligence actually is. And in some cases, there are problems that, you know, your two-year-old can solve. Even if, you know, like a toddler who is pre-linguistic,

like they can't speak yet can solve problems that the most sophisticated AI cannot. And so like that's, I don't know, like it just, I think there's so many exciting opportunities here. Like we are far, far, far away from having all of the problems solved. Yeah. And I think that brings me to kind of where your book is really focused, which is this is happening. It will change our relationship to computers in a very serious way. And it will change our relationship to work in a serious way. And that will have some impact, particularly on rural America, that we have to actually think about now.

There's a relationship there that I just want to pull apart, which is we don't know how it's going to work. Right. I mean, that's the conversation we've been having. It's a lot of very smart people are working on the mechanics of AI right now and having conversations about where the current approach might be failing or succeeding or inventing new approaches. But if you abstract that out, everyone is pretty convinced it's going to work. And now you're talking about what are the consequences of it. Yeah. Well, I think everyone is convinced that AI is going to be able to accomplish increasingly useful things over time.

Like there's more of a question around whether or not you ever get to artificial general intelligence. So like that is that's AI that is in the full. full general sense, the equivalent of a human being cognitively. So some people, you know, think that we're, you know, pretty close to that. And some people think that we're, you know, decades away from that. And some people think that it's never going to happen. What do you think? So what, what I think is, this is, this is, this is, I hate to say it this way because I sort of sound like a politician. But like, the thing that I really do believe is, when you believe or if you believe that AGI is going to happen, I believe that pursuing it is

extremely worthy. It is driving a bunch of advancement in technology that is going to be useful, that is useful, even if you never get to full AGI. And like, even the study of AGI itself shines a light on the nature of human intelligence, which I think is very well understood. Like, this is sort of one of the complicated problems about having a discussion about artificial intelligence because, you know, we make all of these false equivalences between human and artificial intelligence. And, like, the fact of the matter is, like, human intelligence itself is extremely poorly understood. We understand an increasing amount about the biology of the human brain.

But, like, human cognition is, like, a really, you know, sort of fuzzy thing. Like, one of the things, for instance, has happened over and over again. through the history of AI is that we set challenges for AI that we think are like high watermarks of human cognition. Then we go solve those problems like beating grandmasters at chess or Go or multiplayer online games or speech recognition or labeling, labeling the objects and still or moving images. These are all things that computers can do just as well or better in some cases than human beings can. And as soon as we, you know, like we have this repeated pattern of as soon as we

solve one of those problems, we're like, okay, well, maybe that isn't actually the thing that, you know, it makes a human being smart. So yeah, look, I think, you know, my opinion is that we have lots of very interesting things coming out of the pursuit of AGI right now that are going to be very, very good tools for folks to go use to build things that matter. Do you think it's going to happen? I ask every self-driving car person when they think self-driving cars will happen. I feel like my next move is to ask every AI person when AGI is going to happen. Well, I really don't know when. I am one of the people who think that we will get there at some point. I don't think it's five years away, though. Fair. So it brings us back to sort of the first part of this conversation, which was we think these things are going to happen and they will have some impact on how we live and work. And that, again, is more what your book is about than the mechanics of how it might happen. So as you were thinking about writing a book like this, what were the things that you really focused on in terms of, A, convincing people, this is happening, it's going to happen,

you have to take it seriously, and B, here are the impacts and choices we're going to have to make. One of the big aha moments that I had early in the process of writing the book is I went home to Central Virginia. and I had this set of preconceived notions about what I was going to see when I chatted with people there who were running their businesses and, you know, like how they were thinking about technology, like how or if they were thinking about AI and the impacts it might have. And I was just really blown away and sort of reminded of something that I knew from my childhood, which is in all of these communities, there are ingenious, industrious people who are. doing very clever things already with technology. And that some of the most successful businesses that you can find in rural and middle America are making use of the best possible tools that they can find.

