By Lynn Wu
Associate Professor of Operations, Information and Decisions
Dan Loney: What sparked your interest in studying how robots are starting to change employment?
Lynn Wu: Honestly, this one has the easiest motivation. At the time, there were so many articles in the press and academia about how robots are going to take over all our jobs and the impending robocalpyse. There were already policy pieces on robot taxes. Bill Gates promoted it. Bill DiBlasio made it a central piece of the presidential campaign back in 2020. And Bernie Sanders had also proposed some kind of a robot tax.
I think it’s really important that we understand what’s going on and have some real, concrete evidence at the firm level to see whether firms actually do lay off people en masse after robot adoption. There were only industry and country-level evidence at the time, and it is really important to study at the firm level because countries and industries do not adopt robots. Whatever the positive/negative effects that you find depends on whether firms that adopt actually laid off people, or is it coming from firms that do not adopt? These facts are very difficult to observe when we are looking at a macro level, at the industry and country level.
Loney: Why do you think these narratives pop up? They have really taken hold with some people over the last few years.
Wu: I think it’s not just the last few years. I’ve seen this trend going on every time we see a potentially transforming technology that could affect work. We see that back in the day of the Industrial Revolution with the Luddite movement. Those people literally burned the looms that automated the process of making garments. It turns out that having these looms, which effectively accelerate the process of making garments, did not make these people lose their jobs. We see an increase in employment for people who can effectively use new automated, mechanical tools like looms. You see the same thing with Excel. It was going to replace accountants. It never happened.
Every time we see a technology that could potentially change the way we work, change our lives, there’s a human visceral reaction to it. We tend to overestimate what the technology can do and think, “Oh my gosh. I’m going to lose my job now.” It’s really important to take it back and think about, “What exactly is that technology doing?” before we make any strong decisions, especially the policymakers and at the policy angle. What are we going to do about this? What are workers going to do about it? What are firms going to do about it?
Who Will Be Impacted by Robot Workers?
Loney: Tell us a little bit about the research that you’re doing in this area to try to get a better grasp on what’s been taking place here.
Wu: Our work is first to study robot adoption on employment and its effect at a firm level. I want to emphasize at the firm level, because only at a firm you can see what happens when firms adopt robots. Do they actually lay off people, or do they hire more people? And what happens to the firm that did not adopt? These kinds of effects can only be observed at the more micro level.
We used the data from our sister school in Canada, which has comprehensive data on robot import and export, which has a very good measure of robot adoption [at firms]. We also have very comprehensive data about their financial performance from their tax filings and surveys that the Canadian government mandated on various firm practices.
What we found is exactly the opposite of what people were expecting. Robots did not replace human workers. In fact, the robot adopters, or the firms that adopted robots, hired more people than it did before. So, how do we reconcile the evidence that we see sometimes at an industry level and the country level that there is a negative effect of robots on employment? It turns out it’s not the robot-adopting firms that are hurting employment. It is the firms that did not adopt robots that are losing the competition. They’re not as competitive as before, and they had to lay off people because they’re losing the market share.
It’s a very different story than the popular press thinking that we’ve got to tax robots to preserve human work. Taxing the firm that got the robots turned out to be the exactly wrong thing to do in this case. That’s one major finding. We’re basically saying that we have to look at these phenomena in greater detail to understand what’s going on here. Without this kind of firm-level measurement and technology measurement, we wouldn’t be able to know this important distinction.
We also have found other effects on employment. It’s not the number that matters. We always think that robots are taking over our jobs, and that’s not the case. We see skill effects. Specifically, robot-adopting firms hired more high-skill workers, many more low-skill workers, at the expense of middle-skilled workers. I define high-skill workers as those with college education. Low-skill workers are the people who barely finished high school. And middle-skill workers are people with high school degrees or associate degrees, who had some kind of more advanced work-related trainings. It’s these middle-skill workers who are being decimated by robots, and that is a big problem.
We also show that managerial, supervisory work has also been decimated by robots. If you look at the average number, it looks great. Employment has gone up. But hollowing out the middle-skill work, hollowing out the supervisory work is a big problem because now the career ladder is broken. How do we incentivize, how do we train, where did the middle-skill work go? You can’t expect all people to get college degrees and become programmers or robot technicians or producers.
Loney: Does the adoption of robots change the dynamics of the work being done by those companies or their success?
Wu: Absolutely. Just adopting the robot itself is not going to be enough. You have to be able to learn how to manage robots in a way that accelerates performance, increases your worker productivity. Let me give you a real-life example from a repair facility at a U.S. electronics firm. They actually experienced a dramatic improvement in their ability to observe productivity after robots were implemented. At this repair facility, they fix electronics. And because robots don’t get physically tired when performing this kind of repetitive task, they can do this job more consistently than humans. As a result, the variance in production has gone down.
This allowed managers to clearly observe individual employees’ behaviors. They were able to see that many human employees were following irregular patterns of being very productive in the morning, compared to the afternoon. But then they do more repairs at later hours, as they were cramming their work at the end of the day.
