AI executives warn that machines will replace human workers across nearly every profession within years. Economists say the data needed to evaluate that claim barely exists. The missing variable: price elasticity of demand — a foundational economic concept that determines whether AI-driven productivity gains expand markets or simply eliminate jobs.
The exposure metric problem
The dominant framework for assessing AI’s labor market impact relies on “exposure” scores — task-by-task analyses of how much of a given job an AI system could theoretically perform. A widely cited 2023 paper by Tyna Eloundou and colleagues at OpenAI estimated that roughly 80% of the U.S. workforce could see at least 10% of their tasks affected by large language models. Separately, researchers at Princeton, the University of Pennsylvania, and NYU built on O*NET occupational data to map AI exposure across professions, finding that highly paid, white-collar roles — legal services, financial analysis, programming — scored among the most exposed.
These figures have become the currency of workforce anxiety. But according to economists who study labor markets, exposure scores measure the wrong thing entirely. As MIT economist David Autor has argued, knowing that AI can perform a task tells you nothing about whether the humans currently doing that task will lose their jobs. The U.S. government has maintained a task catalogue containing thousands of individual job tasks for over two decades, and researchers routinely use it to calculate exposure scores. But exposure is a measure of theoretical capability, not economic consequence.
The variable that actually matters
The critical question is what happens after AI makes a service cheaper and faster. If AI cuts the cost of legal document review by 60%, does demand for legal services expand enough to absorb displaced paralegals — or does the sector simply shrink its workforce? That depends on price elasticity of demand: how much consumer demand increases when prices fall. And as Yale’s Budget Lab has cautioned, economists lack comprehensive data on price elasticity across most professions. Without it, every prediction about AI job displacement is essentially a guess decorated with task-level statistics.
Consider two industries where elasticity data does exist and where the implications for AI are sharply different. In tax preparation, demand is largely inelastic — nearly every employed American who needs to file taxes already does so. When TurboTax and similar software dramatically cut the cost of basic returns, the sector didn’t see a surge of new customers; it saw a reduction in paid preparers. AI-powered tax tools are likely to extend that pattern, compressing the workforce further rather than growing the market. Contrast that with graphic design, where platforms like Canva had already demonstrated high price elasticity before generative AI arrived. As design costs fell, demand exploded: small businesses, solo entrepreneurs, and nonprofits that previously couldn’t afford professional visuals began producing them at scale. AI image generation tools are now amplifying that dynamic, expanding the total market for design services even as per-unit human labor declines. Same technology thesis, opposite labor outcomes — and the difference is entirely determined by elasticity.
AI leaders speak in certainties; economists see a data void
The contrast between industry rhetoric and economic preparedness is stark. Anthropic CEO Dario Amodei wrote in his October 2024 essay “Machines of Loving Grace” that AI could broadly substitute for human cognitive labor within a narrow window of years, with transformative effects across medicine, science, and governance. Economist Daron Acemoglu of MIT, who won the 2024 Nobel Prize in Economics partly for his work on technology and labor, has warned that rapid automation without institutional adaptation could suppress wages and erode career ladders for a generation. Meanwhile, the economists tasked with evaluating these competing visions acknowledge they lack the empirical foundation to adjudicate between them.
Building that foundation would require what Yale Budget Lab researcher Zak Mustapha has described as a massive coordinated data collection effort — extending the kind of granular price-and-demand tracking that currently exists for grocery items (via supermarket scanner data partnerships) across the entire service economy. No government or institution has yet undertaken anything close to that scale.
The structural gap
The pattern is familiar: technology companies set the narrative timeline for disruption while the institutions responsible for managing its consequences operate without adequate information. As we’ve previously explored, the absence of this data creates a policy vacuum that defaults to industry framing — where displacement is treated as inevitable rather than conditional on measurable economic dynamics.
Without price elasticity data, policymakers are designing workforce strategies based on exposure metrics that leading labor economists consider inadequate. The result: significant investments in retraining programs and policy interventions guided by metrics that measure theoretical capability, not actual economic outcomes. The data that would clarify the picture exists in fragments — scattered across supermarket scanners and academic pilot programs — while the decisions it should inform are being made now.
Feature image by Vitaly Gariev on Pexels