So a good example of this is, you know, like one of my childhood friends is a manager at this company that does precision plastics machining. So they make these very intricate parts out of plastic for customers all over the United States. They literally set this business up in a building that used to be a textile mill, which is one of the three industries that got sort of hollowed out by globalization in the 80s and 90s and in the part of rural central Virginia where I grew up. And that entire business only exists because of technology and the humans who are able to harness it to do these.

interesting things. So they use the internet to market their services and to communicate with their customers. And then they're using these very high-performance CNC machines to actually make the parts. And they have these high-skill workers there who are programming the machines to build the things that their customers want and need. And it was just a reminder to me that, you know, this thing that my grandfather taught me when I was super young. that people can accomplish incredible things when they have the tools to accomplish them. Yeah, because I have an enormous amount of faith in human ingenuity. And as I look at these businesses in my hometown,

it's not like they're using huge amounts of AI right now, but it's very easy to see how everything that they're doing is going to benefit from AI, where they're going to get more competitive, they're going to create more of these high-skill jobs that are going to have this positive impact on the economies of these communities that are spread out across the country. That in a certain way, like for manufacturing, for instance, there's almost like a Moore's Law effect going into these machines. So the amount of capital used to acquire a machine that has capabilities to produce something valuable is going up. So, like, you know, for a small business loan, you can go, you know, get yourself a manufacturing capability, whether it's these CNC machines or 3D printing or, you know, like these increasingly agile manufacturing technologies where you can start a really competitive business, you know, doing very interesting things serving markets that exist, like where you don't have to, you're not having to invent things out of whole cloth.

And all of that benefits from, from things like AI. Like AI makes the machines more capable, which means that they can be used by businesses to make themselves more competitive and to like do more interesting things. So that was like this big aha moment that I had. And as soon as I saw a handful of these in my community, then I started looking around and like you can sort of spot these companies everywhere. And I will say like the other revelation too and like something that I just sort of knew, but like wasn't really. as focused on it as I should have been is the very first machine learning project that I did in 2004, I was coming into the field new, like I had computer science degrees at that point and had been programming since I was 12 years old. I was 30 something at the time. So I had been doing this stuff for a while.

But it still took me, you know, weeks and weeks of staring at graduate level textbooks and research. papers to get the knowledge into my head that I needed to go do this project. And then I spent six months writing a bunch of complicated code to like do this particular thing. And, you know, when I think about what it would take now, because of open source software and cloud platforms and all of the educational materials that are available online, a high school kid in a weekend could do the same thing that I did 16 years ago in six months with like this high barrier to entry. And the barriers to entry to using the tools of AI are going down very fast, which means

that like there are going to be more and more opportunities for these businesses, the same way that they're employing this very advanced manufacturing technology to create opportunity for themselves and their employees and their communities. you're going to be able to use these same tools. And the thing that's standing in their way is like prosaic stuff, like educated kids in high school. And again, like it's not hard stuff to learn anymore. And it's, you know, just super prosaic stuff like broadband access.

Like you cannot expect to connect yourself to a digital future where opportunities are going to exist in the form of technology platforms. If you can't even connect to the internet, it's just insane that we're having this conversation right now in 2020. I mean, I want to talk about nothing but rural broadband policy and broadband access. But I want to stay on, and we're going to, because you run white spaces at Microsoft too, right? My colleague, Brad Smith, runs white spaces, but my teams are heavily involved. It's a bunch of technology that came out of Microsoft research. I definitely want to talk about white space internet, but not right this second.

I want to come back to a scene from your book where you're talking about high school kids. There's a big data center in Virginia. There's a senior book where you're gazing at the enormous data center you have there, and then you go to a job fair and no one is signing up to work at the data center because I don't know what it is. That seems like just the most unexpected result, right? Microsoft came to town. They built a huge building. They've got the sign up at the job fair that says come work at the huge.

fancy building and no one shows up. How is that happening? How do you get over that hump where it seems like everyone knows that tech is the future, but the actual stand in this line to get the job application part of it doesn't seem to be, that loop doesn't seem to be closed? Yeah. In a sense, that is the biggest reason why I wrote the book, because I think the thing that you need in these moments are stories and role models that appeal to people who are trying to imagine their own future. Like I know when I was growing up, my dad, my granddad, and my great-grandfather were all carpenters.

They, you know, they made their living in the construction industry. And even though I knew pretty early on that I really love programming and I love technology and like I wanted to try to figure out a career doing that, I didn't have many role models. Like there wasn't anyone in my family who was an engineer who had. had a degree in science or technology or had a career in science and technology. And, you know, I had these moments in high school when I was thinking about my life and how I was going to earn a living and, you know, how I was going to have a family and where I was going to live where it didn't seem like a crazy possibility to me that I was going to go into the family

business. And my my dad was like stridently against it. And I don't want to leave anyone with with the wrong impression. I think there's an enormous amount of dignity and like just critical needs across the board for, you know, for folks to have careers like the one that my dad had. He, for his own reasons, was determined that I wasn't going to do it. But it had appeal. And like, I just, I really do appreciate the quandary that these, these kids are in. Like when you, you know, when everybody in your community is doing a particular set of things and like those are the paths that you can see to like a future life for yourself asking someone to imagine doing something that seems super abstract like even if you've got the promise of you know some better economic outcomes

