The AI economic transformation has begun. In May, IBM declared that it had fired hundreds of employees and replaced them with artificial intelligence chatbots. Over the summer, Salesforce let go of large numbers of people thanks to AI; UPS, JPMorgan Chase, and Wendy’s are also slashing head counts as they automate more functions. College graduates are having a harder time finding entry-level jobs than they have in nearly a decade. And these trends are just the beginning. In survey after survey, corporations across the world say that they plan on using AI to transform their workforces.
Artificial intelligence will likely create new employment opportunities even as it disrupts existing ones, and economists disagree on whether the net effect will be job losses, job gains, or simply restructuring. But whatever the long-term consequences are, AI will soon become a major political issue. If there is significant disruption, officials will be confronted by workers furious about jobs lost to machines. Voters will make their frustrations known at the ballot box. Politicians will therefore have to come up with plans for protecting their constituents, and fast.
To create an effective strategy for addressing large-scale AI disruption, however, policymakers will need to understand how workers themselves perceive the technological threat. In November 2023, we surveyed 6,000 Americans and Canadians to gauge their level of concern about AI-induced mass layoffs and how the government should deal with the issue. Our findings revealed the scale of the challenge: respondents ranked fears about AI taking their jobs ahead of all other concerns about the technology, including its potential for military use.
When it comes to policy preferences, though, there is ground for both optimism and pessimism. On the positive side of the ledger, most respondents favored measures like retraining programs and expanded safety nets—technocratic fixes that economists believe can work. But on the negative side, many also supported new trade restrictions and immigration barriers, strategies that could make the problem even worse and that governments may well be tempted, for political reasons, to adopt. Multiple countries, after all, responded to the layoffs created by offshoring with harsh tariffs and more deportations—even though neither technique worked. If they are serious about solving the problem and not incurring another round of populist backlash, policymakers should start rolling out the right responses now, before AI layoffs ramp up and while the most effective solutions still command widespread support.
THEORY AND PRACTICE
To determine how voters want the government to manage AI layoffs, we did not conduct a simple poll. Instead, we wrote up 81 scenarios involving either AI adoption or offshoring in which the economic shock had different effects on employment and prices. In one scenario, for instance, AI reduced smartphone prices by 50 percent while eliminating 25 percent of factory jobs and creating 25 percent more data science positions; in another, prices remained unchanged while customer service jobs decreased by 25 percent and factory employment stayed constant. We then gave respondents four of these scenarios to examine, each randomly chosen. We also presented respondents with a menu of possible policy responses—retraining programs, an expansion of the safety net, regulations to govern the economic shock they had seen (either AI deployment or offshoring), trade barriers, and immigration restrictions—and asked whether they supported each one. Respondents were randomly asked to evaluate either AI or offshoring scenarios, not both, allowing us to compare whether voters responded differently to domestic technological change versus foreign competition, and whether similar economic tradeoffs generated similar policy preferences across different types of disruption.
The results were clear. Regardless of what political party they belonged to, respondents in both countries ranked worker retraining as their preferred policy. Average support clocked in at four out of five, where one represents strong opposition and five strong support. Regulation of AI was the second most popular policy, also with broad support across the political spectrum. Expanding social spending, meanwhile, came in third—albeit with much less support among Republicans in the United States and slightly less support from conservatives in Canada.
These outcomes were encouraging. Ask economists what policy they would recommend in response to AI-driven layoffs, and most would say retraining, regulation, or social insurance. The logic is simple. Technological change can be slowed, but it is almost impossible to stop, and so the best thing governments can do for affected citizens is to give them new skills, set sensible guardrails, and create new unemployment benefits.
The problem is that governments today rarely put these policies into practice. In response to recent economic shocks, such as when trade slashed manufacturing jobs from wealthy countries, most states did not set up large retraining systems. The regulatory picture is equally grim. Despite the AI boom, few governments have passed comprehensive legislation related to AI—the European Union’s AI Act being the notable exception. And safety-net expansions look even less likely, particularly at a time when many governments are laden with debt. In fact, Washington is slashing social programs, including public health insurance and nutritional assistance, as part of U.S. President Donald Trump’s signature One Big Beautiful Bill Act.
Optimists might hope that as AI-induced disruption increases, policymakers will feel compelled to invest in retraining, social programs, regulations, or some combination thereof. But history suggests that the pressure to regulate and compensate could actually wane as the years go by. When it comes to economic dislocation, politicians face what social scientists call time-inconsistency problems. Before a disruptive technology is widely adopted—or a trade agreement is signed—those who stand to benefit have strong incentives to promise compensation to those who will lose out, in order to secure political buy-in. But once the technology is deployed or the agreement is in place, the incentives to follow through evaporate. Reversing the change becomes too costly for the state. The balance of power often shifts decisively toward the winners, who no longer need to placate the losers. The result is that compensation gets underfunded, poorly implemented, or abandoned altogether.