Interestingly, after robots are implemented, they are able to track the individual employee’s productivity more easily for two reasons. First, the type of errors robots make are very different from that of humans. Because of the differentiating errors, it makes it easier for us to figure out which one is which. Robots are more likely to make consistent errors compared to humans, again making human errors easier to identify.
Robots also provide precise data about their own performance, which made it easier to isolate both the positive and negative side of performance changes caused by human behaviors. This data-generating capability allows managers to monitor their productivity much better than before and detect weaknesses in the production processes.
So, it’s not just adopting the robot itself. It’s all the other things they’ve done to detect and figure out the production process. In this case, the manager wasn’t even aware of this cramming behavior described earlier until after they adopted robots to observe these processes. And as a result, they changed a lot of their work processes along the way to further reduce the errors once humans and robots were working together. Overall error rates in the facility have dramatically improved. This is an example of how robots can be used to improve productivity and how to manage the workforce appropriately to capture, to further increase the effect of the robots.
The Effects of Robots and Generative AI on the Workforce
Loney: How are careers going to be impacted by the further adoption of robots? I’ll play that off of the comment you made earlier about the impact being significant on middle managers. That’s an important stepping stone in a person’s professional career. If we have fewer of those, the structure of leadership changes.
Wu: That’s a really important point, and it’s a really hard problem to solve. In my research I mentioned earlier, because you have many more lower-skill workers, many more higher-skill workers at the expense of middle-skill work, the type of manager you need is going to be very different. These managers need to understand how a robot works to be able to test these hypotheses. Just like the example I gave about repair facilities. They need to fundamentally change the way they work, fundamentally change the way they monitor and reward and hire employees.
There are two effects. No. 1, we simply need fewer managers and supervisors than before. Because you can manage many standardized workers and lower-skilled workers at the same time, as opposed to higher-skill or middle-skill workers. Furthermore, the type of management skills you need is going to be different. So, it’s a big problem. Now, we have no middle-skill work, or less of them, and much fewer supervisory work. Where do people go? The entry-level work is supposed to be a stepping stone to move up in the career hierarchy, and now you can’t move up anymore. You’ve got middle-skill, you’re gone. You go to the supervisors, and you are gone. Then it’s very hard to move up to high-skill work, right? That requires extensive training. This is a very big challenge for managers, where you have to think about how to build a new career ladder now. The existing one may not work.
That’s why you see lots of unionization going on in the workforce, from Amazon warehouses to Starbucks, everywhere. It’s because you can’t use a career ladder as a motivating force for people to work at an entry-level job at lower pay in exchange for future career advancements. How do you build that back? If firms do not build it back, then we’re going to see unions becoming more of a mainstay in our society again.
Loney: If we’re expecting that we’re going to see more companies adding robotics into their operational structure, does that answer the question about whether or not companies can even avoid having robots in the first place?
Wu: The moment one firm adopts, they become more effective and more competitive. That means everyone else, in order to stay competitive in the marketplace, has to adopt these new technologies. Even the biggest, very profitable, very innovative firms have to catch up. We see in Google’s case, when Microsoft released ChatGPT, Google is scrambling to do the same thing, incorporating every aspect of that technology in their products. That is something that firms just cannot avoid. The Luddites burning the loom is not going to work in this case. Probably never would work.
Loney: How will the advancement of ChatGPT play a role in the corporate structure as we move forward?
Wu: I think the robot thing is the tip of the iceberg. ChatGPT, the large language model, is going to accelerate that process tremendously. Because if you think about what ChatGPT and these large language models are targeting, it’s exactly that middle-skill work. And because these technologies are really good, when you are an expert in that field already, it accelerates your work. But who was doing that work for you before, if you were a senior person? It’s people below you. Now, you can use the ChatGPT to do a lot of it for you. So, it’s precisely that middle-skill work that is being targeted. And that’s exactly the same effect of the robots. There was a recent paper by OpenAI and also our colleagues at Wharton that showed that the skill set that’s being targeted by ChatGPT is middle-skill programming jobs and writers.
Loney: Does the longer-term outlook for middle-skill jobs look very bleak at this point?
Wu: I think existing middle-skill work is in trouble. But new middle-skill work will be created. For example, there’s prompt engineering, something you’ve never heard of until maybe a few months ago. These engineers are literally trying to make ChatGPT do what it’s supposed to do. We’ll need robot technicians to fix the robots, process engineering to observe processes, and see where robots can be used in the production processes. All these things are probably going to be new tasks. Over time, these new tasks will evolve into new career opportunities, just like 20 years ago when there was no social media manager, right? That’s a new job that was created as a result of the technologies.
But the important problem is not necessarily the new jobs that will be created. I guarantee you that new jobs and new tasks will be created. It’s the speed at which we can retrain the existing workforce to leverage that. The last time we had this kind of dramatic technology change was probably the Industrial Revolution, the steam engine being replaced by electricity. That took 30, 40 years to complete. And that means existing managers retire, the existing workforce retires. This time, we are not going to have a 30, 40-year horizon. We’re going to have a five, 10-year horizon. How you retrain that existing workforce is going to be a huge challenge for everyone.