Kevin Scott: Microsoft CTO

If you include research it becomes even more difficult. So like I think broadly speaking Microsoft's product portfolio includes everything from big cloud business, a big set of SaaS services. We have sort of the original or some of what are among the original productivity software products that everybody uses. We have an operating system business. We have a hardware business where we make everything from computer mice and headphones to high-end personal computers and laptops. We have a fairly broad-ranging research group where we have people doing everything from economics research. There's this really smart young economist Glenn Weyl who my group works with a lot who's doing this research on these things called radical markets. He's written an entire technical book about this whole notion of radical markets. So the research group sort of spans from that to human-computer interaction to artificial intelligence. And we have GitHub, we have LinkedIn, we have a search advertising and news business, and probably a bunch of stuff that I'm embarrassingly not recounting in this list.

Yeah, gaming for sure. I was having a super fun conversation this morning with Phil Spencer. When I was in college, there was this game that LucasArts made called Day of the Tentacle that my friends and I played forever. And we're doing some interesting collaboration now with the folks who made Day of the Tentacle and I was completely nerding out with Tim Schafer, like the guy who wrote Day of the Tentacle this morning. Just a complete fanboy. It happens a lot. Microsoft has been doing so much stuff at such breadth for such a long period of time that most of the time my job is very serious and sometimes I get caught up in how amazing it is to be able to have the conversations that I have with the people I get to have them with.

You know, and I think it's a reasonable set of questions to be asking. So what Glenn and — one mode of thought there, if you have doubts that the markets are actually working you can sort of tip towards like okay let's become more socialist and have central planning and governments or some other central organization is making a bunch of decisions about how work gets done and where the investments and where the outputs of those investments get distributed. Glenn's notion is like lean more into the market-based mechanism. So for instance, this is one of the more radical ideas, like suppose that you had a radical pricing mechanism for assets like real estate where you could be bid out of your position in your home. If somebody came along and said I can find higher economic utility for this piece of real estate that you're running your business in, then you either have to bid to sort of stay or the thing that's got the higher economic utility sort of takes over the asset. Which would make it very difficult to have the same sort of rent-seeking behaviors that you've got right now because if you did speculative bidding you very quickly lose a whole lot of money. And so the prices of the assets would be sort of very closely indexed to the value that they could produce. And because you'd have this sort of real-time mechanism that would force you to sort of mark the value of the asset to the market, then it could be taxed appropriately. You couldn't sort of sit on this thing and say, oh, this house is only worth ten thousand bucks when everything around it is worth ten million.

Yeah. And Glenn does a much, much better job than I do at selling it. I probably picked the world's worst example, but it's intentionally provocative. This whole notion, I'm not sure whether I like this notion that we could have a set of market mechanisms where I could get bid out of my property. But you know, if you're thinking about something like Elizabeth Warren's wealth tax for instance, you would have — I mean it'd be really interesting how you would actually set the price on the assets and you might have to have a mechanism like that if you put a tax like that in place.

I think it's sort of one of the most important questions in technology, maybe even in society right now, about how is AI going to develop over the course of the next several decades and what's it going to be used for and what benefits will it produce and what negative impacts will it produce and who gets to steer this whole thing. At the highest level, one of the real joys of getting to do what I do at Microsoft is Microsoft has this heritage as a platform company. And so Bill Gates has this thing that he said a bunch of years ago where the measure of a successful platform is that it produces far more economic value for the people who build on top of the platform than is created for the platform owner or builder. And I think we have to think about AI that way. It has to be a platform that other people can use to build businesses, to fulfill their creative objectives, to be entrepreneurs, to solve problems that they have in their work and in their lives. It can't be a thing where there are a handful of companies sitting in a very small handful of cities geographically who are making all the decisions about what goes into the AI and then on top of all this infrastructure build all of the commercially valuable uses for it. That's bad from an economics and sort of equitable distribution of value perspective, sort of back to this whole notion of do the markets work. But I think it's also bad from an innovation perspective because I have infinite amounts of faith in human beings that if you give folks powerful tools they will go do interesting things. And it's more than just a few tens of thousands of people with the interesting tools, it should be millions of people with the tools. So it's sort of like you think about the steam engine in the late 18th century. It was maybe the first large-scale substitute for human labor that we've built — a machine. And in the beginning when these things are getting deployed, the folks who got most of the value from the steam engines were the folks who had capital. They could afford to build them and they built factories around them and businesses and the experts who knew how to build and maintain them. But access to that technology democratized over time. Now an engine is not a differentiated thing. There isn't one engine company that builds all the engines and all of the things that use engines are made by this company and they get all the economics from all of that. Fully democratized — they're probably sitting here in this room and even though they don't, there are probably things like the MEMS gyroscope in both of our phones, little engines sort of everywhere.

