The Biggest Economic Issue No One Is Talking About

How to address job loss and disruption from AI

In the next decade, artificial intelligence will probably replace the jobs of millions of American workers and reshape the working experience for millions more. Surveys show that Americans are worried about the impact of AI on their jobs and think it will do more harm than good to society.

Job losses and disruption from AI could be the biggest economic issue in the 2028 presidential election. And yet, no national politician in either party has proposed anything resembling a comprehensive plan to address the issue. That leaves a big political opening—and a chance for enterprising policymakers to take the lead in defining how our economy works in the coming decades.

In this post, we lay out the latest evidence on AI’s likely effect on America’s labor market and identify several types of proposals policymakers could consider to harness AI’s potential benefits while protecting the interests of American workers. While there is no silver bullet solution, we argue that empowering workers to have more say in how AI is adopted in their workplace and imposing new corporate taxes to fund income supplements for workers in certain fields are the best options available.

AI is likely to have a big impact on the American job market in the next few years—and Americans are worried about it

Estimates vary, but at the high end, both Goldman Sachs and McKinsey project that roughly 7% of American jobs—almost 12 million—could be fully replaced by AI within the next decade. Roughly 15 to 20 million Americans get laid off each year, so 12 million job losses in a decade may not seem like much. But the fact that there is a single cause for these layoffs—and as we explain below, the fact that the job losses are likely to be concentrated in particular industries and in particular cities—will likely raise the salience of the issue. As context, consider that 12 million job losses from AI would be nearly five times the size of the job loss from the China Shock of the early 2000s, which fundamentally remade American manufacturing and reoriented the politics of the Midwest.

Moreover, these job losses could come sooner than anticipated if a recession hits in the next few years. Past recessions have encouraged companies to turn to automation to replace workers. Researchers have found that since the mid-1980s, 88% of automation-related job losses in the US have occurred in the first year of recessions. With some economists projecting a 40% chance of a recession in 2025 alone, it’s quite likely that companies could pull forward AI deployment and accelerate the projected job losses within the next year or two.

But job losses are only part of the story. In the next few years, millions of workers are going to see their day-to-day job tasks change because of AI. Customer service agents might spend less time answering routine questions and more time handling complex cases. A software engineer might spend less time writing code and more time validating and revising AI-generated output. A graphic designer might spend less time on production and more time on strategy. Researchers at the Brookings Institute project that 30% of all workers could see at least 50% of their tasks change because of AI, while a different study suggested 80% of workers could see at least 10% of their tasks change.

Certain types of jobs will bear the brunt of these changes. Unlike previous waves of automation that largely affected blue-collar workers, widespread adoption of AI will change white-collar jobs the most. Workers that do more routine, administrative work—like call center workers and clerical workers—face higher risk of outright job loss. Coding, consultant, and financial analyst roles are also highly exposed to AI. Microsoft and Google, for example, now generate nearly 25% of their code using AI. Meanwhile, other STEM-related white collar jobs are likely to change significantly from the use of AI, but not be displaced entirely.

These concentrated effects mean certain cities and regions will be affected more than others. While factory towns have shouldered the losses from previous rounds of automation, Brookings’ recent report and Anthropic’s usage data suggest that the regions most exposed to AI will be urban, office-oriented areas with high levels of tech and information jobs—cities like San Francisco and Seattle. The Dallas Federal Reserve adds that New York City and Boston, with their high concentrations of highly-educated workers, may see more augmentation from AI adoption, while other metro areas with higher concentrations of call center and human resources jobs could see more job loss.

The job market impacts from AI adoption are already starting to appear in the data on recent college graduates. A May report from Oxford Economics found that 85% of the increase in unemployment since mid-2023 can be attributed to a reduction in hiring of people just entering the workforce. The decline in employment is concentrated in professional, scientific, and technical services, finance and insurance, and information technology—fields that are highly exposed to AI adoption. And according to a recent report by SignalFire, AI adoption has contributed to the top 15 tech companies cutting hiring of new graduates substantially since 2022.

Those closest to this technology are raising even more dire warnings. Anthropic’s CEO Dario Amodei recently predicted that AI could eliminate half of all entry-level white-collar jobs within five years. And a top executive at LinkedIn, Aneesh Raman, recently warned that AI is “breaking the bottom rung of the career ladder” by threatening the types of jobs that have historically served as stepping stones for young workers just beginning their careers.

Given these impacts, it’s not surprising that Americans are concerned about how AI will affect their lives and their jobs. A recent survey from Pew found that far more Americans are worried or overwhelmed by AI’s potential impact in their workplace than hopeful or excited about it.

And a recent YouGov survey found that far more Americans—across party lines—believe that AI will have a negative impact on society than a positive impact.

