Empowering people with AI (with Kevin Scott)

Thank you for having me and for the very kind and generous introduction.

The biggest influence on my beliefs and – maybe the better way to say it is just sort of my ethos and values – were my family. We certainly weren't well off. We had, at times, a lot of financial difficulty and at best my family was always paycheck to paycheck. But I don't think we really paid much attention to that. Or at least that's not the thing that we were always talking about. My dad was a construction worker, and my mom hustled with a bunch of little side jobs that she did. She was a tax preparer. She ran a nursery school for the kids at church for a brief period. She did a whole bunch of work, but her primary job was taking care of the family. I just had the luxury of being around a bunch of people who were hardworking, very curious, who always worked with their hands, who were always tinkering with something and who felt like they had an obligation to their community to do something valuable, that our job was to take care of each other. People who cared for their community and who were serving others in their day to day lives as well, all the time.

I've thought a lot about this idea of the American dream over the past handful of years. And I think one of the things that a lot of people may be getting confused about is confusing the idea of a dream and a promise. The thing that both my parents and I believed when I was growing up – and I still believe this and my family back in Virginia still believes this – is that there is this motivating idea that if you go figure out some way to be valuable to your fellow human beings, you try to understand what they want and need, and then you go do work and work as hard as you possibly can to try to meet those needs of your fellow human beings, that good things will happen to you. I tell this to my kids all the time… CTO of Microsoft or Head of Global Business for Microsoft or whatever these big titles are like, there's so much luck involved in that. That's a crazy thing to try to fixate on and plan for. But I would have been okay no matter what my outcome was by just applying this set of ideas that I can work hard, try to discover what's valuable to do, and then going to do that thing, and it really isn't a promise. And it was kind of implicit in the way that my parents talked to me about this and in how they acted. It's like, you just have to go do the work and you’ve got to go grind through it. And, you know, sometimes it works and sometimes it doesn't. And it's maybe even more valuable to do something and fail and learn something from the failure and then get up and go again. Sometimes success teaches you nothing.

Pretty much. I didn't even know the word entrepreneur when I was growing up. But I think that’s what I was surrounded by, a bunch of entrepreneurs.

Part of it is pragmatic. I was constantly trying to figure out how I was going to support myself and support my family, because I got a scholarship that partially covered the cost of going to school, but didn't completely, so I had student loans. I was on my own financially, and part of that was my decision and part of it was my parents’ financial circumstances. I just had to take care of myself and eventually I had to take care of my entire family. And so part of it was a pragmatic thing. I enjoyed both equally and I think that's a tremendous luxury because sometimes it's hard to even find the one thing that you really enjoy. And so I picked computer science. And the story that I told myself about picking computer science is that there's a way to do computer science that is artful, that is grounded in the humanities, and that I wasn't really forsaking anything by making this choice. And, truth be told, I'm interested in a lot of things. I drive my wife nuts sometimes with the amount of stuff that I'm interested in. So I think it's fine having a vocation that you go and focus on and try to make yourself as good at it as you possibly can and also like it's great for you, the person, and maybe even for the vocation to have these other things that you're interested in as well.

It sort of depends on what stage of your career and your life that you're in. I don't know whether I personally would have done anything different because I like to understand things as deeply as humanly possible. I've always been fascinated with tools. Young Kevin would have looked at these AI systems and would have wanted to understand what's behind the scenes of them and how they worked. But I think the advice that I would give people more generally is like, find whatever it is that you really, really, really have a passion about, that is valuable to other people and try to use these tools that you now have available to you to go do that thing. I have a 13-year-old and a 15-year-old who are like thinking about what it is that they want to be. My 15-year-old daughter has wanted to be a cardiothoracic surgeon for the past year. It’s sort of a weird thing that she got fixated on when she was really young, because she watched too much Grey's Anatomy and she had a couple of doctor parents of her friends in school. She had decided that she's not a tech person, but her mind has been changed over the past few years. The way that she's using tools like ChatGPT and GitHub Copilot to help her solve these medical problems and healthcare problems that she's really motivated by is quite interesting. I think that that's the mindset. It's like all of these problems we need to go solve, all of these things that matter and that are important, and we have increasingly powerful tools to go do them. You don't have to be a programmer anymore to get the computer to do unbelievably powerful things for you in service of what it is you're trying to accomplish.

