Why Automation Creates New Jobs Instead of Just Destroying Them

The Robot That Hired a Human

In 2019, the economists Daron Acemoglu and Pascual Restrepo published a paper that should have made every headline about automation a little more honest. Their argument was simple, and it cut against the grain of both techno-optimism and techno-panic. They said that automation does not just destroy jobs. It also creates them. But here is the twist: the creation does not happen automatically. It depends on a specific kind of innovation that most companies, and most economies, have neglected for decades.

The paper, published in The Journal of Economic Perspectives, is titled "Automation and New Tasks: How Technology Displaces and Reinstates Labor." It has been cited over 2,000 times. And it offers the most useful framework I have seen for understanding why your grandfather could walk into a factory in 1965 and leave with a career, while your cousin with a bachelor's degree in 2025 might be competing with an algorithm for a gig.

The key is not how many jobs technology kills. It is what technology does next.

The Two Forces That Actually Matter

Acemoglu and Restrepo argue that the labor market is shaped by two opposing forces. They call the first one the displacement effect. This is the part everyone talks about. A robot welds a car frame. A software program processes insurance claims. A self checkout kiosk replaces a cashier. Labor is pushed out of a task. The worker loses that job.

But there is a second force, and it is the one that usually gets ignored. They call it the reinstatement effect. This happens when technology creates entirely new tasks that only humans can perform, at least for now. A new task might be designing the robot that welds the car. It might be maintaining the software that processes the claims. It might be managing the logistics network that the self checkout kiosk feeds into.

The point is not that every displaced worker gets a new job doing something adjacent. The point is that the economy as a whole can generate new kinds of work that did not exist before. The question is whether it does so fast enough to offset the jobs that disappear.

Acemoglu and Restrepo (2019) show that the balance between these two forces determines whether automation helps or hurts labor. If displacement outpaces reinstatement, workers lose. If reinstatement keeps up, the labor market can actually expand.

How the United States Broke the Balance

The authors did not just build a theory. They tested it against real data from the U.S. economy. They looked at industry level data over several decades and measured how the task content of production changed. They wanted to see whether automation was pushing labor out of tasks faster than new tasks were pulling it back in.

What they found was not subtle. Over the last three decades, the displacement effect accelerated, especially in manufacturing. Factories automated more aggressively. The tasks that humans used to do were handed over to machines at a faster rate. At the same time, the reinstatement effect weakened. The economy stopped generating new tasks at the pace it once did.

The result was a net loss of labor demand. The authors found that the slower growth of employment over this period was accounted for by an acceleration in the displacement effect, a weaker reinstatement effect, and slower productivity growth compared to previous decades (Acemoglu & Restrepo, 2019).

This is the part that gets lost in the public debate. People argue about whether robots steal jobs. But the real problem is not that robots exist. It is that we stopped inventing new things for humans to do.

Why New Tasks Matter More Than You Think

The reinstatement effect is not a small thing. It is not just about a few software engineers writing code for the robot that replaced the welder. It is about entire categories of work that did not exist fifty years ago.

Think about the rise of the internet. In 1990, there were no web developers, no SEO specialists, no social media managers, no cloud architects, no data privacy officers. Those jobs did not exist. They were created by technological change. The internet automated some tasks, like looking up information in a library, but it also reinstated labor into new tasks, like building and maintaining the infrastructure that made the information accessible.

The same happened during the Industrial Revolution. Machines replaced weavers, but they also created jobs for machinists, engineers, and factory managers. The reinstatement effect was strong enough that overall employment rose, even as specific jobs vanished.

Acemoglu and Restrepo formalize this. They show that when new tasks are introduced, they shift the task content of production in favor of labor. These new tasks always raise the labor share and labor demand (Acemoglu & Restrepo, 2019). That is a strong claim, but it is grounded in their model. New tasks require human judgment, creativity, and adaptability. Machines are not good at those things, at least not yet.

What the Data Actually Says

The authors did not just rely on theory. They built an empirical decomposition to measure how much of the change in employment was due to automation versus new tasks. They used industry level data from the U.S. Bureau of Economic Analysis and the Bureau of Labor Statistics. They tracked changes in the task content of production across industries over time.

Their method was clever. They looked at the share of output that went to labor versus capital. If automation was displacing labor, the labor share would fall. If new tasks were reinstating labor, the labor share would rise. Then they compared these shifts to actual employment growth.

What they found was that the decline in the labor share over the last three decades was driven primarily by automation, especially in manufacturing. The reinstatement effect, while still present, was not strong enough to compensate. The authors note that the slower growth of employment over this period is accounted for by an acceleration in the displacement effect, especially in manufacturing, a weaker reinstatement effect, and slower growth of productivity than in previous decades (Acemoglu & Restrepo, 2019).

This is not a prediction about the future. It is a diagnosis of the recent past. The economy has been automating faster than it has been creating new tasks. That is why wage growth has been sluggish and why the labor share of income has fallen.

What the Research Does Not Prove

This is where I have to be careful. The paper does not say that automation is always bad for workers. It says the opposite. Automation can raise productivity and create new opportunities. The problem is not automation itself. It is the imbalance between displacement and reinstatement.

The paper also does not prove that the reinstatement effect will always be weaker than the displacement effect. That is a historical observation about the last three decades, not a law of economics. It is possible that the balance could shift back. New technologies like artificial intelligence could create entirely new categories of work that we cannot yet imagine. But the authors are not making that prediction. They are describing what has happened, and the data is sobering.

The paper does not address policy directly, but the implications are clear. If the problem is a weak reinstatement effect, then the solution is not to slow down automation. It is to accelerate the creation of new tasks. That might mean investing in education, research, and infrastructure. It might mean changing how companies are incentivized to innovate. The authors do not prescribe policies, but their framework makes the diagnosis.

What This Actually Means

• ▸If you are a policymaker, stop focusing on whether robots replace people. Start asking whether the economy is generating new tasks fast enough. That is the metric that matters. Acemoglu and Restrepo show that the reinstatement effect is the key to a healthy labor market. If it is weak, no amount of retraining will save workers.

• ▸If you are a business leader, do not assume that automation always cuts costs. It might, but it also destroys the human capital that your company depends on. The authors find that automation reduces the labor share in value added (Acemoglu & Restrepo, 2019). That means workers capture less of the value they create. Over time, that erodes consumer demand and social stability. It is bad for business too.

• ▸If you are a worker, do not panic about every headline that says robots are coming for your job. But do pay attention to the direction of innovation in your industry. Are companies investing in new tasks that require human skills, or are they just automating the old ones? The answer will tell you more about your future than any prediction about the number of jobs lost.

• ▸If you are an educator, teach students to do things that machines cannot. That does not just mean coding. It means creativity, problem solving, and the ability to define new tasks. The reinstatement effect depends on humans inventing work that did not exist before. That is a skill in itself.

• ▸If you are a journalist, stop writing stories about robots taking jobs. Write stories about why we stopped inventing new ones. That is the real story. Acemoglu and Restrepo gave us the framework. It is time to use it.