Discussions about AI’s impact on the workforce tend to focus on what the new technology will mean for white-collar workers, many with higher incomes. But new research by the Federal Reserve Bank of San Francisco shifts the conversation by examining on-the-job exposure to AI among lower income workers.
AI isn’t just a story about high-earning knowledge workers. It’s also reshaping the jobs of low-income workers. If we want AI to expand economic opportunity rather than narrow it, understanding how AI could impact lower-income workers matters.
Lower-income workers, particularly those in roles that require people to be physically engaged, might be able to utilize AI in a way that improves the quality of their job. However, many low-income workers face challenges that could make it harder to adapt to AI without better resources and support. For example:
To better understand key findings of this new research, I sat down with Elizabeth Kneebone, assistant vice president of research in Community Engagement and Analysis at the Federal Reserve Bank of San Francisco, and co-author of the report. The Community Engagement and Analysis team works to understand the economic experiences of lower-income households and communities to inform the Federal Reserve’s work. Here’s my conversation with Elizabeth:
Brooke: Your brief finds that more than 6 million workers in lower-income households are highly exposed to AI on the job. Can you briefly share what you mean by “AI exposure,” and what that exposure might look like for a worker’s day-to-day tasks?
Elizabeth: You can think of “AI exposure” as how much of a person’s job could be helped—or replaced—by generative AI tools like ChatGPT, Claude, Copilot, or Gemini.
To figure this out, we looked at different work activities and asked two key questions: First, how likely is it that AI could either assist with or take over this task? Second, how important is that task to the overall job? Looking at these measures across all occupations, we were able to rank jobs based on the extent to which their most important tasks are AI-friendly. Those are what we call “highly exposed to AI” – basically, the top 25% of occupations where AI could make the biggest impact.
The kinds of tasks that fall into that AI-friendly category include things like gathering information, processing data, scheduling, controlling machinery, and handling administrative work. These are all activities where AI tools are already pretty capable. Although, it’s worth noting that this doesn’t mean other jobs aren’t affected by AI at all, just that the “high-exposure” jobs we’re looking at could be especially susceptible to change given what people do day-to-day.
Brooke: Workers in lower-income households account for 20% of all workers with relatively high AI exposure. What did your analysis reveal about the types of occupations and industries where lower-income workers are most exposed, and were there any patterns that surprised you?
Elizabeth: As might be expected, AI-exposed workers tend to cluster in computer-based jobs where these kinds of tools can be used to help or automate tasks. For instance, Office and Administrative Support jobs topped the list for AI exposure, regardless of worker income. But it’s striking that, while about one in three AI-exposed workers overall are in office and administrative jobs, fully half of lower-income workers with high AI exposure work in those occupations, which include jobs like secretaries, administrative assistants, receptionists, and office clerks.
When you zoom in on those lower-income office and administrative workers, the leading industry they work in is Healthcare and Social Assistance—think of administrative assistants in doctors’ or dentists’ offices or in community food or housing organizations. In keeping with that finding, it also stood out to us that these workers are much more likely to be employed by nonprofits: About 21% are, which is more than double the average for AI-exposed lower-income workers overall.
Brooke: Because AI is digital, we sometimes think of it as something that will impact workers without regard for “place.” But your research found that AI exposure among workers in lower-income households varied considerably across the country. What do you think is behind that variation?
Elizabeth: It really comes down to the fact that every place has its own mix of jobs, industries, and workers. And that shapes who’s exposed to AI and how much.
For example, lower-income workers make up a larger-than-average segment of AI-exposed workers in states like Delaware and Virginia but a smaller-than-average share in states such as Nevada and North Dakota. And the types of AI-exposed jobs also look different depending on where you are.
To shed more light on these variations across the country, we recently released an interactive data tool that lets users explore data on the types of jobs and workers with high AI exposure in each state.
Brooke: Your qualitative research underscored that workforce development must focus on giving people skills to work with AI, complementing the technology with human skills. What are those key human skills?
Elizabeth: Critical thinking was routinely raised as the number one skill. It has long been in demand by employers, but our respondents—who included leaders from workforce and training organizations, community colleges, and other nonprofits and employers—expect it to become even more essential as AI becomes part of everyday work. They see those critical thinking skills as especially important for helping workers use AI tools effectively and make good decisions about when and how to rely on them. And they noted those skills can help equip their clients, students, and employees to be flexible and able to work alongside AI, rather than narrowly developing skills tied to one specific constantly changing technology.
Brooke: Stakeholders also expressed that workers in low-income households will likely need additional support beyond AI skilling to transition to an AI economy. What kinds of support did they elevate?
The digital divide came up a lot. The stakeholders from workforce and training organizations we spoke with are worried that workers who lack strong digital skills or don’t have reliable internet access will fall even further behind as AI tools become standard. And they noted that, while these tools will eventually become widely adopted, right now we’re in this tricky transition period. They were grappling with the question of how to help people adapt, especially workers who aren’t in school or formal training programs but are already seeing AI impact their jobs—or even losing their jobs because of it.
Several participants suggested the need for additional supports and resources to help workers deal with AI-related job disruptions. They noted that kind of assistance could not only help lower-income workers navigating AI-related job losses, but it could also support higher-earning workers who might face downward mobility as their roles change or disappear.
