Stanford Professor Erik Brynjolfsson on How AI Will Transform Productivity

ELISE HU: That was professor, researcher, and author Erik Brynjolfsson. He has spent decades examining the ways that information technology is transforming business and the economy. His research has increasingly focused on artificial intelligence, and he’s going to share his perspective on AI’s potential to transform work. Here’s my conversation with Erik. 

ELISE HU: Thanks for doing this, Erik.

ERIK BRYNJOLFSSON: Good to be here.

ELISE HU: Why don’t we just have you introduce yourself and what you do.

ERIK BRYNJOLFSSON: I’m Erik Brynjolfsson. I’m the director of the Stanford Digital Economy Lab. I’m a professor here at Stanford in the department of economics and business school, and mainly the Stanford Institute for Human-Centered AI.

ELISE HU: Well, we all read the headlines, and it’s clear that something really interesting is happening with AI right now. So Erik, how would you describe in layman’s terms this moment and what’s going on?

ERIK BRYNJOLFSSON: Well, there’s definitely something big happening. I think a lot of the excitement is on ChatGPT, and maybe DALL-E, and these are both examples of a new class of AI called foundation models. That includes not only these large language models that can write stories or poetry, email, ads, and many other types of text, but also, like DALL-E, they can make images, there are others that can make videos, audio, and even write computer code. These technologies have the potential for really transforming the economy, I think, creating trillions of dollars worth of value, but they can also be very disruptive.

ELISE HU: What’s happening exactly to make these AI breakthroughs possible?

ERIK BRYNJOLFSSON: Well, it’s really a confluence of three things. One is just a lot more computer power, orders and magnitude, more computer power than we had 10 or 20 years ago. The second is a lot more data; that may be the most important thing. Over the past couple decades, almost everything has been digitized. And that provides the raw material for these machine-learning engines. And last but not least, we have much better algorithms. People have figured out new ways of using these data and applying computer power to them to answer questions that we couldn’t answer before. 

ELISE HU: It is exciting. It also, of course, leads to so many questions, which then dovetails with your career, which is focused on the economic impact of digital technologies. What would you say is the through-line that connects your research and inquiries and this particular moment that we’re in?

ERIK BRYNJOLFSSON: Well, for a long time, I’ve been interested in how computers are changing the world, since I read Isaac Asimov and other science fiction. And when I went to grad school, my professor asked me to plot computer power in the economy, and every time I plotted it, there were these exponential curves growing really, really rapidly, which was way back in the 1980s. But I could see, if this continues, anything like that, just astonishing things are in store. And indeed, things have continued pretty much along those lines, and we’re beginning to really change the world.

ELISE HU: What kind of astonishing things are bearing fruit? What are we seeing right now?

ERIK BRYNJOLFSSON: Well, lately, I think these large language models, or foundation models, have been just very striking. They’re able to generate new kinds of content that previously only humans could do. I talked to the inventors of these technologies, the people developing them, even they are surprised at some of the capabilities. So these emergent properties, having it understand right from wrong, or be able to create new kinds of insights, or to speak in different voices, even to write computer code or play chess. These are things they didn’t expect it to be able to do, but were surprisingly good if you ask it the right questions. 

ELISE HU: Okay, this of course, is the WorkLab podcast, so we want to know how this is going to have an impact on workplaces. What do you feel like leaders in companies right now need to know about the potential of AI.

ERIK BRYNJOLFSSON: Let me start by saying that these technologies are racing ahead at really an unprecedented rate. The past couple of years have been some breakthroughs. And our organizations are not keeping up our skills, our institutions, even our laws are falling behind. And in that gap between what the technology can do or what the technology demands, and what our organizations and our human creations are doing, there’s a bigger and bigger set of challenges and a bigger and bigger set of opportunities. We need to close that gap. And we shouldn’t do that by slowing the technology, we should do that by speeding up our adaptation. The changes that are happening right now, they’re affecting almost every part of the economy, and many of them are quite different than what happened in the previous 10 years. In the past, we had relatively slow growing productivity; I predict faster growing productivity. In the past, we had growing inequality; I think these tools can, and hopefully will, lead to more widely shared prosperity. If we play our cards right, the next decade could be some of the best 10 years ever in human history. 