Yeah.

And the tieback to this radical markets thing — my team has been working with Glenn on this, and Jaron Lanier actually. Jaron is like the sort of father of virtual reality, he's one of the most interesting human beings on the planet, a sweet, sweet guy. And so Jaron and Glenn and folks in my team have been working on this notion of data is labor, or they call it data dignity as well. The idea is that if you — again going back to this industrial analogy — if you think about data as the raw material that is consumed by the machine of AI in order to do useful things, then we're not doing a really great job right now in having transparent marketplaces for valuing those data contributions. We all make them explicitly — you go to LinkedIn, you sort of set up your profile, that's an explicit contribution, you know exactly the information that you're putting into the system and you put it there because you have some nominal notion of what value you're going to get in return. But it's only nominal — you don't know exactly what value you're getting in return, the service is free, low amount of perceived value. And then you've got all this indirect contribution that you're making just by virtue of interacting with all of the technology that's in your daily life. So what Glenn and Jaron and this data dignity team are trying to do is can we figure out a set of mechanisms that let us value those data contributions so that you could create an economy and a set of controls and incentives that would allow people to — maybe even in the limit — earn part of their living through the data that they're creating. And you can sort of see it in explicit ways. There are these companies like Scale AI and there are a whole bunch of them in China right now that are basically data labeling companies. So you're doing supervised machine learning, you need lots and lots of labeled training data and those people who work for those companies are getting compensated for their data contributions into the system.

Correct.

I don't know. It's hard. It really is. We don't have as much transparency as I think we need in how the data is getting used. And it's super complicated — as technologists we sort of appreciate some of the subtlety there. The data gets created and then it's not valuable atomically. The data exhaust that you give off or the explicit data that I'm putting into the system isn't super valuable atomically. It's only valuable when you sort of aggregate it together into large numbers. It's true even for these folks who are getting compensated for labeling things for supervised machine learning — you need lots of labels to train a model that performs well. So that's one of the challenges — how do you figure out, because this data is getting combined in so many ways, through these combinations, how the value is flowing.

The good thing is it started off as a platform company and it's still a platform company. The parts of the business that are thriving and most successful are those that are building platforms. The mission of the company has changed in a very interesting way. Back in '89, '90, they were still on the original mission, which was put a PC on every desk and in every home. It was basically about democratizing access to this new personal computing technology, which when Bill started the company, integrated circuit microprocessors were a brand new thing and people were building homebrew computers from kits like the way people build ham radios right now. And I think this is the interesting thing for folks who build platforms in general. Bill saw the opportunity there and what personal computers could do and it was a reach — you just sort of imagine where things were when they started the company versus where things are now. In success when you've democratized a platform, it just sort of vanishes into the platform. You don't pay attention to it anymore. Operating systems aren't a thing anymore. They're super important, completely critical, and when you see one fail you just sort of understand, but you're not waiting for the next operating system thing in the same way that you were in 1995, right? In 1995 we had Rolling Stones on the stage with the Windows 95 rollout, it was like the biggest thing in the world, everybody lined up for it the way that people used to line up for iPhone. But eventually — and this isn't necessarily a bad thing — the success is that it becomes ubiquitous. It's everywhere. And human beings, when their technology becomes ubiquitous, they just sort of start taking it for granted. So the mission now that Satya rearticulated five-plus years ago when he took over as CEO, our mission is to empower every individual and every organization in the world to be more successful. And so again, that's a platform mission. The way that we do it now is different. We have a hyperscale cloud that people are building their applications on top of. We have a bunch of AI infrastructure that people are building their AI applications on top of. We have a productivity suite of software like Microsoft Dynamics, which some people might not think is the sexiest thing in the world but it's helping people figure out how to automate all of their business processes and workflows and to help those businesses using it to grow and be more successful. It's a much broader vision in a way now than it was back then. Back then it was a very particular thing. Now we live in this world where technology is so powerful and it's such a basic fact of life that it exists and is going to get better and better over time, or at least more and more powerful over time. So what you have to do as a platform player is just much bigger.