like I think that's hard and so like we have to do a better job of of showing kids like what these paths could look like and why they're interesting and like why you know it's not just about the money I don't think you can't tell them it's like oh you know you're you're going to be able to earn more money having a career in tech than following the path of your farmer parents or your, you know, your oil industry parents or, you know, your parents who are, you know, mechanics or like whatever it is that their path was, you have to show them why it's going to be fun and interesting and compelling and the work you're doing is going to matter to other human beings. It just can't be this abstract thing, I don't think.

Support for this show comes from Shopify. Starting something new isn't just hard. It can be really scary, too. So much work goes into this thing that you're not entirely sure will even work. But here's a better thought. What if it did all work? What if your instincts were actually right all along? Shopify wants to help you get there.

They're the commerce platform behind millions of businesses worldwide and nearly 10% of all e-commerce in the U.S. from established brands like Allbirds and Hines to companies just getting started. Their design tools make it simple to create the exact online presence you're envisioning with hundreds of ready-to-use templates available. And with built-in marketing tools, you can launch full email and social campaigns in just a few clicks. So you can connect with customers wherever they are. It's time to turn those what-ifs into with Shopify today. You can sign up for your $1 per month trial today at Shopify

shopify.com slash vergecast. You can go to shopify.com slash vergecast. That's shopify.com slash vergecast. Support for the show comes from Upwork. The days of doing it all, all by yourself, are over. There's no romance and burning out while you're trying to scale. Instead, you can check out Upwork. Upwork helps grow your business by giving you fast access to specialized talent across more than 125 categories so you can fill skill gaps, launch projects faster, and scale without committing to full-time headcount. And finding the right talent is easy. You can browse profiles, review past work, and get help scoping the role so you can get

started quickly. Seriously, you could connect with the right freelancer in just a few hours, especially when you sign up with Business Plus. Their AI-powered shortlisting pairs you with the top 1% of talent in under six hours. no endless searcher required. You can visit upwork.com right now to post your job for free. That's upwork.com to connect with top talent ready to help your business grow. That's upw-R-K.com.

Upwork.com. So we're obviously talking. We're in the middle of the pandemic. We're both sitting at home. Millions of people are unemployed in this country right now. The common narrative around AI is it's going to take even more jobs away. And your argument in the book is like, that is absolutely not true, we'll actually create jobs. How do you make, I mean, that seems like a hopeful argument to make

in this moment, but it also seems like the right time to make that argument. How would you frame that conversation? Yeah, I think it's complicated. I mean, the reason that I wrote a book, rather than try to have this argument in, you know, on social media is that I do think it's, it's like very, very complicated. And there's a lot of nuance there. And it's just dangerous to, you know, sort of reduce things down to, you know, one point of view or the other. But like I do, I very powerfully believe that AI is going to result in not just more, you know, sort of public good and societal benefit than perhaps any other tool that human beings have ever invented. I also think that it will create a huge number of new jobs. If it doesn't,

it will be the only technology that we've ever invented that hasn't had that property, which would make it very strange. And so, you know, I think it creates a job, jobs in a whole bunch of ways. So one is, you know, when I, again, wrote my first machine learning program and, like, did my first real machine learning project in 2004, there was no such thing as a data scientist. And like now, it's one of the fastest growing hottest jobs in the world. And so there are already all of these jobs that are being created. by AI, like their whole companies that do nothing other than prepare data and, like, help with the training of AI models.

Like, there's a bunch of stuff at a variety of different skill levels that people are doing now and earning a living from that those jobs wouldn't exist if it weren't for AI. Give me an example. Well, so there's this company called Scale AI that's here in Silicon Valley that, is doing a bunch of work to help train supervised models for things like the autonomous driving companies. And so, like, they've got a very interesting business that only exists because you've got a bunch of people who are working on supervised models for a bunch of different applications. And, like, one of the hard things about doing supervised machine learning is, like,

you have to have people who are teaching the models how to do what you need them to do. And the teaching right now is relatively primitive. It's mostly labeling and sort of data, data cleaning. But in the future, we are going to have much more sophisticated ways of machine teaching. And like, this is one of the hopeful things, you know, I think about AI in general, is programming. Programming's barrier to entry is higher than machine learning, I would argue, in the sense that to become a programmer, you actually have to understand a fair amount about the sort of complexity and idiosyncrasies of computers, digital computers.