ELISE HU: What should companies be thinking about? What should leaders be thinking about in order to adapt quickly enough or be agile enough?

ERIK BRYNJOLFSSON: Agility is key. So you’re going to have to have people who understand the technology, but also understand the business needs, what the customers are looking for. And that’s a rare combination. In particular, I think what we need to have is an ability to change our business processes and our organizations, and not simply bolt on the new technologies to the existing way of doing things. Very rarely, is it possible to have a plug-and-play use of the technology. In almost all cases, the big benefits come from doing new things that we hadn’t done before. And that requires a lot more creativity on the part of managers and entrepreneurs than simply saying, what are we doing now? And how can a machine replace a person?

ELISE HU: You mentioned there’s nervousness about AI eliminating jobs because of its potential for productivity gains. And because machines can often be substitutes for human labor, this could also mean that workers could lose power and become increasingly dependent on those who control the technology. But you’ve sketched out a different vision, something called complementary AI. Can you talk a little bit about that?

ERIK BRYNJOLFSSON: I mean, that’s probably the most common question I get. And the reality is, is that yes, there are a lot of possibilities where the technology can replace some existing jobs. But I don’t think that’s the main effect, and that’s not the main opportunity. The bigger opportunity is this complementary AI. What that means is enabling people to do things that they hadn’t done before. In fact, if you look through history, most technologies have ended up complementing humans rather than substituting for them. The people who were weaving cloth early on were worried that the spinning jennies would replace them and drive down wages. It turned out they were right, the wages for those skilled artisans did go down. However, in most cases, the wages of workers have gone up, because in most cases, the technology has amplified what people can do. The way, say, a bulldozer allows a person to move more things physically, or software has allowed people to affect a lot more types of problems than they could previously, and that means that they increase wages. So over the past couple hundred years, have wages gone up or down? Well, they’ve gone up about 50-fold. I should note that it’s not inevitable, it doesn’t always happen—the past 20 years have in some cases been a divergence from that pleasant trend from the previous 200 years. Many kinds of labor have actually seen the hourly rate go down, where people who have a high school education or less are earning less in real terms than they were a couple of decades ago. Those with college or professional or graduate educations have seen continued increases in wages. So we’ve had a divergence. I think the big challenge for us going forward is whether or not we can use these technologies in a way that creates shared prosperity and doesn’t have this increased polarization or inequality. 

ELISE HU: What are a few actionable things that we should be thinking about, or thinking about doing, to help close that gap?

ERIK BRYNJOLFSSON: Yeah, I’m glad you asked that question, because I think this is the big challenge for our society coming up. The reason I started the Stanford Digital Economy Lab was to help close that gap, you know, not by slowing down the technology, but by speeding up our adaptation. And there are a number of policy things we can do in terms of investing more in education and focusing more on having people ask the right questions, be creative, and less on the rote tasks that the machines can do really well. There’s a big role for technologists to rethink the way that they develop the technologies. Alan Turing was a great researcher, and he came up with this evocative idea of the Turing test, which is, can we make an AI that is so human-like, so similar to people that we can’t tell the difference between them. And I think that’s inspired a generation of technologists. Also, I think it’s exactly the wrong thing to do. In fact, it can lead us into a trap, which I call the Turing trap. I’ve looked at it more closely, and when you have a technology that imitates humans, it tends to drive down wages; when you have a technology that complements humans, it tends to drive up wages. So we should not be making machines that are close images of ourselves, we should be making machines that are as different as possible from us and allow us to do new things. It’s a different approach to technology. And most importantly, I think managers and entrepreneurs need to rethink the way they’re using the technology. Don’t just look at your existing processes and think, oh, how can I replace this worker with a piece of software or an AI? It’s okay to drive down labor costs. I mean, it’s great for us to be able to get things cheaper. But there’s way more upside in doing new things, or delivering things in an entirely different way. That takes a little more creativity on the part of managers but ultimately leads not just to more total output and more value created, but also leads to more broadly shared prosperity because you’re keeping humans as part of the production process and not replacing them. And if all three of those groups—policymakers, technologists, managers and entrepreneurs—each pursue that kind of path, I think we’re going to have some of the best years ahead of us that we’ve ever had.