That's probably early days.

Oh for sure.

Oh sure. When we look at what the platforms of the future could be, it's fairly obvious that AI is one. But we also think of mixed reality and quantum as these two interesting, potentially —

Okay. So let's get crazy then.

And it's having impact right now. One of the more interesting things that's happened with mixed reality over the past couple of years that I didn't clearly see is that it's become the computing device for folks doing their work who haven't used any computing device at all to do their work before. So technicians and service folks and people who are doing machine maintenance on factory floors. Because they're mobile and they're out in the world and they're working with their hands and sort of servicing these very complicated things, they don't use their mobile phone and they don't carry a laptop with them and they're not tethered to a desk. And so mixed reality — where it's getting traction right now, where HoloLens is selling a lot of units — is for these sorts of applications, for these workers. And it's become — the people love it. They're like oh my god, for them it's like the same sort of productivity boost that an office worker had when they got their first personal computer.

Yeah, I mean I think it's super, it's everywhere. We run these review meetings now where it's me and Satya and members of Satya's leadership team and a cross-functional group of folks across the entire company who are working on either AI infrastructure or have some substantial part of their product work using AI in some significant way. Now the important thing to understand is when you think about how the AI is going to manifest in an experience for something that's going to make it better, I think you don't want the AI-ness to be the first order thing. Whatever the product is and the thing that it's trying to help you do, the AI just sort of makes it better. And this is a gross exaggeration, but people get super excited about where the AI is showing up in products. And I'm like, do you get that excited about where you're using a hash table in your code? It's just another tool. It's a very interesting programming tool, but it's an engineering tool. And so it shows up everywhere. We've got dozens and dozens of features now in Office that are powered by fairly sophisticated machine learning. Our search engine wouldn't work at all if you took the machine learning out of it. Increasingly things like content moderation on our Xbox and xCloud platform.

No, no, no. It's like anti-bullying stuff.

Yeah. Correct. But it's really targeted towards a gaming audience. So it's a very particular type of thing where the line between playful banter and like legitimate bullying is like a subtle one.

Yep.

Yep.

The very different kinds of communication going on on the two platforms.

I think in a certain way if you have what I would call vertical social networks it gets to be a little bit easier. So if you have a clear notion of what your social network should be used for or what you are designing a community around, then you don't have as many dimensions to your sort of content safety problem as you do in a general-purpose platform. So on LinkedIn, the whole social network is about connecting people with opportunity, whether it's helping them find a job or to sort of find mentors or to help them find their next sales lead or to just sort of allow them to broadcast their professional identity to their network of peers and collaborators and professional community. That is in some ways very broad but in other ways it's narrow. And so you can build AI machine learning systems that with those boundaries are capable of making better automated decisions about what is an inappropriate or offensive comment or dangerous comment or illegal content, when you have some constraints around it.

Same thing with gaming social networks, for instance. It's about playing games, about having fun. And the thing that you don't want to have happen on the platform is why bullying is such an important thing. Bullying is not fun, so you want to do everything in your power to encourage that not to happen.