And they're non-intuitive in a whole bunch of ways. And so, like, you have to learn all of this complexity. And then you have to learn how to translate human understandings of problems into terms that the computer can solve them. And, like, it requires, like, a whole bunch of practice and education and whatnot. One of the promises of machine learning is you sort of transfer or you transform this problem of harnessing the power of the machine from a programming challenge to a teaching challenge. And we all know how to teach.

Like your two-year-old can probably teach another two-year-old how to do something. Like it's just an innate part of how human beings work. Like we are all learners and we're all teachers. And so the prospect of having a set of. of tools where you can get a computer to do work for you by teaching it how to do something, super exciting. And like, not only will that create jobs, it will make it possible to incorporate AI machine learning into businesses and a bunch of ways that we're not imagining. You know, so the real, you know, potential benefits of AI, like I think are sort of twofold.

So there's one that it lets you solve a bunch of problems that are difficult or impossible to solve otherwise. And like, we can talk about that in a few minutes just in terms of like some of the like biosciences things that are happening at the intersection of AI like that are addressing the pandemic that we're at right now. But the other thing that like plays where AI I think creates real opportunity is where we use it as a tool to sort of uplift human beings. Like imagine if you could make a list of the 10 most repetitive, annoying, mind-numbing things. that you do every day and like you had a piece of software that could go capably solve those problems for you. You know, just like you would with word

processors or spreadsheets or whatnot, you would welcome the existence of that software because like if you're anything like me, I know that you believe that you could get so much more done in a day if you could have just a little bit of help. And like that's the idea with AI. It's not that like, oh, we're going to build, you know, Commander Data and, like, he's going to come in and do your job for you, you know, at half the price and, like, never call out sick. Like, that's not the, that's not the thing I think people need to be worried about. I feel like you just really threw some shade of data right there. Oh, data's awesome.

Like, I don't, I don't know whether you have watched Picard. I have not yet. Yeah, so I won't spoil it for you, but let me tell you there will be tears for you in the, the, the, the, finale and they are related to data who I think is actually one of my favorite characters in modern fiction. Interesting. Yeah, I mean, look, data in Star Trek as a narrative device is more about humanity than about AI. So like that character, like let them write all of these stories that sort of shines a light on what it, what it is to be a human being than you would otherwise be able to do. And like, I think that is a very interesting thing about AI right now. Is it like is shining a

light on what it is to be human? And a lot of the people who are suffering right now, the joblessness being created by the pandemic, the jobs that are being lost, I don't think are easy to replace with AI. Like they are like in almost to a job. Like they are the hardest possible things to imagine AI being able to do because they are direct, you know, human-to-human contact jobs. And I don't think we want them to go away. No, I don't either. I think that the way I sort of broadly been thinking about it is this is a, I've never lived through anything like this.

Most people haven't. It is an opportunity to reset a bunch of assumptions about how the society works coming out of this. And what makes this unique and interesting time to talk about AI for me is, well, you could just bake AI into those assumptions about how you're going to reset society. You're going to say there's a bunch of dangerous jobs that actually we don't think people should go back to. We think there should be robots or there's a bunch of extremely menial repetitive tasks that we don't think people should do anymore. That should be an AI task. There's a bunch of dangerous

healthcare tasks. You just brought up healthcare. There's a bunch of very dangerous healthcare tasks. Or tasks, I think in the book you bring up cardiology, right? I think you're an investor in a cardiology startup. There's a bunch of, hey, actually, we need to listen to people's lungs, like a lot now. And we need that to be more perfect than ever to say, okay, this is a candidate for testing. AI can help there. That stuff to me seems additive. But I think the conversation about AI, if there are lots of other people who are making this claim, is it's going to take away a bunch of jobs that were there before. And maybe those workers aren't going to have the skills to go the next level up into the jobs you're talking about. Maybe. Maybe. I mean, the two things that I

will say directly about that is I think it is harder than people imagine to take any AI system that we have right now and to make it a full and complete and equivalent substitute for the human beings doing a job. Like, it's easy to imagine how a piece of AI automates a task. It's very difficult, like even in the narrowest of things, you know, like customer support, for instance, to imagine, like, how, uh, how an AI agent can, uh, be a full substitute for humans. And, and, you know, the people who understand this, businesses who understand this can get better outcomes. So, you know, like if you are, you know, a, you know, sort of an evil oligopolis, uh, you know, and you've got, you know, I'm not suggesting that these people actually exist,