Potential options for policymakers

Policymakers face a serious challenge. AI could enable breakthroughs in medical research and treatment, increase productivity and economic growth, reduce some consumer costs, and help the US outcompete other leading economies. But, absent significant intervention, millions of American workers could see their jobs degrade or disappear in the next few years. We know what has happened in the last two decades: automation has helped drive down the share of national income that goes to American workers. What policymakers need now is a set of policies that helps the US harness potential benefits of AI while ensuring that—this time around—American workers don’t feel disempowered by technological change and instead see their slice of the pie grow.

We see several potential categories of policy intervention:

Compensate the “Losers”

In prior periods of labor market turmoil, our government has adopted a compensate-the-“losers” approach. In short, this approach entails letting the labor market change play out, identifying the specific workers who have lost jobs, and then trying to compensate those workers either with direct government support or with retraining or upskilling programs intended to help them find future employment with comparable pay.

These types of efforts have a poor track record. For example, in 2002, as China was entering the World Trade Organization and the impacts of the “China Shock” were becoming evident to US policymakers, Congress expanded the Trade Adjustment Assistance Program (TAA) to help support workers displaced by globalization and trade. Initially established in 1960, TAA provides training and modest subsidies for workers displaced by trade liberalization. However, the TAA program was woefully inadequate for addressing the scale and persistence of the China Shock. Multiple studies have found that TAA participants are worse off, as measured by future wages and benefits. A Brookings study found that TAA has helped fewer than 75,000 new workers per year—a tiny fraction of those who lost jobs from new trade deals. Between 2001 and 2004, an average of only 64% of participants found jobs when they participated in TAA, and those who were able to find new jobs had earnings that were 20% below their prior jobs.

Similarly, in the aftermath of the 2008 financial crisis, there was a push to retrain newly unemployed workers for coding and software development careers. Coding bootcamps proliferated, with promises of six-figure salaries after just three months of training. The Obama administration launched initiatives like "Computer Science for All" and "TechHire" specifically to funnel workers into programming careers, while community colleges created accelerated software development degree programs. Many bootcamp graduates discovered they lacked the depth of knowledge to compete with formally trained computer scientists, often landing in lower-tier positions or struggling to find employment at all. And many entry-level coding jobs that were targets for these retraining programs—like basic front-end development, routine back-end coding, and quality assurance testing—are now precisely the roles being replaced by AI.

In short, this approach does nothing to respond to worker concerns about protecting their autonomy and livelihoods, and has a spotty track record of being able to help set laid-off workers on new and lucrative career paths. It is an inadequate response to the labor market challenges that widespread AI adoption creates.

Transparency or Notice Requirements

Another category is greater transparency and notice requirements for employers adopting AI in the workplace. For example, policymakers could impose minimum notice periods (e.g., 6-12 months) before AI implementation that affects employment. In practice, this could mean adjusting the WARN Act process for notifying employees of forthcoming layoffs, as some policy experts have suggested. By extending notice period for mass layoffs from 60 up to 90 or more days, particularly in sectors most vulnerable to AI, people theoretically would have more time to retrain and prepare for a new career. Alternatively, policymakers could require companies to disclose how many workers they are laying off or shifting to new work because of AI adoption, which could help younger people choose different career paths and help policymakers design further interventions.

We are skeptical of the usefulness of these types of laws as the primary response to AI’s labor market impact. They do little to actually protect worker autonomy while also imposing requirements on companies that can be burdensome. Transparency and advance notice can be valuable, but they are insufficient for addressing the broad labor market transformation AI is likely to produce.

AI Restrictions

A more aggressive approach would be to ban or severely limit the introduction of AI in certain sectors. Some jurisdictions in the European Union—as well as American cities like San Francisco, Boston, and Portland—have already passed restrictions on the use of AI for law enforcement, particularly related to facial recognition. There have also been several proposed bills, including in California and Indiana, that would ban autonomous trucks by requiring human operators. California has also taken steps in recent years to regulate the use of AI in the judicial system by providing guidance to judges on how they should use AI without mandating an outright ban.

We are supportive of guardrails to ensure the implementation of AI is safe and ethical—especially in areas like law enforcement, criminal justice, healthcare, finance, and education. But we do not believe wholesale bans on AI adoption are the right general approach for addressing the impact of AI on workers. While projections vary widely—from Goldman Sachs forecasting a 7% increase in global GDP to MIT's Daron Acemoglu's more conservative 1% GDP boost over ten years—the evidence suggests that AI could spur more economic growth. PwC found that productivity growth has nearly quadrupled in AI-exposed industries since 2022, rising from 7% to 27%. We should try to capture as much of that economic boost as we can while channeling the benefits into better outcomes for workers.