ELISE HU: How could it eventually lead to adaptation of human capabilities? Technology changes us, right? Like, I don’t remember phone numbers anymore because I don’t have to. I don’t really read a map anymore. I mean, these are obviously very reductive ways that technology has changed me, but in what ways would AI change who we are?

ERIK BRYNJOLFSSON: Well, this isn’t the deepest point, but I find that it’s already affecting the way I assign homework and the way that the students do their homework. As you’ve probably heard, these tools can make it very easy to generate an essay based on a prompt, and the essay may be quite good. So a lot of professors, a lot of high school teachers, are wondering, how can I assign essays for students to write if the technology is just going to do it for them? I think the answer is it can and should change the way we’re doing it. I mean, a few people are saying, so we’re gonna have to find a way to detect them and ban people from using it. I was disappointed, one of the big AI conferences even had a requirement that none of the submitters—AI researchers—were allowed to use these tools when they submitted their papers. I think that’s the wrong approach. A better approach is to, as you say, redefine what it is that we’re expecting from people. And so, if when I’m teaching my class in the spring, I’m going to tell the students, go ahead and use the technology, but I expect your essays to be that much better than the kids’ last year. In fact, I’ve already put the questions through the ChatGPT and the other tool, so I know what a conventional answer would be. That’s your starting point. 

ELISE HU: What kind of roles or professions do you see potentially being most transformed by the AI that we’re seeing these days?

ERIK BRYNJOLFSSON: That’s a great question. It’s a trillion dollar question. I think it’s gonna affect almost all of us. You know, having worked with large language models, I see that a lot of creative work is tremendously being affected. I was just talking to a CEO, who was trying to figure out what the right KPIs were going to be. So he went to ChatGPT and had it suggest some based on his company’s goals, and it came up with a great list. He said he didn’t use them verbatim, but it was a great spur to doing it better. I’ve seen people use it to help design new kinds of pools, new kinds of songs, come up with all sorts of creative work. I’ve used it myself in some of my research writing. It’s helping people at all parts of the spectrum, not just the less skilled information workers that were affected by earlier technologies.

ELISE HU: Does this idea that humans essentially have to really lean into that which makes humans unique, the ability to ask the right questions, the aspects of humanity that we have, like perspective and surprise that machines do not have… does this portend a change to the larger labor force and the economy?

ERIK BRYNJOLFSSON: I think there are some really disruptive changes coming to the labor force and the economy. And I don’t fully understand them all, I’m trying to study them. One of my big projects is to go more in depth to some of the changes. But I see some of the broad outlines. And I think, as you suggested, asking the right question is key. There’s a new profession called prompt engineering, which is literally telling the language models what you want it to do. And it turns out that, depending on how you ask those questions, you can get better answers, more accurate answers, more insightful answers, more creative answers if you structure it the right way. And that may be really where humans can add the most value. One of the things that I said in some of my books is that as these tools become more and more powerful, that means, almost by definition, that we have more power to change the world. And that means that our values matter more than ever before. So it’s time for us to think more deeply about what it is we want the world to look like and how we want to use these tools to reshape it.

ELISE HU: What do you recommend that leaders be saying to their teams, their employees, their staff who are worried about this and worry about the challenges or the existential threat that’s posed by AI.

ERIK BRYNJOLFSSON: I wrote something with Andy McAfee in a Harvard Business Review article a few years ago, I said that AI is not going to replace managers, but managers that use AI are going to replace managers that don’t. And I think that’s even more true today. So my first piece of advice is to have everyone on your team get familiar with these tools—they’re kind of fun to play with. And so I was talking to one executive and he was declaring a full day where he asked everyone in his company to just spend the time playing around with the tools and get a sense of what they can do for their jobs and for their company. That kind of familiarity is going to create a lot of new opportunities—figuring out the new things that the technology can do for individual workers, not just taking stuff off the shelf that’s already been developed by some start-up. And in some cases, developing things that go on top of it to make it more useful for particular business needs that you have.

