Apollo经济学家：AI冲击工资增长而非就业

Westin: This is a story of puts and takes. Until recently, most of the talk about the effect of AI on employees has been based on theory. But now we are starting to get some data. Torsten Slok is chief economist for Apollo and recently co-authored a paper analyzing those data for hundreds of occupations. And though it is early going, the initial results indicate that those most affected feel it in the size of their paycheck rather than whether they get to keep their job. -It is now possible to look at what happened before ChatGPT came along and what happened after ChatGPT came along.

And what we did was that we categorized the employment and wage numbers into occupations and asked what occupations were exposed a lot to AI and what occupations are exposed less to AI. And then we looked at as a natural experiment, what happened before ChatGPT came along and what happened after ChatGPT came along. So when you look at the 300 different occupations that we looked at in terms of AI exposure, the conclusion is, at least where we sit right now, that those occupations that are highly exposed to AI, they are seeing weaker wage growth, but the employment effect on everyone is insignificant.

Westin: What surprised you about the results that you might not have expected going in? -Well, the debate around AI in the labor market for now some time has been about labor displacement. In other words, there's a lot of worries about that when AI comes along, then work process is going to get automated and therefore a lot of workers are going to lose their jobs. What we found was that that effect is not quite here yet.

Instead, let's also not forget that when AI, of course, becomes more easily available as a tool, now it's also easier to open a new business. So if you look at the weekly data from the census for business formation, you are at the moment seeing business creation in the U.S. is at the highest level ever in U.S. history. So on the scale, you have on the one hand an automation and replacement effect that certainly saying that the labor market should get worse.

But on the other hand, so far, the dominating effect has been that there is also a much more dynamic economy where people can now invent ideas, use agents, use loops, graphs to come up with ideas and as a result, create more businesses that ultimately likely will also create some more employment. Westin: And talk about the labor market getting worse. As I understand your study, it said it gets worse more on the wage growth than it does actually in people losing jobs.

But it's not evenly distributed. -Yeah, there are some occupations that are more negatively impacted. We chose and this was somewhat randomly to categorize the pockets of occupations into those occupations that are more than half exposed to AI and those that are less than half exposed to AI. In other words, try to cut the sample into different categories of who is it that has high exposure, who is it that has low exposure, again, by this actual users data.

And that does show you that those who have higher exposure generally saw lower wage growth. So in other words, you can begin to worry about that maybe AI, because it is replacing knowledge workers, is going to create some downward pressure on wage growth. So far, not on employment, but it's going to create some downward pressure on wage growth. Westin: You say in your report that it's still early days, that the effects may grow over time.

Do you have any sense where we are in that process and how those effects may grow? -We don't quite know at this point because the market is trying to figure out the answers to that question literally every day, namely, how quick is the AI payoff going to come? At the moment, enterprises are investing a lot of money in AI. But when you look at the actual margins for the S&P 493, meaning not the Magnificent Seven, margins have not gone up yet.

Most people, including me, expect that we should begin to see some improvements in margins over the next several quarters. But at this point, it is still very early days where businesses are trying like an S-curve to figure out, well, now we have a new technology. Now we're trying to figure out how to use that technology. And once we then have a way to use that technology, we will likely see another step higher in the S.

And the implication, of course, is that those industries that will be able to implement and adopt AI will, of course, be the next winners as this technology continues to develop. Westin: Based on what you've seen so far, including this study, what does it say for economic policy? What should people in Washington be focused on right now? -Well, the challenge in policymaking at the moment is that we just don't know which scenario we're looking at.

Let's say the unemployment rate goes to like 10, 15% because people lose their jobs. Then, of course, economic policy should be focusing on that problem. If, on the other hand, that the dominating effect here becomes that business creation is creating so many more jobs and therefore the unemployment rate, which is the consensus expectation, would actually be going down. Well, in that case, policymaking does not need to worry about displacement of workers.

Then they should instead maybe worry about the economy overheating. And another set of economic tools from policymaking will be needed. So that's why the problem where we sit right now here in 2026 is we just don't know whether the scenario at this point is one where the labor market is going to get a lot weaker or it's going to get a lot hotter. And for that reason, therefore, the best thing for policymakers at this point is just literally like the market to wait and see.

Westin: Are you as optimistic about AI at the end of this study as you were at the beginning? -I'm very optimistic about AI because I do think that the effect of creating a more dynamic economy, many more opening up new businesses in consulting, in finance, in legal services, and those new businesses will compete with the incumbents. And that will continue to put more and more competitive pressure on the economy. And that's something that we should all be very interested in because that creates more jobs, that creates more businesses.

And ultimately, as a result of that, I truly believe that AI is a miracle drug that will both create higher productivity and also create higher employment. So I'm very optimistic on what AI will bring to the U.S. and the global economy. Westin: Whether AI eventually spells riches or ruin for workers, the numbers so far tell only part of the story. Inside companies, executives are already deciding which jobs can be automated, which workers can be retrained, and how best to allocate labor costs.

Diane Gherson is a senior advisor at Boston Consulting Group, and she confronted those choices when she was head of human resources at IBM. -For certain jobs where there's high attrition, those are the jobs that are being automated, the first, with AI. Those jobs are being replaced at a lower rate. The demand for those jobs is lower, and it's your classic demand supply. So people are taking jobs at lower rates in those job categories, customer service being an example or business services.

Those are the types of jobs where you need fewer of them because the work can be done by fewer people faster. So to the extent that they are being hired, yes, they may be hired at a lower rate, but I don't see companies cutting pay. The only area that I would see this happening is in areas where you have contractors, right? So what you're seeing is there's a lot of data available, there are data brokers that can tell hospitals that are hiring traveling nurses which of the traveling nurses have low credit scores or credit problems.

And so they can offer a lower wage and know they'll get it. Or ride hailing services, if there's a driver who previously took a ride at a lower rate, they'll keep being offered that lower rat