New Technology, Same Anxieties: On AI

Every general-purpose technology arrives with the same prediction. This time, jobs are really gone for good. Tractors, ATMs, and spreadsheets were all supposed to lead to employment issues. None of them did; even when jobs were replaced, widespread job collapse did not. Early in the country’s history, for example, over 90% of people were employed in farming; today, around 1%. Is there widespread unemployment because of ag-tech? Do we have less food than ever before? Of course not.

AI is now getting the same treatment, with tech executives and lawmakers alike floating everything from mandated retraining programs to universal basic income to government equity stakes in AI firms as ways to soften the blow.

Nicholas Thielman and I have a new piece out with R Street Institute, making the case that this framing gets the policy problem backwards. The supposed solutions are all treating mass job replacement as a given. Instead, we point out that regulatory barriers are already making it harder for workers to explore opportunities in the workforce, AI or no AI. If anything, reducing those burdens will make humans more able to both realize the productivity gains and better compete with AI-replacement.

The evidence so far doesn’t support the “job apocalypse” story. Where AI complements human work, it has tended to raise employment and wages. Where it substitutes directly for tasks, the effects are more modest and concentrated, mostly showing up as fewer entry-level openings in a handful of highly exposed occupations instead of mass displacement. To be fair, though, it is too early for any “full” effect to be realized, but that is true for anyone making broad claims about AI. If history is any guidepost, rules and institutions that favor more flexibility and freedom allow for the best of both worlds: taking advantage of productivity gains from technology, and tasks that are replaced by technology are supplanted by better, higher-paying careers.

So far, the aggregate labor market effects remain small. That’s consistent with what we found in an earlier, longer working piece on this same question. Agriculture went from 40% of the workforce in the early 1900s to 1.2% without mass unemployment, ATMs increased teller employment rather than eliminating it, and spreadsheets turned bookkeepers into higher-paid, more productive analysts. In each case, the technology was allowed to diffuse in a reasonably open institutional environment, and the labor market did the rest.

Occupational licensing can be a major policy lever. This is the part of the piece I think deserves the most attention, and we plan to build on. Research on regional economic shocks (recessions, trade exposure, automation) consistently finds that how well a place recovers depends less on the size of the shock and more on the institutional environment it lands in. Areas with freer labor markets and lighter regulatory burdens bounce back faster.

Licensing is one of the oldest and most pervasive of these barriers: the share of the U.S. workforce required to hold an occupational license has grown from 5% in 1960 to 25% today. One study found that reducing occupational licensing can offset more than 90% of the negative effect that automation exposure has on workers’ income mobility.

My colleagues Ed Timmons and Clay Routledge make a notable point in a recent op-ed. I will close with their last point in their piece that is worth sitting with: “The era of AI will reward adaptability. Policymakers should respond to workers’ anxieties not by trying to shield them from change, but by giving them the freedom to better navigate it.” In other words, let’s give humans a chance!

Read the full R-Street piece here.