but it's sort of the, it's sort of the caricature that a lot of people paint. Like if you think that your job is to ring every bit of cost out of a system that you're running as humanly possible. And like you're thinking of AI, you know, is this tool that helps you wring out the cost. And you're not thinking about how it is that you can use human ingenuity, human capital, human industry, and talent to help you solve problems. Like, I think you're going to get into trouble. Like, even with customer service, like, I've, I've chatted with people who've tried to do things in both ways to, like, make it a full substitute because, like, they just sort of see

their customer service as a cost center that they need to optimize away. And there have been other people who say, okay, well, maybe if I take my existing customer service people who are an asset to my company and I use a bunch of AI tools to help them do the things that they are uniquely able to do. Like, I can get better customer service for my customers. I can have happier customer service agent. I mean, so like, let's say you've got a bot that does the tier one customer support. Like, it answers the, you know, the most obvious questions. And then its job when it can't is to like very, very quick. quickly traffic direct to a human being who can actually do rich, complicated problem solving,

who can, like, understand what a customer is going through, not just their problem, but like, the emotional state that they're in, who can, like, radiate empathy and, you know, just sort of reciprocate some of the emotion, like form a connection. Like, hey, I can't do that yet. And, like, I don't see the time horizon where it can. And so, you know, like, again, if you are thinking, about it is like, oh, this is ready to go. It's the full substitute for human beings. I'm going to use it to optimize it. You're going to just end up with crappy customer service and you're going to get sort of caught into this trap where like that is not a situation that you can innovate from. Whereas if you are thinking about like, okay, like I love these people that I have who are working for my company are fantastic and like I want to give them better tools so they can do an even better job. Like my God, like you're going to get so much more. You're going to get so much more. out of that than this other option. So let's talk about healthcare, because it's something you do focus on the book. This is a particularly poignant time to talk about healthcare.

How do you see AI helping broadly with healthcare than more specifically with the current crisis? Yeah. So I think there are a couple of things going on. One, you know, I think is a trend that I wrote about in the book and that is just getting more obvious every day that we need to do more. And so that particular thing is that if our objective as a society is to get higher quality, lower cost, health care to every human being who needs it, I think the only way that you can accomplish all three of those goals simultaneously is if you use some form of technological disruption. And I think AI can be exactly that thing. And you're already seeing an enormous amount of progress on the AI-powered diagnostics front.

And just going to the crisis that we're in right now, like, one of the interesting things that a bunch of folks are doing, including, I think I read a story about the Chan Zuckerberg Initiative is doing this. So, like, the idea is that if you have ubiquitous biometric sensing, so, like, you've got a smartwatch or a fitness band or, like, you know, maybe something even more complicated that can. sort of read off your heart tick data that can look at your body temperature, that can measure the oxygen saturation in your blood that you can sort of like basically get a biometric read out of like how your body's performing and it's sort of capturing that information over time, that we can build models, diagnostic models that can look at those data and determine whether or not you're about to get sick, you're sick, and, you know, sort of predict with reasonable accuracy like what's going on and what you should go do about it.

You know, the great thing about that is like it's not possible. Like you can't have a like a cardiologist following you around all day long. There aren't enough cardiologists in the world even to give you a like a good cardiologic exam at your annual checkup. And so like you can have this great societal benefit by having like this sort of ubiquitous diagnostic capability where you can determine whether or not people are ill. before they even know that they're ill when it is cheaper to treat the illness and where you get to better outcomes. And so like the CZI thing that I was referencing, they have these rings that they have developed that gather biometric data. And like the idea is, and like, you know, we don't know whether this will work yet or not. but the idea is like could you from biometric data predict when somebody has COVID-19 before

they're symptomatic. And if you could do that, like even if it wasn't a foolproof test, like, you know, that would be a great signal that says go get yourself tested. Like go self-quarantine. Like go, you know, maybe get on one of these palliative like potential remedies that will help you, you know, fight the infection in a way that doesn't land you in the hospital. And so, like, I think that is, it's a very, like, this isn't a far-fetched thing. It's not, like, you know, there is a path forward here for deploying this stuff on a broader scale, and it will absolutely lower the cost of health care and help make it more widely available. So, like, that's one bucket of things.

