The most recent employment figures paint a rather bleak picture of the labor market and the apparent havoc that AI is wreaking on it. Following warnings about unemployment among recent graduates earlier this year, the latest report suggests that the impact of AI is reaching a broader group of workers. More than 150,000 layoffs occurred in October, making it the worst October in terms of layoffs in more than two decades, and around 50,000 of them have been attributed to AI. Overall, there have been more job cuts in 2025 than in any year since 2020.
It’s too early to say to what extent AI is really to blame for these job losses, even if companies blame it in public statements. A team of researchers from the Yale Budget Lab and Brookings has argued that the overall labor market is not being disrupted more by AI than by the Internet or PCs, and that recent college graduates are being displaced due to sector-specific factors. However, Anthropic CEO Dario Amodei has predicted that AI could eliminate half of entry-level management jobs. So which one is it?
There’s a lot we don’t know about what will happen with AI in general (looking at you, AI bubble), and it’s too early to say whether AI will actually deliver on its most ambitious promises or be more transformative than past technological revolutions.
But to shed some light on the employment question in particular, I called Neil Thompson, senior research scientist at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). He’s been studying everything from why diminishing returns in frontier models will shape the future of AI to how automation changes the value of work. Our conversation has been edited for length and clarity.
In recent years, his work has pushed back about the idea that automation is always bad for workers and that AI will kill all our jobs. But in recent months we have seen tens of thousands of job losses attributed to AI. What is happening?
My guess is that we have two different phenomena happening at the same time. One is that AI is becoming more prevalent in the economy. I think in some cases, like customer service, it’s probably pretty legit. In fact, these systems seem tremendously good at those tasks, so there will be some jobs that will be taken over by these systems.
At the same time, I would be surprised if these systems could do as many things as the job loss numbers imply. So I suspect there’s also a combination of people deciding to cut jobs and put some of the blame on AI, or they’re cutting jobs in advance with the goal of doing more AI. They’re driving their businesses toward that and seeing what’s going to happen.
Why is there so much dissonance among those who say AI will take away half of our jobs and those who say AI is not the reason Are we seeing so much turmoil in the labor market?
Many people talk about an incredibly rapid change: an increase in capabilities that could do things that humans can do. For most companies, adopting these systems comes with very high last mile costs. Someone using ChatGPT only on the frontend is very different than “now we run our business and trust that every time the system runs, it will run well.” That’s a different level. It is often necessary to provide specific data. There are a lot of costs that come with that. So these last mile costs can be very significant and can really slow down adoption even when the systems are quite good.
Aside from that cost, there is also the question of whether a system is good and whether a system is good enough to be better than a human being. They are not exactly the same.
Earlier this year you published a paper with his MIT colleague David Autor, who used experience as a framework to understand how automation affects the value of work. Historically, it’s not all bad, right?
When we think about automation, we have in mind a kind of doom scenario, where, as automation occurs, the number of jobs that exist in that occupation decreases, the wages in that occupation decrease, and you think, “wow, this has been a pretty terrible story.”
But if you look at the last 40 years of automation (this is not AI automation, it’s just computerization and things like that), we know that many routine tasks were automated through this process. If we look at the people who had routine tasks, what we found is that many of those things were automated, but their salaries did not go down either. Some went up, others went down. That’s kind of a puzzle.
What we think is happening is that, when automation occurs in a particular occupation, it really matters which of the tasks in that occupation are automated. In particular, if you have automation of highly specialized tasks (i.e. the most skilled things you do) that has an effect, and if you have automation on less skilled tasks, you will get a different effect.
Can you give me a couple of examples?
Let’s think about taxi drivers. The most expert thing you did was know all the roads in a city. You knew all the little back roads. You knew all the little shortcuts. You were the expert on that. Then Google Maps and MapQuest come along and suddenly anyone who knows how to drive a car can do it pretty well. In that case, your most expert tasks became automated. Because the most skilled things disappear, their salaries go down.
But, contrary to this version of the doom cycle, wages go down, but the number of people in that profession goes up, because now, a lot of people who didn’t know all the streets before, can suddenly drive an Uber.