ELISE HU: When you bring up business needs, that reminds me of the potential for AI to help us deal with maybe some of the more unpleasant aspects of work. I’m talking about increased complexity and pace of change, information overload, too many meetings, just the drudgery or more tedious tasks. Where do you see AI helping solve some of those pain points?

ERIK BRYNJOLFSSON: Well, AI can do a lot to help solve and also exacerbate them depending on how we’re using them. But one way they could help solve them is that, just as the tools can generate new text from simple outlines, it can also go the other way around. You can give it a long article, you can give it even a book, and it will distill down the essence of it. And you can have it connected to that report in such a way that when you want to double click, zoom in, on one piece of it, it’ll bring you to the relevant part of the document. So it’s like an incredibly smart research assistant, or maybe yourself spending weeks going through stuff, so you get a set of notes that are very relevant. This is all done in an automated way. 

ELISE HU: Yeah, I can see this really useful to law firms, right, that are involved in big litigations.

ERIK BRYNJOLFSSON: Absolutely. I mean, there’s already, for some years, there’s been tools to help with document discovery, sifting through and finding key words or phrases. But now it can go further and understand the concepts that are in there and summarize them and even come up with counter arguments as needed.

ELISE HU: Amazing. This conversation reminds me of that famous John Maynard Keynes prediction a hundred years ago that we would only be working 15 hours a week. Because over time, according to Keynes, thanks to machines and technology and new ideas, people would get more productive. And the machines and the technology would take over for a lot of the more menial or tedious tasks that humans were doing. Why was that prediction wrong?

ERIK BRYNJOLFSSON: Yeah, that’s a great quote, and I encourage anyone who hasn’t already done it to go ahead and read his essay, “Economic Possibilities for Our Grandchildren.” And first, I should say he got a lot right. So his prediction was based on the assumption that productivity growth of a couple percent per year would continue exponentially improving living standards for the coming century. And it more or less has. There have been some ups and downs, but by and large, our growth has matched what his prediction was. The difference is a failure of imagination of all the things that were created. So wealthy people of his era, they lived in manors, and maybe on occasion went foxhunting, but there wasn’t a lot else they could do with their wealth, and so it sat around. But now we have all sorts of other gadgets and fun things you can spend your money on. And another thing is more sociological. I’m an economist, but I’ve come to recognize that a lot of people get meaning from their work, and rather than retire, many people feel like they want to continue to contribute in some way. Or maybe they feel compelled to do it. So for both those reasons, people have continued to work quite a bit, even though our productivity is vastly higher than it was when John Maynard Keynes wrote those words.

ELISE HU: So if that’s the case, and there’s always going to be this sense of human striving, and AI helps increase productivity, then what will humans be doing?

ERIK BRYNJOLFSSON: Well, for a long time, I don’t think we’re gonna have any shortage of work to do. There’s no danger of mass unemployment. When I look around in terms of healthcare, child care, elder care, cleaning the environment, invention, art—there are so many things that humans are uniquely good at doing. Perhaps the most important one is what Pablo Picasso pointed out is, asking the right questions. There is going to be potentially a jobs quality problem of, are we going to get the wages paid right? Are we going to have enough of the kinds of jobs that are really rewarding to do? And we can work on that and do a better job, I think, but if we use the technology mainly to complement people rather than substituting people, I think we can have a situation where most people have ways to contribute to society and the technology is amplifying those capabilities. And too many people, I think, for a failure of imagination again, assume we’ll be producing the same things but with fewer and fewer workers. And that’s certainly one way to increase productivity. But we can do a lot better by increasing the amount that we produce. And that can be done, not just in terms of quantities, but also in terms of quality, or new kinds of output, as well.

ELISE HU: How can folks better understand the AI possibilities that are with us now and maybe play with these tools?