I think it's a tough problem in general. It's one where I think eventually we're going to have to have some sort of clarification from our policy makers about what it is that we should be doing, like where the lines are, because it's tough. Like in democracy, right? You want some sort of democratic involvement. People should have a say in where the lines are drawn. You don't want a bunch of people making unilateral decisions. And we are in a state right now for some of these platforms where you actually do have to make unilateral decisions where the policymaking isn't going to happen fast enough in order to prevent very bad things from happening. But we need the policymaking side of that to catch up, I think, as quickly as possible, because you want that whole process to be a democratic thing, not some sort of weird thing where you've got a non-representative group of people making decisions that have national and global impact.

Yeah. I mean, with fake news, for instance, and deep fakes and fake news generated by humans, we can talk about that. I think that is another very interesting level of complexity. But if you think about just the written word, right, we invented papyrus what, 3,000 years ago, where you could sort of put word on paper. And then 500 years ago we get the printing press, where the word gets a little bit more ubiquitous. And then you really didn't get ubiquitous printed word until the end of the 19th century when the offset press was invented, and then it just sort of explodes. And the cross product of that and the Industrial Revolution's need for educated citizens resulted in this rapid expansion of literacy and the rapid expansion of the word.

We had 3,000 years up to that point to figure out, like, what's journalism, what's editorial integrity, what's scientific peer review. And so you built all of this mechanism to try to filter through all of the noise that the technology made possible to get to something that society could cope with. And if you think about just the PC, it didn't exist 50 years ago. And so in this span of half a century, we've gone from no ubiquitous digital technology to having a device that sits in your pocket where you can sort of say whatever is on your mind.

It's not surprising that we haven't figured out what to do yet. But we've got to really lean into this set of problems because we basically have three millennia worth of work to do about how to deal with all of this, and probably what amounts to the next decade worth of time.

Yeah, I think we've articulated a clear point of view. Brad Smith wrote a blog post last fall, I believe, that sort of outlined very specifically what our point of view is there. We believe that there are certain uses to which face recognition should not be put, and we believe again that there's a need for regulation there. The government should really come in and say where the lines are. And we very much want figuring out where the lines are to be a democratic process. But in the short term, we've drawn some lines where we push back against uses of face recognition technology. The city of San Francisco, for instance, I think has completely outlawed any government agency from using face recognition tech. And that may prove to be a little bit overly broad. But for certain law enforcement things, I would personally rather be overly cautious in terms of restricting use of it until we have defined a reasonable, democratically determined regulatory framework for where we could and should use it.

The other thing there is we've got a bunch of research that we're doing and a bunch of progress that we've made on bias. There are all sorts of weird biases that these models can have, all the way from the most noteworthy one where you may have underrepresented minorities who are underrepresented in the training data and then you start learning strange things. But there are even other weird things — I think we've seen in the public research where models can learn strange things, like all doctors are men, for instance.

It really is a thing where it's very important for everybody who is working on these things before they push, publish, launch the experiment, push the code online, or even publish the paper that they are at least starting to think about what some of the potential negative consequences are of some of this stuff.

This is where the deep fake stuff I find very worrisome, just because there are going to be some very good beneficial uses of GAN-generated imagery. And funny enough, one of the places where it's actually useful is we're using the technology right now to generate synthetic visual data for training some of the face recognition models to get rid of the bias. So that's one super good use of the tech.

But it's getting good enough now where it's going to challenge a normal human being's ability to — it's very expensive for someone to fabricate a photorealistic fake video, and GANs are going to make it fantastically cheap to fabricate a photorealistic fake video. And so what you assume you can sort of trust versus be skeptical about is about to change. And we're not ready for it, I don't think.

Yeah. And it could be, you know, I think we will have technology that may be able to detect whether or not something's fake or real. Although the fakes are pretty convincing even when you subject them to machine scrutiny. But we also have these increasingly interesting social networks that are under fire right now for some of the bad things that they do. One of the things you could choose to do with a social network is you could use crypto and the networks to have content signed, where you could have a full chain of custody that accompanied every piece of content. And so when you're viewing something and you want to ask yourself how much can I trust this, you can click something and have a verified chain of custody that shows, oh, this is coming from this source and it's signed by someone whose identity I trust.

