EO - AI Is Killing the Career Ladder. A Stanford Economist Explains What Comes Next | Bharat Chandar

Transcript

Intro

**** · We're seeing that the jobs that are more exposed to AI, the young workers in those jobs are seeing 16% slower employment growth. So that's pretty large. The structural change in AI capabilities that are impacting the labor market, that's not going to be a temporary change. If we really unlock AI's capabilities for helping people learn, it could be much easier to switch between different professions. So I'm really hopeful that we end up somewhere closer to a career lattice that works for workers as opposed to a career ladder where there's much more risk about this technological change. I'm Barat Chandra. I'm an economist at the Stanford Digital Economy Lab and I study how AI is impacting work. I would say over the past year and a half or so, I do feel one of the most important questions in labor economics today is about AI's impact on the labor market.

**** · Once I really started using the tools more and understood their capabilities, that became the focus of my research agenda because it felt one of the most important questions impacting society potentially in the future.

**** · I released a study with my collaborators Eric Bolson and Ryu Chen. We studied how jobs were changing in jobs that were more exposed to AI versus less exposed to AI. and we were tracking millions of workers across the United States using data from a payroll company called ADP.

Canaries in the Coal Mine - What is happening to entry level workers?

**** · One of the key findings there is that overall we were not seeing major differences in employment changes for jobs that were more and less exposed to AI. However, when we focus on young workers, we do see more of a divergence there where the jobs that are more exposed to AI, such as software development, customer service, more administrative roles, we were seeing employment declines and jobs that were less exposed to AI, we were still seeing some continued growth and employment.

**** · And for more experienced workers as well, we were still seeing employment growth that was pretty much on trend.

**** · We're seeing that the jobs that are more exposed to AI, the young workers in those jobs are seeing 16% slower employment growth. a lot of people just starting off in their careers and they're finding it a hard time in doing that. And the reason we chose the canaries and the coal mine title is I think consistent with that view. We want to be tracking these outcomes because we think they could be indicative of potentially future transformative impacts of AI. How much of this is being driven by AI and how is that going to change going forward? We can't be sure whether this is just temporary change in the economy or if it's a structural change being driven by AI. Now, we did test some of the most plausible alternatives that we could think of. So, that includes interest rate changes.

**** · Jobs that are more exposed to interest rate changes are less exposed to AI. One way to think about that is things transportation and construction are very exposed to interest rate changes, but they're really not very exposed to AI. So, that's one key thing and that makes me think that it's probably not interest rate changes that are driving our results. Other things that we tested include tech over hiring. So, we can take out the tech sector, we get similar results. Take out computer jobs. We tested some of these different alternatives and we were still getting this very similar results. So, you're saying, if it's a structural change in AI capabilities that are impacting the labor market, that's not going to be a temporary change. That's potentially going to be a long run change. And the longer that we can track this and if those trends still seem to hold up, that would be indicative of AI potentially impacting work. Now, that said, we don't have an experiment where we can compare a world with AI to one without AI and do a clean comparison.

**** · There's definitely a lot more work to be done here to really tease apart the impact of AI.

Young Workers Lost Their Edge. Where’s the new one?

**** · If you think about what young workers are doing when they're entering the workforce, a lot of it is implementation, doing things that rely on the book knowledge that they learned while they were at school. Whereas the things that they don't have as much experience with an ability to do is relying on the tacet knowledge or the experience that you can only get by doing things on the job. also more social interaction and more strategic thinking. Tacid knowledge I think of as things that rely on a lot of hyper local context or strategic thinking or social interaction or things that you only build via experience on the job. So those are the types of things that are maybe not written down as much in a book. For young workers, it's more directly overlapping with the AI capabilities. And those could be the situations where more experienced workers might have a relative advantage compared to AI and also compared to young workers. When it comes to training young workers, it's totally that firms will want to hire young people if they want to have a middle management or more experienced staff going forward. Now the issue here is even though that they have some incentive to do that so that they have workers in the future, they might not have enough incentive to do that. So they might not hire as much young people as they should from a social perspective and they might not train them as much as they should. And the reason that's the case is because those young people don't have to stay at the company forever.