I was reflecting on this actually this morning… I'm learning how to do ceramics right now. And the way that I learn is like, I can go watch a bunch of videos and read a bunch of books and I can absorb it all, but when it comes to the practice of the thing that I'm trying to learn, I have to actually go do it because I don't know how to prioritize all of this information I've stuffed in my head and I think the way that I learn is like, other people do it, but it is idiosyncratic there are other ways that people learn that are more effective for them than how I learn. And so this idea that you could have an infinitely patient tutor that could adapt to your particular learning needs very individually, I think it’s just a really interesting, powerful idea.

Maybe I could just pull back for a second and talk about the history of AI. We’ve had an extraordinary, difficult time across the entire history of the discipline of AI. So when Turing wrote that paper, the term artificial intelligence hadn't been coined yet. It was five years later at a workshop at Dartmouth, when bunch of these mathematicians and computer scientists and information theorists sat down and said we want to build a machine that in many ways can replicate what a human brain does and we’re going to call this study, this new field, artificial intelligence. And we have had a really hard time over the entire history of the discipline even defining what artificial intelligence is. We’ve got this track record of we think that we understand what an intelligence is like, what our own intelligence is, and then we go build machines to replicate some aspect of it. And as soon as we accomplish the thing, then we redefine what the measure of intelligence is. We used to think that the apotheosis of human cognition are these very challenging games like chess and go. In 1997, IBM built a system called Deep Blue that beat Garry Kasparov, who was the world champion chess player at the time. And it was stunning, nobody thought it was possible. As soon as it happened, we figured out that this really isn't the breakthrough that is going to tip us over into this idea of artificial general intelligence like this solves a very narrow thing. Funnily enough, I think chess is now a more popular human pastime and hobby. You could even think about it as a sport, given how competitive it is and how much joy people derive from competing and watching the competition, even though since 1997 there has never been a human chess player as good as a computer at playing chess. We just don't care anymore. I think that's a really important thing to keep in mind. And to answer your exact question, the most stunning breakthrough in my career I think probably happened recently. It was when GPT-4 finished training, and this idea that we had that you could follow a particular path of developing artificial intelligence was going to result in an AI system that was very generally useful for doing a bunch of cognitive tasks. That was a big a-ha moment. I had a strong sense that it was coming, but I didn't know it was coming this fast. And the thing doesn't think. Going back to your Turing quote, he said two distinct things. One is he's talking about thinking and one he's talking about, can we write a piece of software that emulates aspects of what our brains do? Absolutely, we can do the latter. But the former is like almost a philosophical thing. When I say, JP, what are you thinking about? How do you think? That’s a very human question. It’s very different from what these software systems are doing.

Well, I will tell you, I was sceptical going into the meeting. And like my scepticism in general is just a personality disorder, like not anything specific to the team. I'm always slightly sceptical of things that I haven't fully understood. And sometimes, like, I will go fully understand them and I'm still sceptical afterwards. But this was one of those rare instances where I went in and… I'd been building machine learning systems at that point since 2004, like very large-scale things, doing very valuable commercial applications of machine learning, and I was not prepared to go into that meeting to have my mind set about what was possible in the next handful of years. And I walked out of it, like, absolutely convinced that we were on the cusp of something very interesting. We were still years away, but they had a framework and a way of thinking about the problem that they were tackling that was very scientific. There was a methodology like, here's the experiment we're going to run. This is what we will learn from it and this is what it will tell us about the experiment we will go run after that. I thought we had the foundations for a partnership where we could do some valuable things for them to support this very disciplined development of a very useful AI system. And the way that it was going to be useful and interesting for us was that it wasn't trying to solve a narrow problem. A lot of the AI systems before then were built to solve a narrow problem, like play chess, play go, predict which ads someone is going to click on. But this was a, we think we're going to be able to build this system and that as a function of scale. It's going to become more broadly useful and you're going to be able to build like dozens and hundreds and thousands and millions of applications on top of it. As a platform company, that's exactly the mission Microsoft has been on for almost five decades now.

