We need a coordinated research program for AI job loss
Preparing for AI disruption will certainly require more research in labor markets
There’s a lot of speculation about what AI-induced job loss will look like in the future. Will AI reduce demand for entry-level work? Will AI reduce the white-collar wage premium? Will AI cause entire industries to simply… vanish?
These are all valid questions, and there are many AI experts and economists working on them. But as an economist who studies labor markets and unemployment programs, I think there is also more research we should be conducting today to make sure we’re ready for a variety of potential job loss scenarios.
As I wrote last month, the existing evidence shows that public sector reemployment programs, which are the exact programs that would need to be scaled up to help solve a potential AI job apocalypse, are not particularly impactful. How do we help policymakers design new programs that could be more effective?
The answer to this problem and many others quite like it could be found if we dedicated more resources to understanding disruption in the labor market. I believe this work would be best accomplished through a coordinated research program that operates inside the Department of Labor.
I’m as interested in the speculative essays and podcasts about AI job loss as everyone else. But it’s time to think critically about how to create more government-backed institutions that are designed to actually solve these future problems.
The questions we need to answer
Before I explain the proposal for this coordinated federal research program, let’s look at some high-level labor market questions that a research institution like this could help answer.
How is AI impacting the labor market? So far, AI seems to have had a limited effect on the labor market. But it’s unclear whether this will always be the case, or if AI is having an effect that is hard to detect. Most of the estimates of AI effects rely on a measure of “AI exposure” by occupation, but we could simply not have an accurate understanding of the occupations most exposed to AI.
Why has US employment fallen behind other nations? Two decades ago, the US labor market was the envy of other large countries, exceeding the employment rates of most of our peers. Today, that situation has reversed, and the US lags behind peer countries. That means a future of AI-induced unemployment could hit the US especially hard. Why is this happening? How can we fix it? These are important questions to answer, even outside the context of artificial intelligence.
Why is unemployment duration increasing? Even though the unemployment rate is low and stable, unemployment duration—how long an individual is unemployed for—keeps rising. The average unemployment duration is currently 25.5 weeks, and it had never exceeded 21 weeks before 2007. This is an increasing problem for both the unemployed and potential employers, and I have not seen a satisfying answer for why.
Each of these challenges is important—but we do not have the proper institutions focused on solving them. Instead, they are dealt with on a piecemeal basis by individual academic economists, think tank staff, and journalists. Each of these players uses data from government surveys or administrative records, but there’s no pipeline that directly leads from problem to diagnosis to solution.
A coordinated research program for AI job loss
Imagine if the government looked at these issues and created new institutions dedicated to addressing them. This would need to extend beyond academics examining a problem and actually empowering the builders and tinkerers to solve them.
For example, right now the federal government simply observes the increase in length of unemployment. Certain academics and labor market analysts propose theories for why this is happening, but the problem continues to go unsolved. A coordinated federal government research program could both observe this problem and fund and restructure reemployment programs to be more impactful.
We already have examples of this working on a smaller scale. The Department of Labor funds states to run “Reemployment Services and Eligibility Assessment” (RESEA) programs. This is a program modeled after the success of a similar program in Nevada, which reduced unemployment in the state by over three weeks on average.
RESEA started as a pilot program in 2005, serving 77,000 UI claimants in its first year. Today, nearly all states run a RESEA program, serving over 1.2 million UI claimants. The Department of Labor’s Chief Evaluation Office works with states to help them evaluate the success of these programs.
RESEA is arguably the most evidence-mandated federal program — the problem is that it is still backwards-looking. States can evaluate new ideas, but there’s no guarantee that the successful programs will be scaled up or that ineffective ones will be eliminated. The Chief Evaluation Office is constrained; it can provide evaluation support and technical assistance, but is not in a position to scale successes or kill failed models.
To prepare for an unprecedented labor market disruption, the Chief Evaluation Office needs a significantly larger mandate with the power to actually ensure that successful labor market interventions are scaled across the country.
There are existing coordinated research programs this could be modeled after, such as the Defense Advanced Research Projects Agency (DARPA). DARPA has been used to push forward the development of new technologies that would be useful for defense. Critical inventions like the GPS and the internet all benefited from DARPA-coordinated research.
Following the DARPA model, project managers within the Department of Labor’s coordinated research program would be empowered to identify, support, and expand programs that have been proven to assist reemployment. For example, if Nevada’s RESEA program was able to reduce unemployment duration by three weeks, more focused experimentation and milestone payments for success could identify specific mechanisms in the program that assist reemployment even more.
Instead of each state experimenting with ideas in near isolation, project managers would fund specific experiments to improve program operations. This could be continued grants to states, but could also include for-profit, non-profit, and academic partners.
The key is making sure that evidence-backed reemployment programs are acted upon. A successful program results in more funding and scaling; a program without success gets cut. This sort of framework allows for iterative, goal-directed research, rather than an ad-hoc program evaluation process.
While the impacts of AI so far are limited, we could be on the verge of unprecedented labor market disruption. A coordinated research program embedded inside the Department of Labor could serve as a critical institution to advise policymakers on the programs that will be needed to solve future issues in the labor market. There’s a lot of speculation about what AI-induced job displacement might look like, but this is a tangible step that could be taken today to solve the problems of the future.
 | A guest post by
|
I suspect some data is being repressed or underreported. Hundreds of thousands have been cut from federal and corporate payrolls. And companies are not hiring if they can get the job done with AI. That has to be a number that someone can measure.