Empowering Workers to Have More Say in How AI is Used by Employers

While we are skeptical of other options, we urge policymakers to consider ways to empower workers so they have more ability to shape how their employers deploy AI in their workplace.

Consider the 2023 Writers Guild of America strike. The strike led to the successful negotiation of AI usage protocols requiring studios to disclose AI use to the public, prohibiting studios from using AI to generate original content in place of human writers, and barring studios from requiring writers to use AI to help create content. In that case, the existence of a unionized workforce permitted workers to bargain directly with management and protect their core interests—including their continued ability to engage in the artistic expression of writing—in the face of potential AI adoption.

This type of firm-by-firm or industry-by-industry negotiation could help workers effectuate their priorities and work through tradeoffs unique to their firm or field. For example, while the screenwriters seemed interested in limiting AI adoption in order to protect the creative elements of their job, other types of workers might happily trade more AI usage in their workplace for higher pay and better benefits. Others might prioritize a four-day work week rather than more compensation. Still others might welcome the introduction of AI with no additional compensation because it eliminates mundane tasks they would rather not do. But Hollywood screenwriters are the exception, not the rule, because only 6% of private-sector workers in the US are unionized. In fact, research from the Brookings Institute shows that the sectors most exposed to AI are also the ones with disproportionately low levels of unionization. The workers that could most benefit from collective bargaining in the face of AI are the ones least likely to have access to it.

Policymakers should explore options for empowering workers in this critical moment. That could certainly involve traditional efforts to make it easier to join a union and collectively bargain, like the PRO Act. But policymakers should also consider new forms of worker empowerment tailored specifically to the rise of AI. One model could be sectoral bargaining over AI use: allowing workers across an industry—not just within a single firm—to collaborate with employers to set standards for how AI is adopted.

A recent precedent is California’s Fast Food Council, which was established through legislation that fast food workers and the Service Employees International Union (SEIU) supported. The council is not a traditional union with collective bargaining power, but it includes worker and employer representatives and has the authority to set binding standards on wages, working conditions, and safety.

Similar models could be used in industries with high risk of AI disruption and low unionization rates—like computer programming, HR, finance, and more—to ensure workers have a say in how AI is introduced. Policymakers could also consider initiatives that require companies to consult with worker representatives before implementing AI systems, drawing inspiration from Germany’s works councils, which provide formal mechanisms for workers to influence technology decisions in the workplace.

New Taxes on Corporate AI Beneficiaries to Fund Important Societal Work

We also urge policymakers to consider approaches that tax the corporate beneficiaries of AI adoption to fund societally important work.

The case for broadly shared AI benefits is even stronger than with past technological revolutions. Unlike the steam engine, assembly line, or computers, AI fundamentally depends on humanity's collective intellectual output. It is trained on billions of books, articles, images, and conversations representing an unprecedented appropriation of human knowledge. While steam engines consumed coal and factories processed raw materials, AI processes human ideas, creativity, and cultural expressions. If our shared human knowledge is the foundation of AI's productivity gains, then workers deserve to share in those benefits rather than simply bearing the costs.

The two-day weekend represents a powerful example of productivity gains directly benefiting workers. It was made possible by the industrial revolution’s transformation of work in the late 19th century. If we are poised to see significant productivity and GDP growth from AI, we should be thinking about what our version of the weekend could be.

As AI drives productivity gains that aren't reflected in employment, new mechanisms can redistribute these benefits to support socially necessary but undervalued work. Rather than jump to solutions like universal basic income, we should prioritize solutions that encourage employment, particularly for jobs that contribute to society and are unlikely to be displaced. Jobs in areas like education, child care, elder care, mental health counseling, and social work are likely to become only more important as we enter this period of transition, and they each suffer from shortages of available workers in part because of lower pay. We should also try to incentivize work that facilitates social cohesion and community-building, especially given the alarming studies around the rising rates of loneliness and antisocial behavior.

In recent years, some countries have implemented strategies to try to redirect the productivity gains from automation to better compensate work with high social value. For example, South Korea has implemented a reduced tax deduction for automation investments that directly replace human workers, using the resulting revenue to fund expansions of their caregiving workforce. There are multiple ways that this tax could be designed, including a blanket corporate tax rate increase, a tax specifically on the AI firms most poised to benefit from this technology, a tax on corporate subscriptions to AI models, or a tax on firms that displace workers in favor of automation or AI.

Conclusion

AI could replace millions of American jobs and eliminate certain career paths entirely, but American policymakers have not even begun a serious conversation about what to do about it. We can’t let helplessness prevail: we can still use the tools of government to produce a better future for workers and for the country. We hope policymakers will consider the ideas we’ve laid out here and start to tackle the issue head on.