There is a lively debate happening across various corners of the internet and the hallways of academia: In the coming decades, will advances in Artificial Intelligence (AI) render human labor unnecessary?

At first blush, economic data suggest little reason to worry. American businesses in recent years have needed more and more humans: 15 million new private-sector jobs have been created since early 2010. At the same time, what economists refer to as labor productivity (economic output per human hour worked) has grown disappointingly slowly. If robots were on track to replace most human jobs the opposite would be the case: The economy would produce more with fewer workers.

Slow productivity growth gives rise to a mirror-opposite concern: Technological development isn’t happening fast enough. Productivity growth, often stemming from new technologies, serves as one of the most important factors in rising incomes, and slow growth over the past several decades has substantially contributed to the lower wage growth over that same time period. AI stands to boost productivity growth, and in turn raise incomes and standards of living. For these reasons, AI should be celebrated, and one of the biggest questions should be how to get more of it.

But this is not to say that AI will necessarily have only positive effects on labor markets. Recent work by the Council of Economic Advisers (CEA), shows that lower-paying jobs are at much greater risk than those that pay handsomely. Eighty-three percent of jobs making less than $20 per hour are projected to come under pressure from automation, as compared to 31 percent of jobs making between $20 and $40 per hour, and just 4 percent of jobs making above $40 per hour, as the graph below illustrates.

Of course, advanced economies have seen vast amounts of innovation in the last three centuries without rendering human labor obsolete. Most of the types of jobs that existed in the 1700s do not exist today, but new types of jobs that no one could have imagined then have taken their place—all because of technological advances. A different trajectory is unlikely to emerge this time around because even though AI has the potential to replace certain human tasks, it will likely also create entirely new fields of jobs.

So 83 percent of the jobs making less than $20 per hour will not simply disappear. But it does mean that during a transition period low-wage jobs could face substantial downward pressure on their wages to remain viable in the face of competition from technological substitutes.

This change is coming at a time when existing trends in the economy present cause for concern. Over the past 60 years, the labor-force participation rate for “prime-age” men—the fraction of those between the ages of 25 and 54 who have a job or are actively looking for one—has declined steadily, from a high of 98 percent in the 1950s to 88 percent today. This raises concerns not only because workers’ productivity peaks during these ages but also because a large body of research has linked joblessness with a host of negative socioeconomic outcomes. (It should be noted that I am looking at men’s history as a guide to the potential outlook for both men and women in the future; the history of women’s employment is dominated by the one-time event of the massive entry into the workforce in the decades following World War II.)

The decline in prime-age male labor-force participation has been concentrated among men with a high-school degree or less and has also coincided with a decrease in their relative wages. CEA analysis suggests that rather than a supply-side effect—that these men just don’t want to work—the erosion in labor-force participation is likely driven by lackluster labor demand, resulting in both fewer employment opportunities and lower wages.

In all likelihood, while AI will not replace all human labor, there will be a period of transition that disproportionately affects these low-skilled workers. This period of turnover, in which workers displaced by new technology find new employment (often in jobs made possible by those very technologies), may last for years and can lead to a combination of lower wages (i.e. inequality) and extended periods of joblessness.

But technology is not destiny: These trends and the magnitude of inequality are not solely determined by technology. Across the world’s advanced economies, the share of income going to the top 1 percent ranges from 18 percent in the United States to less than 10 percent in France, Italy, and Japan (as shown in the figure below). These countries all have similar technological capacities, but different institutions, cultures, and policies.

Similarly, the fraction of the prime-age male population in the labor force in advanced economies ranges from 87 percent in Israel to 96 percent in Switzerland, with the United States below average at 88 percent. These differences in participation are enormous—to put them in context, the difference in the unemployment rate between good economic times and recession in the United States can be as little as 2 percentage points. And again, there is no major technological difference that explains the large discrepancies in employment rates among these countries.

This suggests that there is no economic reason that the United States cannot address inequality and increase employment while enjoying even higher levels of technology and productivity than seen today. What matters is how labor-market institutions—such as job-training programs, relocation assistance, licensing regimes, and so on—cope with these changes, support the creation of new jobs, and successfully match workers to them. An overhaul of the policies that shape labor markets—such as unemployment insurance, education and training, the tax system, collective bargaining and wage protections, and so on—could greatly improve the functioning of the U.S. labor market, regardless of whether automation and AI materialize as expected.

If labor-market institutions can be strengthened so that low-wage workers are able to prosper in the face of AI disruptions, then the question will be less how to avoid the advances but how to encourage them such that productivity growth can return to a positive trajectory.

Public policy can play a role in encouraging that growth too. To be sure, the private sector will be the main engine of progress on AI. But as it stands, there is an underinvestment in basic research—research conducted for the sole purpose of furthering the scientific knowledge base—in part because it is difficult for a private firm to get a return from their investment in such research. Federal spending can help make up for this shortfall. In addition, public policy can support developments in cybersecurity and privacy, both of which are critical ingredients for an economy increasingly reliant on technology. Finally, policymakers need to continue to make sure that there is enough competition in the market so that workers are paid fair wages, consumers enjoy lower prices and a wide variety of goods, and companies keep investing in research and development.

With the right conditions, AI is something to be welcomed, not feared.