The other bucket of things is, like, just some mind-blowing science that gets enabled when you intersect. AI with the leading edge stuff that that people are doing in the biosciences. And like, I'm happy to talk more about that if, you know, if folks are interested. I'm absolutely interested. Give me an example. So, you know, like two things that we have done relatively recently at Microsoft. One is one of the big problems in biology that we've, we've had that immunologists have been studying for years and years and years is whether or not.

you could take a read out of your immune system by looking at the distribution of the types of T cells that are active in your body. And from that profile determine what illnesses that your body may be actively dealing with. What is it prepared to deal with? Like, you know, what might you have recently had? And that has been a hard problem to figure out because, like, basically, you're looking, you're trying to build something called a T-cell receptor to antigen, map. And, you know, like, we now with like our sequencing technology, like, we have the ability to, like, get the profile so you can sort of see what your immune system is doing. But, like, we have not yet figured out how to build that mapping of, like, the immune system profile to

diseases, except, like, you know, we're partnering with this company called Adaptive that is doing really great work with us, like bolting machine learning onto this. problem to try to figure out what the mapping actually looks like. And like we, we are rushing right now a serologic test, so like a blood test that we hope will be able to sort of tell you whether or not you have had a COVID-19 infection. So I think it's mostly going to be useful for understanding the sort of spread of the disease. Like I don't think it's going to be as good a diagnostic test as like a nasal swab and like one of the sequence-based tests that are getting pushed out there. But it's like a really interesting.

And the implications are like not just for COVID-19, but like if you are able to better understand that immune system profile, like the therapeutic benefits of that are just absolutely enormous. Like we've been trying to figure this out for decades. The other thing that we're doing is like when you're thinking about SARS-CoV-2, which is the like the virus that causes COVID-19 and that is, you know, sort of, you know, ranging through the world right now. We have never in human history had a better understanding of a virus and, like, how it is attacking the body. And we've never had a better set of tools for precision engineering, potential therapies and vaccines for this thing. And like part of that engineering process is like using a combination of simulation and machine learning and like the, you know, these cutting edge techniques of biosciences in in a way where you're sort of leveraging all three at the same time.

So like we, you know, we've got this work that we're doing with a partner right now where I have taken a set of super computing clusters that we have been using to train natural language processing deep neural networks at just massive scale. And those clusters are now being used to search for like vaccine targets and therapies for SARS-COB too. That's incredible. Yeah. It's it's you know and look we're we're one among a huge number of people who are like very, very quickly searching for for both therapies and potential vaccines and, you know, we, there are reasons to be hopeful, but like we've got a way to go. But it's just unbelievable to me to sort of see how these techniques are coming together. And like one of the things that I'm hopeful about as we deal with this current crisis and, you know,

think about what we might be able to do on the other side of it is it could very well be that this is the thing that triggers revolution in the biological sciences. and investment and innovation that has the same sort of decades-long effect that the industrialization push around World War II had in the 40s that basically built our entire modern world. Yeah. I mean, I keep coming back to this idea that this is a reset on a scale that very few people living today have ever experienced. Yep.

Like you said, out of World War II, a lot of basic technology was invented, deployed, refined, and now we kind of get to layer in things like AI in a way that it's a way that it's is quite frankly remarkable. Yeah. I do think, I mean, it sounds like we're going to have to accept that Cortana might be a little worse at natural language processing while you search for the protein surfaces. But it's a trade I think most people will make. I think that's the right trade off.

I agree. So I promise that we're going to talk about white spaces. This is one of my favorite little bug bears. I've been covering it for 10 years. Do you want to tell people what it is or you want me to do it real fast? Yeah. So the basic idea is that in rural places, it's very, expensive to get wired broadband infrastructure into everyone's home.

What we've done with white spaces is we use unused TV broadcast spectrum to set up wireless transmitters and receivers that bring broadband into these sparsely populated areas of the country. Super cheap. It's like very easy to deploy. You can put a You know, a transceiver at someone's home. The technology is less than a couple hundred bucks now. And the receivers, the big antennas that you have to put up are like also not expensive at all. And so like we believe that this might be a scalable way to get the sparsely populated parts of the country like dramatically better connected than they are now. Even, okay.

So Microsoft has been added for 10 years as near as I can tell. Does it work? It does. I've never seen it work. In 10 years, I've never seen it work. Really? Maybe I just kind of come visit y'all and you can show it to me working. You should come visit.

So we have it working in several communities right now. And granted, there are ways to make it better. But, you know, this one to me is very personal. So my mom and brother live in this town of 250 people, live rural central Virginia. And they have really good broadband because they are. are lucky enough to live about 100 yards away from the local telco exchange. And so they've got, they actually have better broadband than I do. Yeah.