**** · They can just go leave to another company. So it's true that they will still want to hire some of them, but they might not want to hire as many as would be beneficial to society. And it's just this mismatch between what is the incentive of the individual private company versus what is the incentive of society as a whole. Now the more optimistic take that I could give here is that if AI really is as capable of helping people learn and as a tool for education maybe could speed up the process at which that happens that could also require a lot of changes in the way that we organize our education system potentially universities or even at a lower level than that to help people learn faster and better. There are three things that I think AI is going to be much less capable of doing certainly in the short to medium term. One physical tasks unless we see a big advance in robotics. Number two is strategic thinking and guiding what needs to be done. And number three is social interaction. I think the strategic thinking is increasingly important and it's going to be even more important going forward potentially because it does seem in the future a lot of work might look guiding AI agents to do implementation while you're telling them and guiding them on what needs to be done. And so that strategic thinking, expressing what it is that needs to be done or what I want to be produced, I think that's going to be a pretty key skill and that's the role of what a manager does within a company. So that managerial work and strategic guidance could potentially be a quite important skill going forward. When I think about young workers, how can they develop those sorts of skills? building and using the tools as much as possible and getting used to that mode of work.

**** · The faster that can happen, the better that they might be in terms of adjusting to labor market disruptions or these technological changes.

**** · I do think it's very helpful to compare AI to some of these historical changes.

From Career Ladder to Career Lattice

**** · So for example, the industrial revolution. I think one comparison between AI and that period that was a case where it was the most skilled workers who faced more risk from the industrial revolution. So one case that comes to mind is the lites who were these skilled textile workers and the new inventions that came about during the industrial revolution led a lot of them to lose their work and those were the more skilled workers in society. Something that you might be seeing that's similar here is that it's more of the knowledge workers in more educated roles that might be facing greater AI exposure. So I think that's an interesting comparison. If we think about things electricity or the IT revolution, so over the course of the 20th century, a lot of those were the opposite where it was this middle skill or low-skilled work that tended to be more exposed to that technology. Whereas the most skilled, the highest educated people benefited a lot more from the development of this new technology. So we still have to see going forward, is AI going to look more the first case or the second case? I do think there's something worth bearing in mind here. So one way that AI might be different than past historical episodes is just the rate of capabilities improvement. Even today it's much more capable of doing different tasks than it was 3 years ago and I do think there's this question about as new work gets created there's new demand for existing work etc. Are those going to be done by humans or are the AI capabilities going to advance fast enough that AI is also going to be doing that work? And I think that's the one area where we could think that potentially AI could be different than prior technologies.

**** · There's been a lot of discussion about how we can use AI to augment workers and make them better off as opposed to potentially just substituting them from the workforce and automating all everything that they're doing. Where I was going with this essay is just trying to suggest one concrete solution that I think could potentially augment workers quite a bit. It using AI as a tool for helping people learn. I think an example of a person who's augmenting themselves with AI now. An example of that would be a startup founder with a really lean team that's able to do a lot more tasks because they have access to the AI. All of the different functions that previously they wouldn't have had any idea how to do. Now they can do it themselves because they have access to these AI tools. I think that's a very good example in fact of augmentation.

**** · Whether you're more automated or augmented really depends on what are the tasks that you're focusing on. Are you increasing the scope of tasks that you can do or are your tasks getting shrunk by the introduction of this technology?

**** · The reason that I think that this could be wonderful in terms of augmentation is that when we think about technology that benefits workers, often it is increasing the set of tasks that they're able to do. In contrast, things that automate work that substitute for workers, those are things that take away some of the tasks that workers have to do and now they have to do fewer things. know I think the goal is to try to find ways to augment workers to make them more capable of doing things. And one of the best ways that we know historically for doing that is by educating them. With education, workers are able to do a lot more than they could do before. I do think there's something worth bearing in mind here. So one way that AI might be different than past historical episodes is just the rate of capabilities improvement. And I think we have an opportunity now for one of the biggest changes in learning capabilities that we've had in 100 years if not longer. And that's in using the AI tools for personalized learning. For me, using AI for augmentation, there's a couple branches to that. And something that increasingly I'm using it for is for math. There are areas where I might need to write down a model or prove something. And it's really, really good at that. The way that's augmenting is that it's easier to check if something is correct than it is to necessarily write it from scratch. And so I also potentially view that as a as a significant way of augmenting my work.

