The One Piece of Data That Could Actually Shed Light on Your Job and AI
AI / AI Policy | 4 min read
Within Silicon Valley's orbit, an AI-fuelled jobs apocalypse is spoken about as a given. The mood is so grim that prominent voices — including Dario Amodei, who has described AI as a general labour substitute for humans capable of doing all jobs within five years — are amplifying widespread worker anxiety. Yet even economists who previously cautioned that AI had not yet cut jobs and may not cause a cliff ahead are coming around to the idea that it could have a unique and unprecedented impact on how we work. The trouble is that the tools for predicting what that impact will actually look like are, by the assessment of the economists themselves, pretty abysmal.
"Exposure alone is a completely meaningless tool for predicting displacement. We need, like, a Manhattan Project to collect this."
— Alex Imas, Economist, University of Chicago
Why "AI Exposure" Is a Misleading Metric
The dominant framework for assessing AI's impact on jobs has been task exposure — the idea that any job is made up of individual tasks, and the more of those tasks that AI can perform, the more "exposed" that job is to displacement. The US government chronicled thousands of these tasks in a massive catalogue — the O*NET database, first launched in 1998 and updated regularly since — and it became the data source researchers at OpenAI used to judge how exposed a job is to AI. Anthropic later used the same data in its analysis of millions of Claude conversations to see which tasks people are actually using AI to complete and where the two lists overlapped. A real estate agent, for example, was found to be 28% exposed. But Alex Imas, an economist at the University of Chicago, argues that this framework produces illusory insight. Exposure tells you which tasks AI could theoretically do — it tells you nothing about whether doing those tasks more efficiently will increase or decrease demand for the workers who perform them.
The Real Question: Price Elasticity of Demand
The most illustrative example from Imas is a software engineer building premium dating apps. AI coding tools allow that engineer to produce in one day what used to take three. The employer, spending the same amount of money, now gets more output. So what happens next — does the employer want more employees or fewer? The answer depends entirely on what happens to demand for the product when AI-driven efficiency lowers its price. If millions more people want premium dating apps at a lower price point, the company grows and hires more engineers. If demand barely increases — if people who don't use premium dating apps simply won't want them even at a lower price — fewer coders are needed and layoffs happen. The economic concept governing this is price elasticity of demand — how much demand for a good or service changes when its price changes. And repeated across every job with tasks that AI can perform, this is, as Imas frames it, the most pressing economic question of our time. The specifics of price elasticity will vary by industry, by product, by customer base — and we are currently operating in the dark on all of them.
The Data Gap — and the Case for a Manhattan Project to Fill It
Price elasticity data does exist for some goods. Grocery items like cereal and milk have well-documented elasticity figures because the University of Chicago partners with supermarkets to collect data from price scanners. But the equivalent data for tutors, web developers, dietitians, lawyers, customer service agents, and the vast majority of service economy jobs that AI is now being applied to simply does not exist in a compiled, accessible form. Such data is occasionally scattered across private companies and consultancies, but not in a way that researchers can use to build economy-wide models of AI's labour market impact. Imas's call is not just for current AI-exposed roles: fields that are not exposed now will become exposed in the future, so the data needs to be tracked across the entire economy — prospectively, not retrospectively. The investment required is real, but Imas argues the payoff would be equally real: for the first time, economists would have a realistic basis for predicting how AI-driven productivity gains translate into job creation or job loss in specific sectors, and policymakers would have a foundation on which to build an actual plan — rather than operating entirely without one. As the article notes, no lawmakers have yet articulated a coherent plan for what comes next.
Key Takeaways
- • Even economists who previously cautioned that AI had not yet meaningfully cut jobs are reconsidering, acknowledging that AI could have a unique and unprecedented impact on how we work. But the tools for predicting what that impact will look like — particularly the dominant task exposure framework used by OpenAI and Anthropic — are, by the economists' own assessment, insufficient to actually predict displacement.
- • Task exposure (the O*NET-derived framework showing what percentage of a job's tasks AI can perform) is a misleading predictor of displacement because it says nothing about what happens to demand for those workers when AI drives down the cost of the output they produce. Exposure is only decisive in the extreme case where every task is AI-replaceable at a cost below the worker's wage — which is not guaranteed given the cost of reasoning models and agentic AI.
- • The key variable is price elasticity of demand — how much demand for a product or service increases when AI-driven productivity allows its price to fall. If demand rises strongly, companies may hire more workers. If demand is inelastic and barely increases, fewer workers are needed and layoffs follow. This plays out differently in every industry and is currently unknown for the vast majority of AI-exposed service economy roles.
- • Price elasticity data exists for consumer staples (cereal, milk) because supermarkets share scanner data with researchers. It does not exist in compiled, accessible form for tutors, web developers, dietitians, lawyers, customer service agents, or the majority of knowledge and service economy jobs that AI is now being applied to. The data is occasionally scattered across private companies and consultancies — not available to researchers for economy-wide modelling.
- • Alex Imas (University of Chicago) calls for a "Manhattan Project" to systematically collect price elasticity data across the entire economy — not just currently AI-exposed roles, but all fields, since roles not currently exposed will become exposed over time. Without this data, economists cannot build realistic models of AI's labour market impact and policymakers have no foundation for planning. No lawmakers have yet articulated a coherent plan for what comes next.