My aunt, who lives three miles away from them, like has, I mean, it's miserable connectivity. I mean, it's like down in the hundreds of, like, Killabod that she has. And so, you know, she's constantly having to either, you know, use the Internet at work or come to my mom's house to do. things on the internet, which is just crazy in 2020. And like, imagine if, I mean, her children are all adult at this point. But like, imagine if she had high school kids. Like, you cannot fully participate in like the modern system of education and get yourself prepared for a full participation in a digital future if you don't have access to the internet as a kid now. Yeah. So I guess my question is how.

So right next to that, you know, we talk to a lot of telecos. We cover a lot of telecom. There's the promise of 5G is going to do everything ever. But it's also going to, you know, a lot of the telecoms have promised 5G will come and cover rural areas. That's actually a very specific promise they've made. How does that interact with white spaces? Look, we don't have a, we don't care how the problem gets solved. It's a good attitude. We have offered up, uh, we, we have offered up, uh, white space as a as a possibility and we're happy to have that used. We're happy to have it superseded by something that's better. We're happy to have it be a complimentary thing.

But I think you have to have real commitment to invest in infrastructure in these places. Like there's nothing miraculous about 5G if you don't choose to go deploy the infrastructure. Yeah. Like nothing at all. And it would be great. There's nothing, nothing I would love more than to have my hometown blanket it with 5G. Like it would, you know, in some ways, if you did that, like, it could spur the same sort of innovation and development as some of these, some of these like downtown areas in rural and

middle America where the local community has decided to invest in municipal fiber. And like those, those investments are, you know, just incredibly beneficial for those communities. Like an equivalent wireless investment in these communities could, like, produce similar sorts of effects in terms of innovation. When you think about, specifically, you know, you're written a book that's basically about how technology is going to change the character of rural America, right? One of the big, and now we're talking about broadband policy, which is, you can just stop me whenever it gets boring because I can do for it. One of the big debates is, hey, actually, there's no business case for blanketing some of these small. There's not enough customers to lay the fiber and get any return from that investment. There's like literally not enough people.

I don't buy that argument. I live in New York City. There's a lot of people. The broadband still sucks. Like that argument to me is like very personal. Do you think there's a different kind of role for our government to play in terms of deploying some of these foundational access AI education opportunities? Yeah, I think so.

One way or the other, like whether it's public or pride. it, and I think it's probably a mix of the two. We have to commit ourselves to having ubiquitous high-quality broadband everywhere. It's a basic public good at this point, given where the future is going. You really can't have, you can't have a straight-faced argument that you're going to have an equitable future unless you have equitable access to internet connectivity. Yeah, look, if the thing that is preventing the markets from filling this role is sort of the cost of deploying it, then like maybe that's a place where the government should go in and try to invest in infrastructure that can change the commercial calculus of doing this stuff. And I think it's entirely possible. Like the white space technology is cheap.

It's not there there's absolutely a business case that you could make. that this or something like it could be part of a solution for getting rural communities better connected. And if we think that it's not good enough, then we know how to solve huge challenging engineering problems. Like, you know, we've got DARPA, we've got the National Science Foundation, we've got all of these institutions that, you know, when we've sort of set these grand challenge problems up in front of people and sort of committed ourselves to investing in the research and the

development to go make them happen, like people will go solve the problems. Like, we just need to make it a priority, I think. So we've been talking on the sort of times ago you've been describing two to five years, right? But I want to get beyond it. When you think about AI in 25 years, and 50 years, what does that look like to you? I think this is really, really sort of tough. I had this exercise with my team recently where I was just trying to predict the next five years. And Arthur C. Clark wrote this book in the 60s called Profiles of the Future. And like this is the place where his three laws got mented. You know, the third law is the one that everybody knows about like any sufficiently advanced technologies indistinguishable for magic.

But the interesting thing about that book is like the point he makes is that scientists and technologists are awful at predicting the future, just awful. But it is very easy, though, without being specific about what you think the set of technologies are going to be. You can sort of well imagine what the trends are that are going to shape the future. And so, you know, like if I look at 25 or 50 years, like I think we're, you know, we hopefully will be using tools like AI to help us deal with some of the big challenges that we'll be facing us. So like there's climate change where I think AI can play a very big role all the way from, you know, materials and material science, which is going to be super important to, you know, optimizing production and consumption of energy. which we should be doing more of already. I think that what I'm seeing already is that we're at the very, very early stages of this,

you know, very beneficial marriage between biology and machine learning. And so, you know, again, even if you're thinking about how you're going to get away from a, like a petrochemical economy, there are all sorts of ingenious things that people are doing right now using a combination of biology. and machine learning to be able to produce the materials that we need to run our lives that are made for petrochemicals right now. But you can actually reprogram things like yeast to go brew these things for us or use protein folding as a way to use the structure of proteins to serve as the substrate for materials that you're generating in a very precise way.

