AI threatens a new jobs shock—hitting white-collar workers hardest and widening inequality—unless education, skills, and policy responses catch up quickly.
Editor’s note: The Red Cell series is published in collaboration with the Stimson Center. Drawing upon the legacy of the CIA’s Red Cell—established following the September 11 attacks to avoid similar analytic failures in the future—the project works to challenge assumptions, misperceptions, and groupthink with a view to encouraging alternative approaches to America’s foreign and national security policy challenges. For more information about the Stimson Center’s Red Cell Project, see here.
Red Cell
Rarely has such an unfolding transformational change been in plain view yet elicited so little forethought of the large-scale social repercussions. While the extent and speed with which jobs are lost is unclear, the United States should take a lesson from its earlier China Shock episode and prepare ahead of time for a wave of lost employment by upping workers’ skills for the jobs that artificial intelligence (AI) will create. Moreover, with inequality already approaching unprecedented levels, US democracy won’t survive with all the benefits of AI flowing to capital as happened during globalization. Some of the productivity gains would be better spent on avoiding a new cohort of losers.
The Middle Class Will Be Hurt This Time
There is no end to the warnings. The IMF has proclaimed that about 60 percent of jobs may be impacted by AI in advanced economies. Roughly half the exposed jobs may benefit from AI integration, enhancing productivity. For the other half, AI applications may execute key tasks currently performed by humans, which could lower labor demand, leading to lower wages and reduced hiring. In the most extreme cases, some of these jobs may disappear. With rapid advances towards agentic AI capable of perceiving the environment around it, reasoning, planning, and taking action to achieve complex, multi-step goals with minimal human intervention, the balance could shift to more jobs lost.
JP Morgan has said that “half of the vulnerable jobs in the United States will be automated away over the next 20 years.” They believe that, unlike with past technological advances, where lower-skilled jobs were made redundant, those most at risk with AI would be “white-collar professional service jobs such as budget analysis and technical writing,” going on to say that “[these jobs] look more vulnerable than childcare work or pipelaying.”
A 2023 Pew Research Center study found that “workers with a bachelor’s degree or more (27%) are more than twice as likely as those with a high school diploma only (12%) to see the most exposure” and be at risk of losing their jobs.
More recently, Dario Amodei—CEO of Anthropic, one of the world’s most powerful creators of artificial intelligence—has made perhaps the scariest prediction so far: “AI could wipe out half of all entry-level white-collar jobs—and spike unemployment to 10-20% in the next one to five years.” He thinks that “AI companies and government need to stop ‘sugar-coating’ what’s coming: the possible mass elimination of jobs across technology, finance, law, consulting, and other white-collar professions, especially entry-level gigs.” Ford Company CEO Jim Farley agrees with Amodei’s prediction on white-collar jobs, arguing on the other hand that demand for trade skills will surge with the AI boom, requiring workers to build and service data centers: “There’s already a massive shortage of trade workers.”
Not surprisingly, “experts are far more positive and enthusiastic about AI than the public,” according to an April 2025 Pew Research Center poll. “Far more of the experts we surveyed believe these technologies will benefit (76%) rather than harm (15%) them personally,” while the “public is far more likely to think AI will harm them (43%) than benefit them (24%).” Among the experts, men are much more likely to be optimistic about AI than women. A majority of experts and the public agree on the need for more regulation.
How Much Should We Worry?
Harvard Business School’s Christopher Stanton cautions that predictions about job loss are “extraordinarily hard.” “We had lots of discussions in 2017, 2018, 2019, around whether we should stop training radiologists. But radiologists are as busy as ever, and we didn’t stop training them. They’re doing more, and one of the reasons is that the cost of imaging has fallen. And at least some of them have some AI tools at their fingertips.”
At the same time, Stanton points to some categories of jobs that will be eliminated. “It looks like a lot of the code for early-stage startups is now being written by AI. Four or five years ago, that wouldn’t have been true at all.” Many of the graduates who went into STEM because it was safe are now finding that some jobs have been eliminated. At the same time, studies of randomized rollouts of conversational AI tools show that lower-performing workers or those at the bottom of the productivity distribution disproportionately benefit from these tools. If these workers have knowledge gaps, the AIs fill in the knowledge gaps.
Jobs most at risk include ones involving repetitive tasks, data processing, and basic customer service: retail cashiers, telemarketers, and many administrative occupations such as scheduling and record-keeping. The World Economic Forum believes that “women are also more likely to have jobs that are being disrupted by automation and GenAI, such as administrative assistants.” WEF also sees fewer women entering the AI field, even though more women than men go to university. The exception is China, where women are involved in public policy making, AI research, and AI education.
Of course, every technological innovation spawns new jobs: McKinsey believes job growth will be more concentrated in high-skill jobs across different professions, from healthcare to science and technology, engineering, and manufacturing. However, other trends unrelated to AI, such as climate change, will shape the future employment landscape. In another report, WEF surveyed 1,000 employers around the world, predicting that about 170 million new jobs will be created this decade, with farm workers at the top of the list, along with AI professionals. WEF explained that “green transition trends, including efforts to reduce carbon emissions and adapt to the climate crisis, will drive growth that will create 34 million additional jobs by 2030, adding to the 200 million farmworkers today.”
Could AI Be Derailed?