ERIK BRYNJOLFSSON: Well, more than two years ago now, I was asked to comment on a paper at a conference with the National Bureau of Economic Research. And I happened to be the last speaker, and it was a conference about AI and how it was changing the economy. So I thought I would not just talk the talk, but walk the walk. I put it through GPT-3, and I asked it to help with my remarks. And I said, do it in the style of Erik Brynjolfsson, and it came up with some pretty good sounding stuff. I have to say, when I first read it, I was like, hey, wow, this is really good. And when I read the second time, I was like, you know, that’s not quite right what it’s saying here. It definitely did sound good. I had the audience listen to that. They thought that was kind of cute, it was fun what it did. But then what I did next, I thought was really interesting. I asked it to redo it in the style of Taylor Swift, just for kicks, and it wrote this beautiful poem with all these evocative metaphors—

ELISE HU: So it’s very Swiftian.

ERIK BRYNJOLFSSON: Yeah, absolutely. And I was like, it captures what I was thinking, but in a much more beautiful way. And I think in a way, actually, the audience got better than they would have with my prose. So that was a real aha moment for me. Some of those metaphors that Taylor Swift came up with, I was like, that’s so good. I wonder where, you know, where GPT got this from. No one had ever said these things before, it was completely original, but also quite creative and evocative. So that’s something that I imagined more people would be doing. I’ve been getting, I don’t know about you and your friends, but I’ve been getting poems from my friends and relatives about me or something else, and some of them are pretty silly. But we’re having fun with it, and I think it’s changing our lives for the better.

ELISE HU: Okay, and just one last thing. What kind of future do you envision if we leverage the potential of AI?

ERIK BRYNJOLFSSON: I don’t think any particular future is inevitable, and so I think the way you phrased the question was exactly right. If we do it wrong, we could have increased concentration of power and wealth, a lot of people losing their economic wellbeing. But if we do it right, I think that it will not only lead to shared prosperity but also lead to a greater rate of invention, more creativity, and people inventing new drugs, new science, new types of structures, new materials that hadn’t existed before with the help of these tools. And I would not be surprised at all if the next decade was one of the most productive decades ever in history, because these tools will allow us to do new things that we never did before.

ELISE HU: And yet, somehow, we’ll still want to work.

ERIK BRYNJOLFSSON: Well, a lot of people will. I think the definition of work will change a little bit. I think we’re fortunate—probably my great, great grandparents would not recognize what I do every day as work, you know, they’d say, I don’t get it, you’re not lifting anything. My hope is that going forward, more of the routine, the boring, the rote parts of the jobs will be done by machines, the parts we don’t like will even be handed over to robots. And we’ll be able to spend more time on asking the right questions. Also on interacting with other people, that I should underscore, that’s another thing that I think humans are uniquely good at and most of us enjoy, which is interacting with other people, relationships. I think most of us wouldn’t want to have a robot taking care of our babies or our grandparents. We want to be able to interact with them ourselves. And that also is a uniquely human skill.

ELISE HU: Well, Erik Brynjolfsson, I appreciate the conversation. I’m sure you can tell how much I love talking through these concepts and possibilities. Thank you so much.

ERIK BRYNJOLFSSON: It’s been a real pleasure. I love talking about it with you.

ELISE HU: Thanks again to Erik Brynjolfsson. I loved that conversation. And that’s it for this episode of the WorkLab podcast from Microsoft. Please subscribe and check back for the next episode, where I will be speaking with Gloria Mark. She’s an author and professor of informatics, exploring how leaders can help their teams regain control of their attention and restore balance. If you’ve got a question you’d like us to pose to leaders, drop us an email at [email protected]. And check out the WorkLab digital publication, where you will find transcripts of all our episodes, along with thoughtful stories that explore the ways we work today. You can find all of it at microsoft.com/worklab. As for this podcast, please rate us, review, and follow us wherever you listen. It helps us out. The WorkLab podcast is a place for experts to share their insights and opinions. As students of the future of work, Microsoft values inputs from a diverse set of voices. That said, the opinions and findings of our guests are their own and they may not necessarily reflect Microsoft’s own research or positions. WorkLab is produced by Microsoft with Godfrey Dadich Partners and Reasonable Volume. I’m your host, Elise Hu, and my co-host is Tonya Mosley. Mary Melton is our correspondent. Sharon Kallander and Matthew Duncan produced this podcast. Jessica Voelker is the WorkLab editor.