I think the thing that you should do, and this is a set of things that we ought to be thinking about doing broadly actually, not just in the rural parts of the United States, but in the developing world and in all parts of the industrialized world as well. I think job one is you actually have to have the infrastructure in place to support people using AI tools to solve the problems that they want to solve. So part of it is like, do you have a platform available? Is it open? Is it getting cheaper and more powerful over time? Can you operate it freely across global boundaries? It means some very basic things, like one of the struggles that folks in my community have is the internet just doesn't work well. My mom and brother live near a local telephone building, like the exchange for Gladys and they have very good internet. And my uncle, who lives three miles away, has 300 kilobit per second internet, which was great in 1990, and is pretty horrendous now. So you can't leverage AI if you can't connect to AI. I think a bunch of the stuff you need to go do is like really educate people to be entrepreneurs in a sense. So how do you take young people and expose them to this full palette of tools that they have to use, and like help them think critically about what the interesting problems there are in the world that they could put those tools to use solving. It’s a very different educational paradigm than the one that we've had since the beginning of the Industrial Revolution, where it was like, you human being need to learn these basic skills. You need to be literate. You need to know a little bit of math. You need to have structure in your life where you can get up in the morning and go do something for this number of hours and work in teams and understand hierarchies and like all of this stuff. That was very industrial revolution sort of learning. And that's not really what we need right now.

Not to interrupt, but part of my job is – and like I’d do this even if it weren't my job – is thinking about long time horizons. And so it's really hard for us to think past the month, the quarter, the year, the next election cycle, whatever it is. We’re very short time-horizon focused. But if you look over decade time-horizons, we've got some very challenging things happening in the world. We have a warming climate which is going to impose a whole bunch of structural changes on the world. We basically have designed a world for one temperature regime, and we're about to enter another one. We have lots of work to go do to adapt to that and to engineer an energy economy that won't make the temperature regime worse than it's going to be. Maybe the one that we don't talk a lot about is demographic change. Almost everywhere in the industrialized world, we are either in population decline or the population growth is decelerating and you can sort of see a point near in the future where you're going to tip into deceleration. I think the United States and France are demographically two of the countries that are in the best situation. The last data I looked at, France was decelerating, but still growing through the 2030s. It’s actually getting almost to decline. Last year we had this pretty bad signal in France. Still ahead of most European countries, but just like in the US, I think we are almost the last one standing. Italy is in decline. Germany is in decline. China's in decline. Japan and Italy in particular are in pretty serious decline. In our lifetime – this isn’t like an imaginary time-horizon where I'm worrying on the behalf of my children, which I do – but like we can also worry on behalf of ourselves. These demographic changes in the world will mean that you just don't have a big enough workforce to go do the work of the world, unless you have major breakthroughs in productivity and major breakthroughs in productivity, meaning technology. Whether it's AI or something else, something has to happen or we will have a profoundly different world than the one that we occupy right now. AI is currently the best shot on goal for productivity that we have.

They totally get to vote. I agree with Bill on a great many things like on this particular point, I agree. The problem I think that you have in democratic society is if you make the choices very brittle that people have to make, then the votes can be really stark. The way that I talk about this and I wrote about it in the book and I talk about it all the time and it's one of the things that I love about Satya so much is like, I think all the time about permission. We get to do what we do because society has given us permission. We don't have a right to do what we do. We currently have permission to do what we do and like the permission can be revoked if we do it irresponsibly, if we are not listening very carefully to what people are telling us. If you have that mindset you basically have more fluid voting because before you have to have extreme votes, people are voting every day by telling you what they do and don't like, and if you're listening and adapting what you're doing, hopefully you get into an equilibrium that people are happy with, that they feel confident about and hopeful about. But you can't just jam stuff into society. It really doesn't in the limit work that way.

Responsible AI is sort of like responsible governance or generally accepted accounting practices. It is about saying this is what we believe responsible AI means. It is being transparent like in the saying, like publish what your guidelines are, and having a set of processes and controls inside of your company that make sure that you adhere to the standard, and then being willing to engage with stakeholders about evolving the standard and like having some degree of auditability of your processes for adhering to it. I think every company that is developing and deploying this technology needs a framework like this. The same way that you’ve got board audit committees and outside auditors that are looking at your books. When the thing gets important enough, you need some kind of framework, so that collectively, everybody can believe that you're doing the thing. Some of the stuff that I think the president called for is perfectly reasonable. One of the things that we've been very, very hesitant about and we've had technology that can do interesting things with voices for a while. And for a while, we have chosen not to deploy that because the risk of people using that in fraudulent ways, in ways that far outweigh the benefits, seems like something we need to think carefully about. And maybe at some point you will do it because, there are also positive uses for the technology. Like one of the things that we've been doing in my team is there’s this neurodegenerative disease called ALS that will ultimately cause, among other things, people to literally lose their voice. And so we have been using this technology to archive people's voices. So before they lose the ability to speak, we get enough of a set of samples from their actual speech so that you could have an AI system give them their voice back. That is an unbelievably powerful and beneficial thing. And so, yeah, with all of this stuff, it's about the balances. It's like, what good does it enable? Like what bad does it enable? How do you make sure that the good far outweighs the bad and that like when the bad is really bad that you can prevent that from happening and very quickly detect and mitigate misuse.