Recent scientific research questions the intellectual advances of Large Language Models (LLMs) and their ability to supersede human intelligence. Studies have cast doubt on whether those models have even a basic understanding of general logical concepts or an accurate grasp of their own “thought process.” According to scientific studies, these “reasoning models can often produce incoherent, logically unsound answers when questions include irrelevant clauses or deviate even slightly from common templates found in their training data.” Eric Schmidt, who has been an AI promoter, now believes US technology chiefs are on the wrong path, believing AI superintelligence may be a chimera.
For a long time, observers have worried about the racial, gender, and other biases that were built into the training data for the model(s). Scientists don’t believe they can eliminate all bias, although they are developing methods to hopefully reduce it. Nevertheless, a few high-profile court cases in which there has been discrimination in hiring due to LLMs’ biased evaluation of the different candidates or where candidates are refused credit could highlight the fallibility of LLMs. Research has shown that LLMs can generate biased outcomes in legal assessments, potentially leading to unfair judgments against marginalized groups.
There are other problems in companies implementing AI. RAND reported that more than 80 percent of AI projects fail, twice the already-high rate of failure in corporate information technology (IT) projects that do not involve AI. RAND cited a number of reasons for businesses wishing to employ AI, including a lack of the right data or that the organization is focusing too much on using the latest and greatest technology rather than solving the real problems that AI might not be adapted to. Finally, “AI projects fail because the technology is applied to problems that are too difficult for AI to solve.”
According to S&P Global Market Intelligence’s 2025 survey of over 1,000 enterprises across North America and Europe, 42 percent of companies abandoned most of their AI initiatives this year—a dramatic spike from just 17 percent in 2024. The average organization scrapped 46 percent of AI proof-of-concepts before they reached production. Those “outliers”that do succeed—often taking multiple tries and making changes—saw great success with increased productivity. AI implementation might be slower and more limited than suggested by the hype, but the benefits will remain attractive to businesses and governments that succeed.
Taking Off the Rose-Colored Glasses
Tech, like globalization in the 1990s, is too often viewed as a vehicle of universal progress, ignoring the effects of creative destruction. Marc Andreessen’s Optimist Manifesto, heralding the advent of unstoppable AI, is comparable to Bill Clinton’s embrace of globalization as a force of nature, like wind or water.
Like other technologies, AI can transform our lives with positive changes, but the excesses should not be ignored. Early nineteenth-century industrialization brought misery, sickness, and cruelty to children as young as five employed in textile factories working 12-hour days. One can still be an optimist without losing touch with reality. AI’s disruption of the white-collar jobs market needs to be tackled; otherwise, it will turn into further fodder for populism.
One way would be to better equip Americans with the needed STEM skills, as well as to institute lifelong learning for adults. The average half-life of skills is now less than five years, and in some tech fields, accelerated by AI, it’s as low as two and a half years. Some large businesses have been reskilling their workforces to better equip them to keep up with rapid tech changes. But so far, it is too little, too late. There is an urgent need to rethink education, prioritizing skills training. One recent BCG study indicates that the investments in reskilling represented only 1.5 percent of those organizations’ total budgets. Even before the AI craze, the OECD was highlighting that as digitalization spread, millions of workers would need to be entirely reskilled. The case is even stronger today.
In general, the United States has let its educational standards slip significantly. The average US student assessment scores“rose through the 2000s, plateaued during the 2010s, and then declined sharply during the pandemic.” Compared to other rich countries, US students rank 28 out of the 37 Organization for Economic Co-operation and Development (OECD) member countries in math, according to the latest tests by the Program for International Student Assessment (PISA). However, US students did better on science, ranking number 12 out of 37 OECD countries.
Besides not achieving the top position in the PISA rankings, what sets US students apart from those of other nationalities—and may help explain the more middling scores—is the significant gap between high and low achievers, which is not the case for other countries. Since 2012, an American Enterprise Institute (AEI) study has shown “a stark separation in trends across percentile groups: The lower percentiles trend downward, while the relatively higher percentiles are closer to zero.” The gap is present across all subjects and has “widened considerably” over the past decade.
Moreover, continuous surveys of the literacy and numeracy of the US adult population (ages 16–65) show steep declines after 2017, like the student scores, though a bit later than the dates 2012-2015 for the declines in the student population. The widening educational gaps reflect the growing class divisions, as the lower student and adult scores appear to correlate with more screen time and less time reading, as well as lower household income. If technology and globalization helped build middle classes in developing countries, they widened the US income and wealth divisions. The United States is among the most unequal in terms of income and wealth among advanced countries, largely due to the rich getting richer. AI will make this worse unless there is a concerted effort to avoid a bifurcated future.
China Shock has turned into a catch-all phrase useful for blaming all ills on China, but originally it referred to not just the manufacturing jobs lost with globalization, but the inability of those out of work to move on and find new, equally rewarding jobs. We know the political ramifications. According to economic historian Carlota Perez, populism is “typical of a midway turning point in the diffusion of a technological revolution.” In the earlier industrial cycle, Henry Ford’s assembly line led to similar or more extreme politics of fascism and communism in 1930s Europe, which the United States largely avoided due to the beginnings of the welfare state under President Franklin D. Roosevelt. Knowing the likely AI impacts on the US working and middle classes, that movie should not have to be rewound and relived. We can’t say this time that we could not see the AI jobs shock coming.
About the Authors: Matthew Burrows and Robert A. Manning
Mathew Burrows serves as Counselor in the Executive Office at the Stimson Center and is co-author of the recent book World to Come: Return of Trump and End of the Old Order.
Robert A. Manning is a Distinguished Fellow at the Stimson Center, working on Strategic Foresight, China, and great power competition.
Image: chayanuphol/shutterstock