I think having that chain of custody, being able to say, here's this video, it may or may not have been produced using some of this deep fake technology, but if you've got a verified chain of custody where you can trace it all the way back to an identity and you can decide whether or not I trust this identity — like, oh no, this is really from the White House, or this is really from the office of this particular presidential candidate, or it's really from Jeff Weiner, CEO of LinkedIn, or Satya Nadella, CEO of Microsoft — that might be one way that you can solve some of the problems. And that's not the super high-tech, like we've had all of this technology forever, right?

But I think you're right, it has to be some sort of technological thing because the underlying tech that is used to create this is not going to do anything but get better over time, and the genie is sort of out of the bottle. There's no stuffing it back in.

Well, look, I think it's one of the reasons why the scientific method and our apparatus of modern science is so good, because you don't have to trust anything. The whole notion of modern science — beyond the fact that this is a hypothesis and this is an experiment to test the hypothesis, and this is a peer-review process for scrutinizing published results — but stuff's also supposed to be reproducible. So it's been vetted by this process, but you also are expected to publish enough detail where, if you are sufficiently skeptical of the thing, you can go try to reproduce it yourself. And I don't know what it is, I think a lot of engineers are like this, where your brain is sort of wired for skepticism. You don't just first-order trust everything that you see and encounter. You're sort of curious to understand the next thing. But I think it's an entirely healthy thing. And we need a little bit more of that right now.

I think there's a bunch of exciting — I mean, on the Office front, there's going to be this increasing productivity wins that are coming out of some of these AI-powered features that are coming. The products will sort of get smarter and smarter in a very subtle way. There's not going to be this big bang moment where Clippy is going to reemerge.

I think there are a bunch of very small ways in which machine learning-powered assistive things are in the product right now. There are a bunch of interesting things — the autoresponse stuff's getting better and better. It's getting to the point where it can autorespond with, okay, this person's clearly trying to schedule a meeting, so it looks at your calendar and it automatically tries to find a time and a space that's mutually interesting. We have this notion of Microsoft Search, where it's not just web search but it's search across all of your information that's sitting inside of your Office 365 tenant and potentially in other products. We have this thing called the Microsoft Graph that is basically an API federator that gets you hooked up across the entire breadth of what were information silos before they got woven together with the Graph. That is getting, with increasing effectiveness, plumbed into some of these autoresponse things where you're going to be able to see the system automatically retrieve information for you. If you know — I frequently send out emails to folks where I can't find a paper or a document or whatnot. There's no reason why the system won't be able to do that for you.

I think it's building towards having things that look more like a fully integrated assistant. But you'll have a bunch of steps that you will see before — it will not be this big bang thing where Clippy comes back and you've got this manifestation of a fully powered assistant. So that's definitely coming. All the collaboration, co-authoring stuff is getting better. It's really interesting — if you look at how we use the Office product portfolio at Microsoft, more and more of it is happening inside of Teams as a canvas. It's this thing where collaboration is at the center of the product. And we built some really cool stuff, some of which is about to be open source, that are sort of framework-level things for doing co-authoring.

We're already a little bit better than that. And so the fact that you're unaware of it means we've got a better job to do helping you discover this stuff. But yeah, it's already got a huge cloud component. And part of this framework stuff — I think we're calling it, we've been working on it for a couple years, so I know the internal code name for it, but I think when we launched it at Build, it's called the Fluid Framework. What Fluid lets you do is you can go into a conversation that you're having in Teams and reference part of a spreadsheet that you're working on, where somebody's sitting in the Excel canvas working on the spreadsheet with a chart or whatnot, and you can embed part of the spreadsheet in the Teams conversation where you can dynamically update and all of the changes that you're making to this object are coordinated and everything is sort of updating in real time. So you can be in whatever canvas is most convenient for you to get your work done.

