“Pro-Worker AI” Means Deregulation

The coming of artificial intelligence means states must remove onerous restrictions on labor.

Courtesy Israel Andrade/Unsplash

Artificial intelligence probably won’t lead to mass unemployment in the long run. Much like automations past, the gains in productivity that will come from it will likely create new, more stimulating work to replace what is automated.

But in the short term, there will be winners and losers. One response, from the heavyweight economists Daron Acemoglu, David Autor, and Simon Johnson in their recent report “Building pro-worker AI,” is to leverage public spending, grants, and antitrust law with the aim of encouraging labor’s share of the returns from AI. Another response, endorsed by the likes of Senator Bernie Sanders, is to slam the brakes on innovation, taxing robots and turning to unions.

But that will make us worse off in the long run, depriving us of the growth that AI promises. The far better alternative is deregulating labor and allowing humans to compete. This would benefit both employees and consumers while also catalyzing the discovery of new, labor-augmenting applications of artificial intelligence.

AI’s perceived advantage over humans comes in part from its higher productivity in many tasks. But it also comes from the fact that the costs of capital can be negotiated much more easily than the costs of labor. The latter is subject to the thicket of regulation that surrounds employment; the former is not.

For example, imagine a paralegal willing to accept the average salary of $80,000 in New York City to gain experience before going to law school. The work is increasingly supported by AI tools like Harvey, which can assist lawyers in their research at a cost of $14,400 per lawyer per year. Hiring an extra paralegal still makes sense; Harvey can’t do everything, so our paralegal, who can help multiple lawyers, still provides a lot of value relative to only having the AI tool. But as AI becomes better, its value rises relative to an additional paralegal. At some point, the firm might decide that AI increases its output more than an additional paralegal.

Our fictional paralegal, still wanting to acquire experience, might have been willing to accept a lower salary. But she’s at a disadvantage relative to AI, because New York requires that her employer provide paid sick leave, paid family leave, and a transportation contribution, plus federally mandated health insurance and a 6.2 percent levy employers must pay to fund Social Security and Medicare. All of these raise total cost to her employer by roughly 40 percent, and can’t be negotiated in the contract.

So too are there other risks and costs that come with hiring a person over a machine. People can bring disputes at the National Labor Relations Board, where arbitration costs can reach $50,000, or a discrimination lawsuit, costing employers $75,000 on average to resolve.

If we don’t make it easier for workers to compete, then as artificial intelligence continues to improve, machines will become even more successful relative to humans. That matters for several reasons.

First, while AI may produce short-term unemployment, it may also force some workers out of the labor force altogether. That can turn workers into long-term beneficiaries of welfare programs. Not only is that bad for the public fisc, but it also deprives these people of the non-pecuniary benefits of work, including the meaning it provides in people’s lives.

Second, as Acemoglu, Autor, and Johnson explain in their report, to the extent that companies are biased toward automation, they divert AI resources from hard problems to easy ones. Automated resources spent on uses of middling value—like language app Duolingo’s choice to replace staff with AI only to face significant backlash—can’t be spent on hard problems—like Google DeepMind’s sequencing of the human genome.

Acemoglu et al. provide some ideas for rebalancing this short-term for automation. That includes proposals like loosening licensing laws so more workers can take advantage of AI’s potential “leveling” impact; using grant-making to develop pro-worker technologies; and notably, “rebalancing the tilted investment playing field in the tax code,” in which they urge “a more symmetric tax structure, where marginal taxes for hiring and training workers and for investing in equipment and/or software are equated.” Bernie Sanders’ proposals go much further, including a “robot tax” on large corporations, increased union membership, paid family leave, and shortening the workweek to 32 hours.

Many of the proposals put forth by Acemoglu et al., including loosening licensing laws, have merit. But other proposals—including most of those from Sanders—would worsen labor’s ability to compete with capital, specifically by increasing labor costs. Taxing automation, meanwhile, would both be practically difficult to implement and would have extraordinary costs given the potential positive payoffs of artificial intelligence.

Which leaves one serious alternative: if humans are to remain competitive against artificial intelligence in the short-term, policymakers must take deregulation of the labor market seriously.

States could, for example, leave certain now-mandatory benefits—like paid sick and family leave—to the discretion of employees. States like New York could also reduce their dependency on payroll taxes for things like transportation, instead moving to a model in which the end consumer pays a larger share of these costs directly.

Finally, despite the political barriers, states that reduce the influence of unions make it easier for people to get hired. Evidence shows that states with the strongest right-to-work laws are also those seeing the greatest population increases, and that over the long run workers under right-to-work laws are much better off.

Artificial intelligence isn’t the first technology to create a paradigm shift. But realizing its growth potential requires allowing individuals to adapt, tinker, and discover its many applications. And keeping humans competitive with it means, first and foremost, letting them actually compete.