Speaker 3
Well, first, before we talk about
Speaker 2
that, I just want to mention how Adam and I became friends, which is that I don't remember whether I tweeted something or wrote it on Bloomberg. This was a very long time ago, 2014, I think. And then Adam randomly called me up because my number was publicly listed, you know, because I was a professor at the time. Adam randomly called me up and wanted to debate about some of these things. And we had a nice conversation. And I think I actually skipped a meeting to do this. And then we talked to each other on Twitter and then have been friends ever since. But anyway, so this is a long tradition. But all right, onto the AI thing. So basically people are starting to do experiments with, uh, you know, AI and productivity. So Eric Brynjolfsson, we all know he did some of these, um, but there have been five or six so far that have been done. I, and, um, all of them have shown the exact same thing, which is AI is a big productivity boost for the low productivity people and a modest to zero productivity boost for the high productivity people on a variety of tasks. So you have essay writing, call centers, coding, and business writing, and just a number of other kind of tasks that you think, and they usually use GPT-4 is the thing they test, occasionally something else, but usually GPT-4. And they find that the lowest performers get boosted while the top performers get either a very small boost or no boost at all. And so this made me really think, you know, since I've been a kid, we've sort of seen nerds, you know, like us get increasing returns. You know, when, when, when I was a kid, there was still a little bit of nerds getting, you know, being second class in society. The idea that nerds would just go in the basement and do something technical while the back slapping bros went and made all the deals, you know, on the top floor of the, of the building and got all the money. And so that was still kind of the way people imagine the economy being, it was already in transition, but, but that was still a stereotype. And I feel like by, by now, you know, by the 2010s, things had completely inverted. Nerds are the masters of the universe. You know, you have to, yes, you still have to know how to make deals and slap backs, but you also now have to know how to code or what, you know, what technical things mean. And it's people from nerdy backgrounds who are getting all the money. And so I thought, you know, it was really the age of the nerd and you can see the college wage premium just explode in the, in the eighties and then sort of creep up in the nineties and two thousands. And so human capital had this increasing return to it. It may, you know, and I think this went along with a general shift in America's position in the world. You know, we went from being the world's factory, where we made all the cars and the steel and the aluminum and all that stuff. We went from being that to being the world's research park, where we wrote the software and did the bio research and you know did the finance stuff and we did all the the knowledge industries and I think that that big change has reverberated throughout our society in many ways and one of the main ways is that you know working as just an engineer for a large tech company you now get paid hundreds of thousands to millions of dollars, you know, while a regular, a regular worker is still getting paid, you know, $50,000 a year, $40,000 a year to just do a regular job. The people with the knowledge at the top of the skill distribution have just done so much better. When I looked at those results from AI, I thought maybe this is the, this is one thing that could help reverse this. Maybe AI will let random, people with very average skills who don't win hackathons learn to code well enough to actually get software engineering jobs. Of course, that's going to drive down the overall return because you have an increase in supply, But maybe it will, you know, just like mechanized, you know, garment making machines in the industrial revolution helped regular people be able to sew things and weave things as well as the masters. Maybe that's coming for AI and maybe the age where human capital was absolutely everything is going to wane a little bit or shrink a little bit and nerds aren't gonna rule the universe quite so much as as we have for the last 30 years. As usual, Noah, you make a compelling case. And, you
Speaker 1
know, going back to our origins or our discussions back and forth, I'm definitely the slightly more crazy one here compared to Noah because his what combines is you know when he writes pieces and he writes a lot it turns out I agree with quite a few of them but sometimes I have questions issues ideas and sometimes debate his patience for Twitter DMs man is just unreal you know this piece I thought was really fascinating for me because I spent a lot of time thinking about this you know my first love in school I was in science, but I ended up focusing on human computer interaction, which is kind of all about what I think of are the irrational, emotional, kind of perceptive ways that humans interact with technology versus the technology itself, which tends to be ones and zeros and very rational. And because of my love of economics and kind of finance, that's always overlaid. Well, how does this play out in the economy? I find these topics really interesting and they get into my work. So when I read this piece, I thought part of it was very correct on two dimensions. It got me thinking, you know, one was we have all these industries, right? All these areas, which we tend to label as low skilled or unskilled. By the way, I don't think that label is accurate or the right way to think of things. But we have all these industries that have actually defied a lot of productivity boost with this recent IT boom, right? Like there hasn't been as much productivity, there's actually been a slowdown on the overall numbers. And so you got me thinking, well, maybe AI is actually a thing we've been missing is that, you know, there's been intermediaries needed between people who aren't as comfortable with technology, don't have as much of a growth mindset about using technology to like doing things the way they do them. And maybe AI could be the thing that breaks through, maybe not for everyone or every industry, but in a lot of verticals that we haven't seen it before. The other thing I thought that was very true is that despite Silicon Valley