At the other extreme, let’s think about proofreaders. Enter the spell checker. A lot of things they used to do are now automated, but it was the least expert thing they did. The significant thing they did was reorganize your paragraphs and make sure you were thinking about the right things and wording things the right way, not the spelling part.
So if you look at what happens to them, their less skilled tasks became automated. What remained was more expert. And so, because they used their expert material most of the time, their salaries have actually increased faster than average, but now there are fewer of them.
So you have this interesting effect where Uber driver salaries went down, but there was more. And for proofreaders, salaries increased and were less. And both have advantages and disadvantages.
So clearly, AI is not the first technology to automate aspects of work in the computer age. But is the same framework of experience valid further back in history? Would we see similar patterns in the Industrial Revolution and in the automation of the work of textile workers?
One of the examples my co-author likes to talk about is skilled craftsmen. Think of the carters, the blacksmith and all those people; They used to be incredibly skilled jobs. And through industrialization, we figured out how to do it on production lines and other places where the average experience was less, but there were a lot more wheels being produced and there were a lot more people involved in wheel production.
And then of course we have a lot of modern examples as automation comes, and some of the things that do get automated, we actually become more expert at the things that we do because we don’t have to do the basic things anymore.
Companies like Google and OpenAI promise that their technology will do much more than automate basic tasks, and they are spending hundreds of billions of dollars on infrastructure to achieve it (call it artificial general intelligence or superintelligence). we are listening a lot There’s been talk of an AI bubble lately, because it’s unclear whether these tools will actually work before the bill comes due. How will we know when AI has proven its effectiveness?
I don’t think the question is really whether AI is going to prove itself. I think it’s clear that these capabilities are improving quite rapidly. I think it will be incredibly useful and I think there will be a lot of adoption. There will be many benefits that will come from it.
For me, the issue in terms of the AI bubble has more to do with valuations. This will be helpful, but is that the correct assessment? It’s going to matter a lot. It’s going to have many of these effects. The question is: are we moving even faster than those effects will occur, or just the opposite?
TO recent Pew Research Center survey showed that Americans are more worried than excited about technology. Why is AI so unpopular?
I want to hesitate to get too much into people’s heads, but I think it’s understandable that people have anxiety about what AI is going to do and how it’s going to change their jobs, because it’s such a powerful tool. I think it will change many people’s jobs, including yours and mine.
I think it’s particularly difficult to face that and not know how much of the work is going to be replaced or how much I’m going to have to adjust in ways that could be painful. I think we’ll learn more about that in the coming time.
There is a second piece that is very, very difficult. Historically, when new technologies appeared and things were automated, humans began to perform new tasks. New tasks are created that did not exist before but are actually important for employment. We don’t really know in advance what those new tasks will be. That lack of visibility is a challenge. But it is worth saying that, historically, a remarkable source of new tasks and new jobs has emerged. And so, I think we should be sure that there will be many of those to come.
There will be a transition. In many cases, we should think of this as similar to previous transformations. The question is how quickly it happens. If it’s medium or long term, we humans are pretty good at saying, “Okay, if these are new tasks that we’re particularly good at and the technology isn’t, let’s adapt to do those tasks.” But if it happens all at once, and many of the transitions and shifts occur in a compressed time frame, that will make it much harder for the economy to adapt.
It sounds like you are saying that there is a fear of the unknown and that there are many unknowns right now. But we have gone through important technological transformations before this one. We just don’t know how long it will take or what we will do on the other side. That doesn’t sound very comforting.
Let me put a little spin on that. Definitely, if we look historically, we have seen patterns where new technologies come into play. There is some turmoil in the economy, some people are hurt by that, and we need to be aware of that. We should hope that that can happen now too. But in the medium term we adapt well.
In terms of AI, I think we can take solace in those historical lessons. And the question is fair: Is AI in any way different from these previous technologies that would make us think we would get a different result?
I think people who think we’re going to get to AGI quickly, their answer would be yes. If you can do everything we can do, and you can do it next year or the year after that, that’s very different from previous technologies. That makes it quite difficult to adapt. If it’s implemented, it does some tasks, it takes a long time to do other tasks, well, then I think we’re much more in a world where we can adapt like we have in the past.
A version of this story was also published in the User Friendly newsletter. Register here so you don’t miss the next one!