Don't get fancy with your labor market fixes for AI

AI could be unpredictable. Our labor market response to it shouldn't be.

Sam Altman, the certifiably UBI-curious CEO of OpenAI, getting fancy at Windsor Castle. (Photo by Andrew Caballero-Reynolds/Getty Images)

The only thing emerging faster than the AI arms race in America is the debate over what to do about the workers who AI will displace.

Many of the loudest voices are also pitching some of the most ambitious ideas. OpenAI CEO Sam Altman is interested in implementing a universal basic income (UBI) — an experimental idea that would require not only a gigantic budget but also massive new infrastructure to develop and implement. UBI sounds like a simple solution, but any new program — particularly one aimed at universality — would need to make decisions about eligibility, implementation, and oversight that are complex and politically contentious.

While politicians and tech optimists are out there pitching programs that are tough to build consensus around — or overly specific to the point of being unworkable — we already have a boring, scalable solution to the problem of mass job loss.

It’s called unemployment insurance (UI).

The best economic policy is flexible, automatic, and responsive to a range of shocks. UI already exists in all 50 states and has a strong track record of responding to technological disruption in the labor market.

We don’t yet have a clear sense of exactly how big AI’s labor-market disruptions will be — or who would be most immediately affected. By bolstering a long-running program that already serves as a social safety net for the unemployed, the economy can endure whatever shocks AI might deliver.

Why build UBI when we already have UI?

So much is still uncertain about AI

We absolutely shouldn’t dismiss the pain of possible technological change. One paper from economic researchers James Feigenbaum and Daniel Gross showed that while the automation of telephone operator jobs didn’t affect future cohorts — who were able to enter new professions — the people who were in those jobs when they were automated did suffer.

We also don’t know who will be impacted by AI. So far, my research with Molly Kinder from The Brookings Institution has suggested no mass disruption to the labor market is currently happening due to AI. This is also backed up by research from the Economic Innovation Group.

And even some early research about disruptions is looking less than clear-cut than it did at first. For instance, work from researchers at Harvard and Stanford both found concentrated declines in employment among early-career workers who are more exposed to AI — but those declines started immediately after the introduction of ChatGPT, which suggests those labor market changes are not really about advanced AI. Companies were certainly not clear then about how to use generative AI for commercial uses, and those jobs may simply be affected by macroeconomic factors.

Early-career workers might be the most resilient — finding it easier to adjust their occupations to a dynamic economy than later-career workers. No matter what happens with AI these workers will need a boost in some way.

There are absolutely people who have lost their jobs in response to technological innovation in recent years. But the labor market has been able to absorb those people and redirect them into new occupations.

As an example, economists Daron Acemoglu and Pascual Restrepo estimated that from 1990 to 2007, the introduction of industrial robots did lower employment. But there was no mass unemployment over this time period. Technology did displace workers, but on average, people in the economy were able to find different jobs.

This is not to minimize the pain of that job loss and transition for those workers — losing a job is terrible and the new one you find may not be as satisfying as the old one.

Some are predicting, however, that AI is a massive disruption on the same scale as the Industrial Revolution, which did seem to create a huge amount of economic pain for certain workers. It seems clear from the records at the time that the average height and weight of the population fell, a strong indicator of falling living standards.

But the disruptions didn’t cause us to ban factories, and here we are, two centuries later: richer and more fully employed than ever. (I feel sure I am much happier as an economist pontificating on the internet than I would have been as a weaver.)

We don’t know what future we’re headed toward — minimal impact, impacts for specific workers, or widespread disruption. After all, Geoffrey Hinton suggested a decade ago that no one should be trained to be a radiologist anymore because AI would replace them. But as an article by Deena Mousa showed, radiology jobs are more popular than ever.

Imagine if we had decided to legislate wage or training subsidies specifically for radiologists and then it turned out we were just funneling resources to a specific group for no reason. We don’t want to decide that early-career workers need help and then find out it was the mid-career workers we should have been worried about.

Let’s turn to our tried-and-true policy solutions

Rather than designing AI-specific policy responses, we should be optimizing our response to economic shock — from AI or anything else — and making sure that it can handle a range of outcomes.

The U.S. needs the type of tool that can handle the Industrial Revolution, the dot-com bust, and a financial crisis.

We need to make sure automatic stabilizers like unemployment insurance are ready to go for the workers who need them, whoever they are.

We don’t have clarity on what jobs could be impacted by AI, which jobs might go away, which ones might become better compensated, and which ones might emerge.

We don’t need Trade Adjustment Assistance for coders. We need a safety net that can respond to a range of scenarios. A good and enduring social safety net should be flexible enough to support the workers who need it without creating preferences for others.

UI is already there for workers who lose their jobs to help them during the transitional period. But we do need to fix the unemployment insurance system. It needs to be shored up. We saw how it sagged under the weight of a massive shock like COVID-19, and still it leaves gaps that need to be addressed because not everyone qualifies for it or can easily access it.

Just because we’re not still seeing headlines about people waiting months for benefits doesn’t mean that the system is working well now. It just means fewer people are relying on it, so it’s not a national story.

If you think mass (or even elevated) displacement is coming because of AI, the first backstop is the UI system. UI is there for workers across industries, no matter the scale of the disruption, and it is a proven strategy.

It still needs fixing and modernizing. We just have to make sure the systems are actually ready to work.

More on AI: