California is tracking AI job losses — but should track gains, too

More than half of Americans fear that they, or someone in their household, could lose their job to AI.

A recent Reuters/Ipsos poll ‌showed that concern about AI-related job losses is widely shared across the population (though Democrats are more concerned than Republicans, who tend to represent working-class voters in the Trump era).

But while AI could create new challenges in the job market, it could also create new opportunities.

In that context, California deserves credit for something rare in public policy: building an ambitious new labor-market measurement tool to monitor AI-related job losses.

The California Policy Lab, working with the state’s Employment Development Department, recently launched a public dashboard tracking AI-related unemployment claims in near-real time. It adds to the toolbox that economists have to analyze AI, along with the Stanford Digital Economy Lab dashboard.

The researchers behind the project have been admirably transparent about what their data can — and cannot — show.

That effort deserves recognition, but there are also major limitations in the tracker.

For starters, it measures only one side of AI’s effect on the labor market: jobs that disappear. It cannot measure the jobs AI creates, the workers it helps retain, or the productivity gains that lead firms to hire more people. A dashboard focused solely on unemployment risks telling only half the story.

Another limitation is that California still records workers’ occupations using the Dictionary of Occupational Titles — a federal system last comprehensively updated in 1991. Modern occupations such as data scientists and machine-learning engineers simply do not exist in it.

Researchers therefore must translate obsolete occupation codes into modern measures of AI exposure, averaging together workers whose jobs may look very different today.

That is not a criticism of the research team. They are making the best use of imperfect data.

Download The California Post App, follow us on social, and subscribe to our newsletters

The same is true of the tracker’s main finding: Unemployment claims rose among workers in AI-exposed occupations beginning in late 2022. The timing naturally invites comparisons with the arrival of ChatGPT.

But late 2022 also marked the collapse of the pandemic-era technology hiring boom. As interest rates rose, companies including Meta, Google, Amazon and Salesforce announced tens of thousands of layoffs after years of expansion. Many of the occupations most exposed to AI were also those most affected by this correction.

Distinguishing between the effects of generative AI and the end of the tech boom is extraordinarily difficult using unemployment claims alone.

Even if every increase in joblessness reflected in influence of AI, however, unemployment claims would still capture only separations — not what happens to employees afterward. That missing half of the equation matters.

In my research with economist Andrew Johnston using administrative data covering nearly every employer in the United States, we have found that industries more exposed to AI experienced faster productivity growth, higher employment, and higher wages through 2024. Other recent studies likewise find surprisingly modest evidence of widespread labor-market disruption from large language models.

A new analysis using company spending records finds that firms investing most heavily in AI expanded employment more rapidly than comparable firms that invested less.

Reasonable researchers can disagree about the magnitude of AI’s effects. But the broader evidence increasingly points toward a familiar pattern from previous technological revolutions: Some jobs disappear, many jobs change, and new opportunities emerge alongside higher productivity.

A dashboard that counts only layoffs cannot capture that process.

Imagine if California had introduced an “Automobile Job Loss Tracker” in 1910. It would have carefully documented every buggy-whip maker and carriage painter who lost work, while recording nothing about the mechanics, assembly-line workers, gas-station attendants, truck drivers, and highway engineers the automobile was creating. The data would have been accurate — and deeply misleading.

California has an opportunity to build something even more valuable. The state already possesses quarterly wage records covering virtually every employer. Those records can reveal whether displaced workers quickly find new jobs, whether their earnings rise or fall, and which industries are expanding.

Combined with modern job-posting data, these existing data sets could identify where demand for AI-related skills is growing across occupations, industries, and regions.

Better still, California could modernize the information employers report in the first place. Replacing outdated occupation codes with the federal Standard Occupational Classification system — or requiring employers to report workers’ job titles on quarterly wage records — would give policymakers a far clearer picture of how work is changing.

Workers facing technological change do not simply need to know how many people lost jobs like theirs. They need to know where opportunities are emerging, what skills employers are seeking, what those jobs pay, and how to qualify for them.

California now has an AI job-loss tracker, but it should now build the nation’s first AI opportunity tracker. That would not only measure disruption, but also help workers navigate it.

Christos A. Makridis is an associate research professor at Arizona State University. He holds dual doctorates in economics and management science & engineering, both from Stanford University.