On March 10, The Budget Lab at Yale, The Hamilton Project at the Brookings Institution, and Peterson Institute for International Economics hosted an event on artificial intelligence (AI)’s impact on the labor market. The event coincided with the release of an analysis from Jed Kolko exploring research on AI and the labor market.
Peter Orszag (Lazard) opened the event with remarks on how AI is reshaping the economy. He articulated three “bets” about AI’s impact on the labor market:
- Although AI has not affected the labor market all that much to date, it will do so increasingly over the next five years.
- AI will accelerate a trend we have already been seeing: a widening spread in corporate performance.
- The view that AI cannot contribute to judgment will increasingly be proven wrong.
Peter Orszag speaks at “AI + work: Understanding AI’s impact on the labor market.”
Next, Jed Kolko (Peterson Institute for International Economics) reviewed the state of research about AI and the labor market. He concluded that the surge of early research (1) is still inconclusive, (2) provides weak signals about the future, and (3) only focuses on one corner of the vast AI research landscape. Kolko discussed each of these conclusions and offered principles for future research: think comprehensively about labor market effects; contribute to collective data infrastructure; make research useful for decisionmakers; and be broad-minded about historical analogues.
The research presentation was followed by a panel with Bharat Chandar (Stanford Digital Economy Lab and Institute for Human-Centered Artificial Intelligence), Martha Gimbel (Budget Lab at Yale), and Nathan Goldschlag (Economic Innovation Group), moderated by Ben Casselman (The New York Times).
Chandar discussed the early evidence of how AI is affecting the labor market. The current literature, Chandar said, suggests that “the aggregate effect of AI on the labor market is quite small.” However, he added that there is greater uncertainty when looking at specific groups of workers, for example, entry-level workers in jobs that are more exposed. Chandar emphasized the importance of continuing to collect data and track whether changes continue to accelerate.
Goldschlag agreed that existing studies are consistent in finding no significant aggregate impact but pointed to various challenges in identifying AI’s effect on the labor market, saying, “Are the tools that we’re using, and the ways that we’re using the data, up to the task? I’m not so sure.”
Gimbel reiterated how difficult it is to identify AI’s current impact on the labor market, stating that “we are expecting both too much of the technology and of the data.” She explained that broad categories in data may not capture differences in adoption across people and firms that “look similar.” Additionally, Gimbel argued that adoption takes time, and it is unrealistic to expect AI to already be having mass effects in the labor market. Finally, Gimbel cautioned against using CEO statements and layoff announcements to assess AI’s impact on the labor market in the absence of data.
To conclude, Casselman asked the panel about their ideal data set or source. The panelists proposed: data more directly linking firm-level and employment data; employee-employer data with occupation information; and labor market and AI adoption data from countries outside the U.S.
This is a summary of “AI + work: Understanding AI’s impact on the labor market.”