mindset around engineering and software, we all talk about how much is that Netflix person making working on AI or, you know, those Google, you know, Google stock units. The truth is, is that actually the data on computer science in general and programming, even going back to when I was in school, I remember one of my good friends, we had a debate when I was just early and it was a classic debate between CS and electrical engineering. And at the time I went to school, electrical engineering paid more. And everyone said, well, of course, electrical engineering pays more. It's IEEE accredited. Why would you pay for something like computer science? Is that even really engineering? This was like Stanford talk in the early 90s. And so, but the truth is, if you were a VB programmer in the bowels of a Fortune 500 company, you weren't actually making a huge amount. You weren't making stock. And so I think that tends to get overstated by these sensationalists. A lot of our discourse lately has been very much not even just on the 1%, but kind of the 0.1% of kind of outcomes. And I think that distorts things. The only problem I had with the piece, and, you know, when I was thinking about it was really it's our definition of skilled and unskilled using proxies like college educated, how much college they went through. There's a lot of data that the government collects and we use the data as economists and finds because it's available and it's been standardized and we've been doing it for a while. But at least in my career, I haven't found that there's a great overlap between those definitions and what leads people to be successful with technology. And that's always bothered me. I mean, these are all correlated. So I'm sure over large data sets, we could have an argument about maybe it's the same thing. But you hear these words in Silicon Valley, a growth mindset, this idea of learning new technology, asking the question of how this can help you do better with it. It's somewhat of a progressive idea in some ways. It's somewhat of a growth idea in some ways, but it fights the conservatism of, no, no, we've always done it this way. Keep doing it this way. This is how to make something of the highest quality. And so across different industries and verticals, you can find people who have a growth mindset, right? Like I've worked with amazing stonemasons who actually are always thinking about better ways that they could lay a path or better way, you know, carpenters who better ways they can do things with wood, et cetera, and they embrace tools. I think that across verticals, traditional unskilled industries are actually filled with people who have a growth mindset, who will embrace technology. I think you're going to find some people who fight it, who don't embrace the new technology and new platforms, new ways of doing things. Now, obviously, I'm very bullish about AI. I think the people who embrace the new technology and figure out how we can do things at a hundredth the cost or a hundredth the time that we were able to do before will inevitably outcompete and build businesses and products that are not only better than what we have now, but actually they will do things that aren't even economically possible right now. So the biggest problem I had with the thesis was just more around the definition of skilled and unskilled. But like I said, it might come out in the averages, but I'm already seeing it like on my team at Daffy. Our CTO did a full walkthrough. Actually, we have a deal. As it turns out, OpenAI uses Daffy internally. So we've had a bit of a front row seat to some of these innovations. But I'll tell you, the engineers I see embracing AI do phenomenal work, and they do it faster with higher quality. The writers who embrace it, the designers who embrace it do phenomenal work. So I'm very bullish about in general. I'm not sure it's as simple as unskilled versus skilled.
Speaker 2
Well, I mean, I think that when we're talking about the real world, obviously, it's going to be more complex. You know, when when we talk about those experiments that people have done to get those, you know, to get interpretable results, they had to define skill in a narrow way, but they didn't define it by education. They didn't say, you know, which of our, which of our test subjects has a degree, you know, they all had degrees. You often, you know, sometimes they're all at the same college even, but they, they measure their initial skill on a task. So they would have them write some code, see how, you know, evaluate how good the code was, and then have them write code with, you know, GitHub Copilot in that case. And so then see how well they did with GitHub Copilot. So you had, it was very narrowly defined skill on a task. And I think that you're right, that in the real world, skill is very broad. There's, you know, you could call entrepreneurialism a skill, work ethic, self-improvement, self-starting. These are all skills that we talk about and that are written on motivational posters, but that we don't have a good way to define for research purposes very clearly. And I think you're absolutely right that in the real world that matters. And these experiments are just like toy experiments in a lab, so it doesn't necessarily translate. But at the same time, I think that getting better on specific skills, you can certainly have bottlenecks. So someone in the tech world who can do everything except code well may be held back by that bottleneck in their skill set. And, you know, of course, the typical response is growth mindset, spend years fixing the bottleneck, spend nights and weekends fixing the bottleneck in your skill set. And maybe you can do that, but it takes time and, you know, there's an opportunity cost to doing that. Now, maybe you have this amazing machine tool, you know, like you have this tool and you can just load it up and use GitHub Copilot to go from a mediocre coder to a decent coder overnight. And now that bottleneck in your skill set is fixed. So if you're running a company and you need to write a little bit of the code yourself because you don't have someone to do it right now, it's not going to suck and it's not going to fail because you have the copilot to help you. And I just made that example up. But I'm thinking that when you have a bit, but of course, you know, AI helps with much, much more than just coding. So if you have, you know, a tool, a general purpose tool like AI that can fix all those little bottlenecks in your skill set, maybe people who, you know, would have been mediocre will now go to decent or