Yeah. I mean, obviously we're a platform company. And so wherever possible, when we're building tools and infrastructure to help ourselves build products and to launch them and operate them safely and efficiently and all the other attributes you want of products, we try to make that infrastructure available to customers. We’re on the second version of our responsible AI standard that we have published. The new US NIST AI framework that they have published is based in part on our responsible AI framework. There are a bunch of companies and partners who are using our AI framework as a jumping off point for their own. We have an increasingly sophisticated set of infrastructure that we're offering in Azure to help people build and operate AI powered products responsibly. It’s a brand new toolchain. It's a way of testing things. It's a way of monitoring things. And increasingly, the tools that you use for our RAI are powered by AI themselves, which is like one of the really interesting things. The more powerful AI gets, the more powerful your RAI infrastructure gets. One thing that I will say, like, since we're talking about policy and regulation, is that there's also… The same way that when you're developing a technology, there's this balance between benefits and harms that you're trying to weigh, there’s also positive and negative for regulation. If you regulate too aggressively and too early, you may curtail the development of technology that could be unbelievably beneficial. The thing that I… I think, again, this is all about societal equilibrium here, like we have to do what we're doing in a way where we have enough societal trust, where we can develop the technology and let people lay their hands on it and evaluate whether it's good or bad themselves, so that we can inform the regulation that actually has to happen. If you don't do that, you're kind of regulating into a vacuum, you're sort of imagining what a future might be, and you don't even know that it exists. You sort of risk two things, like regulating the wrong things. You think very thoughtfully about regulation, you pass something and it doesn't produce the effect you intended. Or you inhibit something that could be very, very valuable.

Yeah. People, people, people is the right thing. That’s the most important thing. It´s more important than the regulation. It's more important than the implementation details. If you are not doing something that is valuable, that is serving the public interest, that is making everyone's lives net better, then what are you doing? You’re honestly wasting your time. You don't need regulation. You need to wake up and stop doing what you're doing.

Well, I think technology in a sense – I mean, there are many people who will disagree with me – but I do think that technology is almost neutral. Obviously when a technology emerges, it absolutely changes us. It changes us in pretty deep ways. If you just imagine, you and I for instance, how we process the world and see the world is deeply influenced by a whole bunch of technologies that have emerged over the past 40 years. My great grandmother, who passed away many years ago, lived to be a hundred years old. She was born in the 19th century and lived through the first stages of the first internet bubble. She went from a world that had no electricity, no cars, no airplanes… She didn't have indoor plumbing. She was in a barely industrialized, kind of Victorian world when she was born, and by the end, we had airplanes, space travel, satellites, internet, mobile phones. So technology has an extraordinary impact on not just how we live our lives, but how we perceive reality. Whether that's good or bad is up to us. It’s choices that we make about how we use the technology, like what we decide to put it to. If you look over the long arc of history, it has been, by and large, positive. And in a sense, it has to be, otherwise you have no progress over very long periods of time. If you choose the wrong path every time, you will collapse. We occasionally choose wrong things and then we course correct. We’ve been building, building, building to a world that is so much more prosperous than the world that you or I were born into, or certainly that my great grandmother was born into, where there’s less violence, less disease… There's all of these things, there's disease and suffering. All of the things still exist. It's just they're much less now than they were decades ago. And, the question we have to ask ourselves as people who build technology is, are we doing the best job that we possibly can to build technology in a way where 40 years from now, our children and our grandchildren will look back and say, wow, this was beneficial, this mattered, this made all of these things that we care about with justice and the human condition better, not worse. We’d better get it right.

