Does AI exposure hit low-wage workers harder in some fields?

The paper's central finding is about the shape of AI exposure, not its average. Using "a novel dataset that links occupational characteristics to measures of AI exposure," the authors find that in male-dominated occupations exposure is "generally concentrated in higher-skilled and higher-paid occupations," while female-dominated occupations "display relatively uniform levels of exposure" across both high-skilled, high-paid and low-skilled, low-paid work. The frame is LLMs and "broader AI technologies," and the question is who sits where on the exposure distribution once occupations are split by gender composition and by position in the skill and wage distribution.

The reasoning the paper gives runs from distribution to vulnerability. Lower-skilled, lower-paid workers are "typically more vulnerable to labour-market disruption" because of weaker bargaining power, lower access to training and reskilling, and greater employment insecurity. If female-dominated occupations carry exposure evenly into that lower tier, a larger share of the exposure lands on the workers least able to absorb it. The authors add, "combined with existing evidence from the literature," that women in these positions are "disproportionately exposed to forms of AI associated with task automation and job restructuring, rather than productivity-enhancing augmentation," and that AI adoption "may place women in lower-paid and lower-skilled occupations at greater risk of displacement, constrained wages and career progression." The framing in the introduction is that AI may "unintentionally" reinforce existing occupational segregation.

This sits on a different axis from Does concentrated AI exposure enable workers to adapt and reallocate?, where concentration means few tasks within a firm are affected. Here the distribution is across occupations ranked by skill and pay, and the point is who holds the exposed positions. It also adds a "who" to firm-level evidence such as Do firms substitute labor for AI at different rates?: that note shows substitution varies across firms, and this paper says the workers behind the exposure are not evenly placed either. The reskilling-access point connects loosely to When does AI actually boost worker productivity?, since both make the route from exposure to benefit depend on training. Does incremental AI replacement erode human influence over society? is the society-scale version of the worry, with no gender dimension.

The excerpt does not say how exposure is measured, how "male-dominated" and "female-dominated" are defined, which countries or years the data cover, or how large the differences are. It measures exposure, not displacement, so the claim that women in lower-paid roles face automation rather than augmentation rests on "existing evidence from the literature," not on the dataset itself. The excerpt also breaks off at "an additional contribution of this work is the distinction between different forms of AI" without saying what that distinction yields. What the evidence supports is a narrower claim: exposure is distributed differently across the two groups of occupations, and that pattern is a reason to check outcomes by gender rather than assume a uniform effect.