Carlo Pizzinelli is an economist in the European Department of the International Monetary Fund. Marina M. Tavares is a Senior Economist in the Climate Change Structural Reforms Division of the IMF’s Research Department.
Artificial intelligence (AI) is set to expand rapidly across a wide range of industries in the coming years, reshaping the labor market while raising concerns about job displacement alongside expectations of productivity gains. At the same time, advanced economies, including the United States, are undergoing a significant demographic shift toward older populations. An important question for policymakers, academics, and other stakeholders in the retirement field is whether AI will ultimately enable older workers to remain productively employed later in life.
Our recent research offers a nuanced picture. We find that, on average, older workers are better positioned than younger workers to benefit from AI. Yet this advantage comes with an important caveat: older workers are also less able to adjust when disruptions occur, which may leave them vulnerable during periods of rapid structural change.
Understanding Exposure and Complementarity to AI
Assessing how AI may affect workers requires a conceptual framework for evaluating its likely impact on a given occupation. Here, two concepts are useful: AI exposure and AI complementarity. Exposure captures the extent to which the tasks in an occupation can be performed by AI tools. Several recent studies propose methods to measure exposure, and we follow the approach developed by Felten and colleagues. Complementarity, introduced in our earlier paper, measures whether AI is more likely to augment workers’ productivity or substitute for their labor.
While AI exposure captures the technical feasibility of applying AI to job tasks, AI complementarity reflects features tied to social preferences and ethical or regulatory considerations, including responsibility for others, independent decision-making, social interaction, and physical presence. Occupations with both high exposure and high complementarity (HEHC) are those in which AI is more likely to function as a supportive technology, enhancing productivity and potentially wages rather than fully replacing workers. In contrast, jobs with high exposure but low complementarity (HELC) face a greater risk of displacement.
This framework provides a practical tool for assessing not only the likelihood that AI will be deployed in a given occupation, but also the potential consequences for the workers in those roles.
The figure below illustrates the relationship between AI Occupational Exposure (AIOE) and complementarity. AIOE measures the extent to which a job’s tasks are exposed to or can be affected by artificial intelligence. Occupations with higher complementarity tend to cluster at moderate levels of AIOE. Roles such as judges and lawyers exhibit both high exposure and high complementarity, whereas positions such as telemarketers and legal assistants combine high exposure with low complementarity.
Where Older Workers Are Concentrated
Using U.S. microdata, we document that older workers are employed in HEHC occupations to a larger extent than their younger counterparts, after accounting for differences in education. These jobs include managerial, professional, and technical roles where experience, judgment, and interpersonal skills remain essential. Meanwhile younger workers, particularly those with a college degree, are more likely to be employed in HELC occupations. These are also predominantly white collar jobs, but they generally entail lower levels of responsibility and independence in decision making.
This pattern suggests that many older workers are positioned to benefit from AI-driven productivity gains, provided they can acquire the skills needed to work effectively alongside new technologies. Moreover, AI-complementary jobs often have features such as greater flexibility and the potential for remote work that align well with the preferences of older workers and may encourage longer careers.
The figure below shows how the distribution of workers across occupation types changes with age, separately for those with a high school education or less (left panel) and those with a college degree or more (right panel). The three lines represent high-exposure, high-complementarity (HEHC), high-exposure, low-complementarity (HELC), and low-exposure (LE) jobs. Among workers with a high school education or less, employment is heavily concentrated in low-exposure (LE) occupations at all ages, although this share declines modestly over the life cycle. At the same time, the share working in HEHC occupations rises gradually with age, while HELC employment remains relatively flat at a lower level. A different pattern emerges for college-educated workers. At younger ages, they are more likely to be employed in HELC occupations, but this share declines steadily over time. In contrast, the share in HEHC occupations increases sharply with age and eventually becomes the dominant category. Employment in low-exposure occupations decreases early in the career and then stabilizes at a lower level.
Overall, the figure indicates that as workers age, especially among the college educated, they tend to move out of more easily replaceable, low-complementarity roles and into positions where AI is more likely to complement their skills rather than substitute for them.
The Vulnerability Beneath the Surface
Despite this favorable employment composition, the paper highlights an important source of vulnerability. Older and middle-aged workers exhibit lower labor market fluidity: they are less likely to change occupations, industries, or locations in response to shocks, and they tend to experience longer unemployment spells following job loss.
As is indicated in the figure below, among workers with a high school education or less, employment is heavily concentrated in LE occupations at all ages, although this share declines modestly over the life cycle. At the same time, the share working in HEHC occupations rises gradually with age, while HELC employment remains relatively flat at a lower level. A different pattern emerges for college-educated workers. At younger ages, they are more likely to be employed in HELC occupations, but this share declines steadily over time. In contrast, the share in HEHC occupations increases sharply with age and eventually becomes the dominant category. Employment in low-exposure occupations decreases early in the career and then stabilizes at a lower level.
Notwithstanding older workers’ greater concentration in HEHC occupations, a sizeable share of them remains employed in HELC jobs. Their reduced adaptability means that labor market disruptions such as AI-driven task reorganization within firms can have outsized effects on displaced older workers. While younger workers may respond by switching careers or retraining, older workers often face higher adjustment costs, including shorter time horizons to recoup training investments and greater barriers to hiring.
Overall, the figure indicates that as workers age, especially among the college educated, they tend to move out of more easily replaceable, low-complementarity roles and into positions where AI is more likely to complement their skills rather than substitute for them.
Policy Implications for an Aging Workforce
These findings point to a clear policy tradeoff. On the one hand, AI has the potential to support longer and more productive working lives by complementing the skills of older workers. On the other hand, without targeted support, AI-driven restructuring could amplify existing vulnerabilities, aggravating the pressures of an aging population.
Several policies can help reap the benefits of AI for older workers while providing support to those most adversely affected. First, lifelong learning and mid-career training are essential to ensure that workers can acquire the skills needed in AI-complementary jobs. Fortunately, being equipped for AI does not necessarily mean becoming a computer scientist. In the vast majority of cases, it will require gaining AI literacy; that is, acquiring the basic skills to be a responsible user of AI in the context of one’s own profession, understanding its potential as well as its limitations and risks.
Second, active labor market policies including job matching, retraining subsidies, and hiring incentives can help reduce the costs of displacement for older workers. Third, strengthening income support and unemployment insurance during transitions can mitigate the risk that late-career job loss translates into premature or involuntary retirement.
As AI continues to diffuse across the economy, the challenge for policymakers is not simply to prevent job loss, but to ensure that technological progress translates into longer, healthier, and more productive working lives.
Views of our Guest Bloggers are theirs alone, and not of the Pension Research Council, the Wharton School, or the University of Pennsylvania.