Speaker 1
great along a wide spectrum of skills in the real world. Yeah. And I think that's a fair point. And this is not new for technology platforms. People at the highest end of skills, like take my co-founder at Daffy, one of my favorite engineers, most talented, probably won more hack days at LinkedIn than any other person. Alejandro Crosa, don't recruit him. He's busy. Oh, no, no, no. We just want to give him a shout out. For example, we deal with Daffy with a lot of government data, a lot of regulated data, which tends to not be on the cutting edge of technology. So recently we had to rip through some IRS data that I think was in some older format of XML, right? 20 years kind of old, like I'm sure it was cutting edge at some point. He could do it. It's draining for him. It takes him away from a lot high value tasks, but it turned out the combination of something like Copilot and letting, you know, AI kind of take a crack at things and then finishing up, he would describe it as something that normally would have taken him a few days to do. He was able to do in a day, right? And I've heard designers say the same thing, where they were building out a whole brand expression, et cetera, that normally would have taken them a couple weeks, but they were able to do it in hours. But I think you're fundamentally right. I think there's more room for improvement for those people who are missing skills, right? And we have 50 years of analytics software promising that business users who aren't good with data, who can't program, who didn't take advanced statistics could actually make sense of the data coming back in business. But the thing that I have a slight calm about is like when we talk about things like inequality or kind of the tertiary, it's not just about the median or talking about the high end. Right. They're these different like they're different percentiles. Right. So the truly exceptional folks at the high end. tends to come from a lot of creativity. What can you do with this technology, right? There's a big difference. Economists tend to look productivity as like, here's a task. Can we do this faster, cheaper, better? That's different than inventing new tasks and new systems and new platforms and ecosystems. Now, eventually they have to converge, right? The, all these things have to go to human needs and things that we build and do, but I'm not convinced yet which side of that equilibrium is going to get more benefit from AI. I still, like what still bugs me in the back of my mind is these definitions of skills versus what I've seen in my career, which is some people embrace new technology and ask those questions and some people fight it. I have family members who are brilliant, who have advanced degrees, top of their field, but fight technology. I don't mean this, the printer config thing, because no one can still figure out how to configure printers. I don't know why, but, but I mean like, you know, resisting kind of new ways of doing things, new technology, new ways of running their business. They like the way they run it. You see this in a lot of existing businesses. I mean, a lot of the opportunity in tech that I love exists because of that. Right. The optimization of what exists versus reimagining from the ground up what could be. And I'm not convinced that AI changes that equation completely. Well,
Speaker 2
I mean, we'll see. Most of the people that you talk to about AI's impact on job markets are talking about replacement. They're talking about, you know, dumb people will get replaced by AI and then smart people have to race to stay ahead of AI. They're thinking of this substitutability. They're thinking of substitution, you know, and they're thinking that for smart people, you're going to have to use your smarts to retain tiny little bits of edge over AI where you're competing. And then for dumb people, you're just screwed, go on UBI, whatever with you. And I just think that that's the perspective that think that these econ results mitigate against. And history, too. Because when we look at history, we've seen this again and again, this fear of human replacement that seemed to be validated when a few high-skilled technical occupations like seam or master weaver got essentially devalued by new waves of innovation. You know, telephone operators, where are they? But then, so technology did replace specific occupations. And so we've had these waves and waves of fear about machines replacing human beings since we started using machines. And if you, you know, from the original Luddites who were just like pissed off weaving workers to, you know, even in the post-war years when unions were strong and the growth was very well balanced, you know, and inequality was low and all these things, right? And people were terrified that new machine tools, new industrial technology would mass replace humans. You can find stories in the New York Times about this. And so that perspective has never really gone away. And there's this idea that this time is different, that this time we've invented a machine that replaces humans instead of a tool that complements them. And that may be true because I can never say it's not true, right? To say it's not true is to just make prophecies that I can't support because no one really knows, right? Maybe this time we've invented the people replacer, which we never did before. But the little experiments that I'm seeing are very encouraging because in these narrow tasks, narrow situations, it doesn't look like a people replacer. And if it's not a people replacer and narrow tasks, how the heck is it going to be a people replacer for whole jobs? And so I just don't see it. I think cyborgs, humans working closely with tools, have beaten the alternatives of just natural unaided human skill and of pure capital use with no human input. Again and again, cyborgs have just won again and again a human using a machine tool can manufacture just really really well um you know and uh and and we've seen this um again and again i and now we're seeing it in these little narrow econ experiments too and so i'm saying i just don't think that this is the moment when we switch from benefiting humans as a tool, from upskilling and upgrading humans with tools, to replacing humans and turning humans into essentially useless dross or horses to be sent to the slaughter for dog food. I don't think we're at that point.