FALSE PROMISE
The biggest risk, however, may not be that governments will ignore effective fixes. It is that they will adopt policies that will backfire. Many politicians, particularly on the populist right, might respond to AI layoffs by trying to restrict immigration and trade—just as they have to past economic problems.
If they do, the argument will be straightforward. If a government can’t shield its people from competition by robots, then at least it can protect them from competition by foreigners. But this zero-sum logic does not hold up in practice. Virtually every piece of research suggests that restricting immigration and trade will not stop companies from adopting AI. In fact, it may hasten layoffs. Reducing trade, for example, will raise input costs, shrink export markets, and heighten policy uncertainty—pressures that make labor-saving technologies like automation more attractive in exposed industries. Reducing immigration will further encourage AI use by increasing labor costs.
Analysts can try to make these likely consequences clear to voters. But protectionism often polls quite well, and substantial numbers of people already support such steps as responses to AI shocks. Overall, support for immigration restrictions averaged 3.4 out of 5.0 in our survey, while trade restrictions averaged 3.2. Among Republicans, the pattern is even more striking: support for immigration restrictions averaged 4.0 out of 5.0, making it their single most popular policy response, even higher than retraining. Trade restrictions came in at 3.5 among Republicans, roughly equal to retraining and well ahead of social spending. If AI layoffs keep rising, those figures could prove to be low-water marks. According to a study by the political scientist Nicole Wu, when Americans are told that robots threaten employment, Republicans become markedly more hostile to immigrants while Democrats turn against trade. Almost no one favors slowing the pace of AI itself.
There are other reasons why politicians might turn to exclusionary policies. One is that there are comparatively few barriers to implementing these kinds of measures. To set up and fund retraining programs, regulate AI, or expand social welfare, most governments would need to pass legislation and appropriate significant amounts of government spending. Deporting migrants, by contrast, rarely requires fresh laws, and can thus be done relatively quickly. Another is that immigration restrictions and tariffs yield clearly measurable results—thousands of foreigners gone, hundreds of millions of dollars in tariff revenue—in ways that other policies do not. Finally, nativism and protectionism offer voters someone or something to blame. It is easier, after all, to be angry at foreign workers and foreign products than it is to be angry at technological progress.
If voters embrace nativist policies in response to AI, they are unlikely to revert to more effective solutions. According to research on European democracies by the political scientists Alan Jacobs and Mark Kayser, when people negatively affected by economic change turn to far-right parties, they tend to stick with them. Politicians who profit by peddling anti-immigrant or anti-trade rhetoric certainly have few incentives to bring voters back to the center. In fact, some states and parties that have traditionally been hostile to immigration, including the Japanese government, are even outwardly promoting AI as a substitute for foreign workers.
GET AHEAD
Many of these findings seem to bode poorly for both the future of work and the future of democracy. But as our survey findings show, the right course of action—retraining, regulation, and social welfare—is also the one that people want most. If they want to respond to popular demand, policymakers could pass laws establishing and funding retraining programs that teach workers how to work alongside AI systems, develop skills in sectors less susceptible to automation, or transition into new roles created by AI. They could set up new income support programs for people caught between jobs. Finally, they could pass laws that regulate AI by requiring transparency in automated decision-making, mandating human oversight for high-stakes applications, and establishing liability frameworks for AI-caused harms, which would slow the most disruptive applications and ensure safer deployment without stifling innovation. Governments could pay for these proposals by taxing large AI companies. This would ensure that the businesses that profit from disruption also help manage its consequences.
These policies would not only help millions of workers. They could also help restore faith in government. By acknowledging workers who lose their employment to AI and offering them assistance, officials would demonstrate to voters that the state can, in fact, address their needs. In doing so, politicians would bolster their own political fortunes. According to research by the political economist Yotam Margalit, during the George W. Bush administration, incumbent parties performed better in counties where a larger share of laid-off workers qualified for retraining programs—evidence that voters’ access to government support mutes their potential political backlash to job loss. (The United States has funded retraining programs, but not nearly enough.)
Time, however, is running out. AI adoption is accelerating, and its deleterious effects on employment are no longer a speculative problem. They are already widespread and they will only accelerate in the months to come. Adaptive policies, meanwhile, will take years to yield results. If governments want to protect their economies—and themselves—they must act now.
Loading…