AI and Labor Markets: What We Know and Don’t Know

I wrote a new essay outlining my assessment about what we know and don’t know about AI and labor markets. There is a lot we don’t know!

Here is a link to the full essay. This post is a quick summary. See the essay for references and more details.

A number of recent papers find small effects of AI on overall employment. It does not seem likely that AI has caused widespread displacement so far.

On the other hand, younger workers in AI-exposed jobs face a challenging job market, as Erik, Ruyu, and I show in our recent “Canaries in the Coal Mine” paper. Other economic forces such as interest rate changes or tech overhiring do not seem to account for all our findings, which suggests AI might be playing an important role. Two new papers show similar findings in the US and UK using alternative data sources. One paper finds no effects of AI on entry-level employment in Denmark. More work should study these questions using alternative data sets and methods. An experiment comparing firms that use AI vs ones that don’t would be ideal. Better data on when firms start using AI would help, too.

A common complaint is that existing measures of AI exposure are poor because they do not have validation on real-world economic outcomes. This is less true than it used to be. Predictions of AI exposure correlate pretty strongly with actual usage. We also have new evidence that show these measures predict employment impacts.

In which dimensions should we expect AI to improve most rapidly? This is an active area of research that people should watch going forward. Understanding how the models are improving can help identify future occupational exposure.

More research should also consider evaluations of AI geared towards augmentation of work rather automation. Technology does not need to replace humans. Model development is shaped by the way we evaluate performance. More evaluations should take human-AI collaboration as a benchmark to optimize for.

We do not know what’s happening outside of the US and a couple countries in Western Europe.

What does it mean for a firm or worker to “adopt” AI? They use it once? Every day? 1% of the company uses it? 10%? How do we make sense of people using consumer applications to do enterprise work? These conceptual questions feed into wildly different estimates of firm adoption from different sources.

Business spend data on AI subscriptions seems especially promising here. We should also encourage AI companies to share data on this to the extent feasible.

Prior technologies decreased labor demand in some occupations. Simultaneously they increased labor demand in other occupations and created new forms of work. Will AI be different? We should build dashboards to track what’s happening.

AI may also affect the matching process between firms and employees. If job postings and resumes are mostly AI-generated, is that good or bad for the labor market? We should track statistics on how efficiently the labor market is connecting workers to firms.

At the same time, we should do scenario planning for potential future impacts. If certain occupations get disrupted, how can we smoothly transition people to alternative careers or opportunities?

For over a century school has served as the starting point for developing the skills needed to enter the workforce. How is AI changing the way students prepare for their personal and professional futures? We know a lot of students and teachers are using AI, but how is it affecting what students choose to study? What careers they choose? What about course curricula? Or overall levels of learning? We need much more data on these questions.

We also need to study how AI may facilitate labor market adjustments. AI may speed up the rate of job obsolescence, but it may also speed up the rate of skill retraining or reduce barriers to entering a new profession in the first place.

High earners tend to have higher occupational AI exposure on average, but they also receive a lower share of their income in wages. Lower earners tend to have lower occupational exposure on average, but a higher share of their income comes from wages. We should track changes in incomes across these different sources to measure how AI is shaping inequality.

We need more theoretical work that allows us to simulate the labor market impacts of policy changes, especially in settings with potentially transformative AI. This would represent a step forward in planning for future scenarios in a way that is informed by data and based on clear assumptions.

—

Here is a link to the full essay. Thanks for reading, and I hope to see more research on these topics!