And like those biology techniques only work when they're combined with machine learning. And so like I think there's going to be this incredible flourishing of those sorts of things. I do think, you know, the thing that that people should be thinking about on the 25, 50 year time horizon is we will have in almost all of the industrial world a demographic inversion. So we are below replenishment fertility rates in almost all of the European economies or countries. And also in China, Korea, Japan. And so, like, we will have an increasingly elderly population and, relatively speaking, fewer workers. And so if you want to continue the, you know, the productivity trends that we've had, like, you actually have to have more technology. So the very thing that people are anxious about, like, you know, machines taking jobs, like, we better hope that the machines come along at some point and, like, be able to do, like, a lot more than they're doing right now because otherwise, like, we won't have enough capacity to run the country and take care of, like, the elderly people who are, you know, in various declining states of health.

So I don't know. That's sort of how I see the future. Like, I think AI is going to play a very, very big role in it. I think it is going to be less about, you know, again, Commander Data, you know, Android's walking around who are, you know, like the full, you know, sapient equivalence of, you know, human beings. But I think over time, and especially at that intersection of biology and machine learning, like we're going to see more and more super interesting stuff happening.

That is a pretty wild vision of things, actually. So here's my last question. I ask it to everybody. with a fancy title. We've, uh, it's, this is really just me trying to get,

it's like my self-help program. We've talked about a lot of things. You have a wide range, right? We've talked about medicine. We've talked about white space internet. We've talked about, uh,

yeast protein, uh, yeast doing protein full. Like we've, the whole range. AI. When do you work? I ask, I literally ask this to,

to everybody who comes on who's like a busy person. When do you sit down? It seems like you have to read a lot. You obviously write a lot. When do you do your focus, work. I schedule it on my calendar. I'm a very early riser. So I, this started when I had like really, really young children. My kids are nine and eleven now, but like when they were infants, they would just be up super early every morning. And so I've never gotten out of the habit of like

getting up at five in the morning. So like I get up at five every day. And I can easily spend three hours doing focus work first day in the morning before the rest of my engineering teams come online. This is not the answer I want to hear is come on, man. But look, the thing, one of the things that I've been doing for a really long time is I got into the habit of doing OKRs. So these are objectives and key results when I was at Google. And so this is basically just sort of a high level framework where you just sort of say this quarter for these three months, like, this is the top three things that I want to accomplish. And like, this is how I know that I have accomplished them at the end.

Like, these are the measures of success. And what I do almost every week is I sit down with my calendar like a week at a time and I look at everything that's on it. And I ask myself whether or not the things are on my calendar are helping me with my objectives and key results for the next three months. And if the things aren't, then like I cancel the meetings and I, you know, like I reserve that time for like a thing that does help me make progress. And like a big part of that is I have my EA calls them TTT blocks. So they're time to think.

And like if you look at my calendar, like there are big blocks of time that are TTT. Like on almost every day this week, I've got at least two hours of time every day that's. like mark t t t i don't know if the listener could hear that clicking but i literally just watched him bring up his calendar on the screen yeah i was just i mean like i'm no it's it's no live like at least two hours every day all right kevin thank you so much this is an amazing conversation i know we've gone a little bit over your time thank you so much for joining us we got to have you back soon oh i would i would uh i'd love to love to come back and this was a this is an amazing conversation thank me thank you for having me on the show yeah uh tell people where they can find your book uh

So the book is reprogramming the American Dream. You can buy it. You pre-order now. It's on sale April the 7th at Amazon, any of your other fine online booksellers. And if you can find an open bookstore, which you probably can't. And you probably shouldn't. You probably shouldn't. So don't go to the bookstore to buy this book.

Order it online. Yeah. And hopefully people will find it interesting. Very cool. Well, thank you so much. We'll talk to you soon. Awesome. Bye. All right, my thanks to Microsoft CTO, Kevin Scott. His book, Reprogramming, The American Dream, is out now. You can go check it out. We'll definitely have him back again in the future. We'll be back on Friday with the chat show. We're back on Tuesday.

We're lining up some big names to talk about what is happening with technology and society right now with the pandemic. Let me know who you want me to talk to you. I love that feedback. You can tweet to me on Matt Reckless. We'll talk to you soon.