Yeah, I think there's so much stuff. One that you mentioned in particular, the pattern I think is really interesting. So one of the things that modern generative AI ought to be very good at is helping everyone navigate the complexity of the world. The example that you gave is a good one. We’ve been doing some work with an organization in the United States called Share Our Strength, whose mission is to feed kids. It’s just sort of amazing how much childhood hunger there is and what the knock on effects of childhood hunger are. A tremendous amount of what gets diagnosed as ADHD is actually just kids being hungry. When you're hungry, you can't focus, you can't learn. It just sort of becomes a snowball effect in people's lives. And so the thing that we're doing with Share Our Strength is we're trying to figure out how to use generative AI to help people navigate the entitlement programs that already exist, that already are funded, the money has been appropriated and it's sitting there unspent because people either aren't aware of it or can't navigate the bureaucracy to sign themselves up for it. And this ought to be a great thing for generative AI to help with. I think that that as a pattern is really, really extraordinarily powerful. I think educational equity is another thing. My daughter goes to a very good school here in Silicon Valley with excellent teachers. My school did the best that it could do back in the 70s and 80s in rural central Virginia. But if I look at my school versus my daughter's school, there’s not even a comparison. The thing that I think AI could do is close that gap to try to make a higher quality of education and learning and enablement more equally available, not just across the United States, but across the world. I think there's some truly exciting stuff happening right now where the fundamental pattern driving progress in AI right now, this notion of self-supervised learning, that you can transform, compute and data and increasingly just compute into AI systems that can solve very complicated problems, is applicable to more than just language. And so we're doing some really fascinating work at Microsoft Research on how you can apply these techniques to physics and chemistry and biology to help with building the next electrolyte for energy storage or designing a therapeutic molecule that will cure a disease. It’s the thing I think is underappreciated because they're more complicated for a layperson to perceive what the progress is versus what’s happening with language agents or linguistic base agents. The progress is extraordinary. These are not little steps we're making. They are just unbelievable in some cases. The biggest jumps forward in progress that we've ever seen. It’s an exciting, exciting, exciting time.

I think what Bill has done with the Gates Foundation is really extraordinary. The thing I think that makes their accomplishment extraordinary is they've just picked a handful of things to focus on and they have just driven very intensely to try to solve. The work that they've done in public health is just extraordinary. They have chosen these acute problems that are not getting solved fast enough that don't have a natural mechanism to get solved. If something doesn't change, they're going to be just as bad or worse 50 years from now. So the thing that we focus on at the Scott Foundation is trying to identify and relieve structural poverty cycles so things that lock people into intergenerational poverty. And the reason that that's our focus is because both my wife and I grew up not terribly privileged. We had some privilege, because even though my dad went bankrupt a couple of times and we had a lot of financial hardship to go deal with, we're still more privileged than someone who was born in equatorial Africa, for instance, in the 1970s. I think you always have to appreciate what you have, but I think both my wife and I, in a sense, look at ourselves and we're like, wow. There are a handful of things that happened to us, that if they hadn't have happened, we would have had very, very, very different lives. The point of the foundation is what can you do to go engineer a helping hand, to go engineer some of the good luck that my wife and I had so that you don't have to have so much left to chance in getting people to snap out of a poverty cycle. We try to invest in organizations that are entrepreneurial, where they’re thinking about how you can take an investment and with leverage, with tools like AI, for instance, you can go tackle a problem and like get a big benefit.

So there's a bunch of stuff, actually. Since we don't have a lot of time, maybe I'll just sort of say the most important ones. I think it is really important for your own positivity and the positivity that you project in the world to be grateful. No matter how crappy your day is, to ground yourself in “what's one thing that I can be grateful for today”? The reason I think it's so important is because once you start feeling gratitude, there's almost a snowball effect to it. Once you've felt the first thing you're grateful for, it's easy to see all of the other things that you should be grateful for, and it helps you be humbler. I think one of the foundations of having compassion in your life, of being able to put yourself in someone else's shoes, like not feeling what they're feeling, but just trying to understand what their point of view is and why they might be doing what they're doing. Even if your knee jerk reaction is like, this is irritating. There’s always a reason for everything, so just a little bit of gratitude can go an awfully long way.

I don’t know. Honestly I'm deeply uncomfortable with anybody paying any attention at all to what I'm doing. I would like the opportunity to work with people who care about what it is that they're doing. The thing often that I'm proudest about. I just got a note this morning from a friend – someone that I hired 19 years ago, actually – they sent me a note on the 19th anniversary of their start date at this thing that we were both doing back then and thanked me. I should be thanking them because the thing that makes me feel the best about what I've done is having some tiny little impact, a positive impact on someone's career. Just seeing the things that those people go do after we are no longer working closely together just fills me with joy. Just knowing that... I kind of don't care what everybody thinks of me, but having those people that I've worked with, like, feel like it wasn't a waste of time working with Kevin Scott. That's what I would like. And obviously I care a lot about my family. I want to support them in their ambitions and dreams and have my wife be successful and my children be successful and do good things in the world and believe that they had a husband and a father who supported them being their best selves.

Thank you very much for having me on. And thanks for putting positive energy out into the world.