**** · Now, on the other hand, things that I don't do with AI, I personally don't really use it for writing. And the reason I don't use it for writing is that writing helps me think and it helps me understand a problem really well when I do it myself. It's not that I don't trust the AI tools to do the writing.

**** · It's more that I would get way less value out of the writing if I didn't do it myself and understood what it was that I was talking about. I think in deciding what we want to delegate and what we want to preserve as human, I think a lot of that depends on what it is that humans want and some of that is about values what is what is wrong. Some of that is also just expressing our preferences. So what do we want to build? What would make us better off? What would make us happier?

**** · What are the things that we would want to use AI for implementation? That's something that we have to express to the AI. I think those sorts of tasks, it's not obvious to me how that's going to be automated, in the short to medium term at least because some of that both it depends on also our reflection. We need to think through what it is that we want. Sometimes we learn about what it is that we want as we reflect on it and as we think more about it. And so that guidance about what it is that we should build, what it is that we should implement, that I view as at least in the short to medium term being more characteristically human than AI. And I view the AI is more on the implementation side.

**** · Imagine that AI really reduces the benefits of learning something new. An interesting thing about this is that's a world where potentially inequality is much lower in the labor market. If the barriers to getting at the top of a field or getting the highest quality output in a given occupation or something, if that barrier becomes much lower because AI can do a lot of the hardest tasks, then that's a world where potentially inequality is lower because the difference between people who know a lot in school and are very capable could be not that different from people who don't try that hard in school. It's interesting because there's this potentially trade-off between inequality and investments in learning. On the other hand, if AI really increases the benefits of this strategic thinking, even these social skills, etc., that could increase the benefits of trying really hard in school because if I can develop the strategic thinking skills, then I'll be really valuable in the labor market.

**** · I would encourage young people, students to use the AI tools as much as they can build with them and really focus on developing that strategic thinking. How do you best make use of these tools? Where are the areas where they're not as good and what are areas in which you as a human can add a lot of value? I think there are some very complicated ways of thinking about how AI might affect thinking going forward.

**** · There are some newer interventions that are happening here education style modes with AI usage to make people focus on the critical thinking skills as opposed to just offloading the task. So there are different platforms. I know Khan Academy for example has one where you can use the AI tool but it's not going to give you the answer. It's going to help you think through how to get to the answer as opposed to just giving it to you off the bat. I do think that we could imagine a world in the future where if we really unlock AI's capabilities for helping people learn, it could be much easier to switch between different professions based on how demand for those jobs is evolving over time. And if some job becomes much more important in the economy, if we can find a way to help people transition faster, that could really unlock a lot of potential. So I'm really hopeful that we end up somewhere closer to a career lattice that works for workers as opposed to a career ladder where there's much more risk about this technological change.

**** · For the first time I struggled to assemble questions that chat GPT would get wrong. I was devastated was thinking how am I going to stay ahead of AI? I think that's the wrong question. My name is Ken Ono. I'm a mathematician and I also work in the space called AI for math. I'm a professor at the University of Virginia on leave and I'm the founding mathematician at Axiom Math. My view on intelligence now has changed quite a bit. We in this world aren't doing the best we can at educating our children.

Next Episode

**** · And I don't say that to be critical of educators. I am an educator. It's always a treat to visit a kindergarten class, a first grade class when it's bring your parent to school day so they can talk about what they do. Oh, I know all the prime numbers or I'm really good at adding that wonder and I want to just bottle up this energy because if we could maintain that wonder in the world and the energy that children have when everything around them is new. Think about where we would be today. The ability and the potential to be someone Romanagen or at least creative in a productive way. I think it resides in us all. Who owns your identity? You do.