I think the thing that you learn as you manage larger and larger scale is that there are three things that are very, very important for big engineering teams.

One is having some sort of forethought about what it is that you're going to be building over large periods of time. Not exactly — you don't need to know that I'm putting all my chips on this one product and this is going to be the thing. But it's useful to know what sort of capabilities you think you're going to need to have to build the products of the future, and then invest in that infrastructure. And I'm not just talking about storage systems or cloud APIs — it's also what does your development process look like, what tools do you want, what culture do you want to build around how you're collaborating together to make complicated technical things. Having an opinion and investing in that just gets more and more important, and the sooner you can get a concrete set of opinions, the better you're going to be. You can wing it for a while at small scales. When you start a company, you don't have to be super specific about it. But the biggest miseries that I've ever seen as an engineering leader are in places where you didn't have a clear enough opinion about those things soon enough, and then you just sort of create a bunch of technical debt and culture debt that is excruciatingly painful to clean up.

The other bundle of things is it's really, really important to have a clear mission that's not just some cute crap you say because you think you should have a mission, but something that clarifies for people where it is that you're headed together.

I know it's probably a little bit too popular right now, but Yuval Harari's book Sapiens — one of the central ideas in his book is that storytelling is the quintessential thing for coordinating the activities of large groups of people once you get past Dunbar's number. And I've really, really seen that managing engineering teams. You can brute-force things when you're less than 120, 150 folks, where you can sort of know and trust and understand what the dynamics are between all the people. But past that, things just start to catastrophically fail if you don't have some sort of set of shared goals that you're marching towards. Even though it sounds touchy-feely and a bunch of technical people will sort of bulk at the idea that you need to have a clear mission — it's very, very, very important.

And that works for companies too. It works for everything. Even down to, you sort of really think about it — our currency, for instance, is a story. Our Constitution is a story. Our laws are — we believe very, very, very strongly in them, and thank God we do. But they're just abstract things. They're just words. If we don't believe in them, they're nothing.

Yeah, look, I think we're entering this time where we have technology that is progressing at the fastest rate that it ever has. And you've got some really big social problems, society-scale problems that we have to tackle. And so I think we're going to rise to the challenge and figure out how to intersect all of the power of this technology with all of the big challenges that are facing us.

Whether it's global warming, whether it's — the biggest remainder of the population boom is in Africa for the next 50 years or so. And global warming is going to make it increasingly difficult to feed the global population, in particular in this place where you're going to have the biggest population boom. I think AI is going to, if we push it in the right direction, it can do incredible things to empower all of us to achieve our full potential and to live better lives. But that also means focus on some super important things, like how can you apply it to healthcare to make sure that our quality and cost of and ubiquity of health coverage is better and better over time. That's more and more important every day.

In the United States and the rest of the industrialized world — Western Europe, China, Japan, and Korea — you've got this population bubble of aging working-age folks who, at some point over the next 20, 30 years, they're going to be largely retired. And you're going to have more retired people than working-age people. And then you've got natural questions about who's going to take care of all the old folks and who's going to do all the work. And the answers to all of these sorts of questions, where you're sort of running into constraints of the world and of society, has always been, what tech is going to help us get around this.

When I was a kid in the 70s and 80s, we talked all the time about population boom — we're not going to be able to feed the planet. And we were right in the middle of the Green Revolution, this massive technology-driven increase in crop productivity worldwide. Some of that was taking some of the things that we knew in the West and getting them distributed to the developing world, and part of it were things like smarter biology, helping us increase. And we don't talk about overpopulation anymore because we can more or less figured out how to feed the world. That's a technology story.

I'm super, super hopeful about the future and the ways where we will be able to apply technology to solve some of these super challenging problems. One of the things I'm trying to spend my time doing right now is trying to get everybody else to be hopeful as well. Because back to Harari — we are the stories that we tell. If we get overly pessimistic right now about the potential future of technology, we may fail to get all the things in place that we need to have our best possible future.

Thank you so much for having me. It was a pleasure.