Speaker 1
No, and you and I are very aligned on this and agree. I come to it from a slightly different angle, but Andy McAfee was actually one of my professors in business school. Oh, really? He's worked with Eric. Yeah, yeah. It just turned out that he ended up going down this direction. Yeah, he taught operations, so I credit him with anything I do know about real physical operations. And also, that was tougher for me than kind of the virtual world, as it turns out. Who knew that Cranberry case? Man, you know, Cranberry is so difficult. Anyway, I haven't studied everything, but I've studied multiple disciplines. And sometimes you see different lenses on the same problem. And sometimes there's reassurance when you see the same, different lenses leading you down the same path, right? You know, philosophy, it can be politics, it can be economics, it can be technology. But I agree with you, like the basic fundamentals of productivity, if we take the economic lens, always leads to fewer people being able to do the same thing, right? Like what used to take 1000 people, if you compound whatever percentage over a period of time, it'll go from 1000 people to 990 to 950 to 900. And it turns out that curve keeps going. As long as you believe you can increase productivity, we could talk about who captures that value. They are now more valuable. And actually that entire vertical, that industry, because we live in a competitive economy, right? Those people could be doing something else by it being more productive. It actually makes that area more competitive, right? The fact that designers can do more actually raises the value of designers, even if you don't need as many designers on any particular job. And that economic competition tends to lead to getting more, not less of those things. So you might have factories might not take 2000 people, they might take hundreds, but you get more factories and you get them in more areas and more verticals where they may not have made sense before. So listen, I think a lot of smart people debate this back and forth, but I'm definitely in the camp that sees AI as a productivity tool, first and foremost. I don't see it as eliminating people. I do think that there is a, and I've seen this in many ways in my career, as short or long as you think it is, where, you know, the people who embrace the new technology and find ways to either solve problems that weren't economically solvable before, or go after valuable problems in a way that's a tenth the price, or 10 times the speed, there is going to be a short term advantage for the people who push in that drive. I'm already seeing it. I mean, just think of translation alone. Think about like when I work with contractors, sometimes there's a language barrier, right? And the person who can speak to the customer in their native language actually does better. I have a good friend actually who speaks fluent Chinese and Spanish. And so it turns out in California, that is a very valuable skillset to have. And as a result, he can talk to customers and do deals that other people can't. What if that friction goes away? What does that mean? Maybe there's other skills that dominate. Maybe the speed you do the project, the quality you do the project, all these other things. So I think people underestimate people in different verticals in general, especially in Silicon Valley. There's such an overemphasis, I think, on software and particularly the slice of software that we do in venture-backed startups. One of the things I like about your piece is, Noah, and talking about the economy in general, is you realize that that's not the biggest piece of the economy. That's not where most people work. It's not how most people make their money. It's not core to most people's lives. And I'm very bullish on technology. There are very few who are more bullish than I am. There's a few, but not many. But when I look at AI, I still see people looking through a very, very narrow lens. That's well said. We've been talking about one of the
Speaker 3
main concerns that people who are concerned about AI have, which is, hey, what's it going to do for labor? Just to list out the other concerns, some people are worried that it's going to, that other humans are going to use AI in very dangerous ways, putting aside any sort of, you know, supernatural, you know, concerns. And other people are worried that AI is going to spread all sorts of misinformation or say very mean things or harmful things towards vulnerable populations. This is kind of the AI ethics community. And then there are other groups of people more like the Ucow, Ucow, Ucow, Ucow, Ucow, Ucow types, um, who are worried that no, actually this could mean sort of, uh, you know, game over for, for humans on, on some timeline if we keep up the, the, the, the, the, the, the rate. And so, yeah, just to list out there. Different groups of people that have different kinds of concerns with AI, some, some stemming, uh, sort of short-term and some, you know, existential. Listen,
Speaker 1
um, unfortunately, some of the logic that Noah put forward actually cuts both ways. If you believe in the cyborgs, if you believe that this isn't about human replacement, it's actually a board amplifying skills and capabilities.