AI exposure is not a jobs forecast

Start with what the measure describes

An occupation can be exposed to a technology without disappearing. Exposure describes the relationship between tasks and capabilities; an employment outcome also depends on adoption, demand, organizational choices and the distribution of new work. Turning one measure into the other skips the mechanism the research is trying to understand.

The IMF staff discussion note Gen-AI: Artificial Intelligence and the Future of Work examines exposure alongside complementarity and distributional effects. The distinction matters: a tool that can perform part of a job might substitute for work, increase a worker’s output or change which skills are valuable. These possibilities call for different evidence.

Ask how gains reach people

Broadening the Gains from Generative AI shifts the question toward fiscal policy. Read it as an examination of policy choices around a changing distribution of income and opportunity. A policy proposal is different from evidence that an intervention has already produced its intended result.

For a specific sector, the next research step is to connect task changes to observed hours, earnings, output and hiring, with an explicit period of comparison. A company’s adoption announcement supplies only one part of that record. The effects on workers require their own observations.

Put adoption in its institutional setting

The OECD Digital Economy Outlook provides a broader technology and policy context. The AI Index offers another entry point for tracing changes across research, performance, investment and governance. These collections are useful for identifying a question and its relevant indicators, rather than collapsing all indicators into a single score for social progress.

NIST’s Generative AI Profile adds a risk-management perspective. A voluntary framework can help structure an inquiry into deployment practices. Its existence does not show that a particular organization has implemented it, and implementation does not by itself prove a labour-market effect.

A research agenda that can be tested

Our synthesis is to follow the chain from capability to adoption, task redesign and measured outcomes. Keep the point of observation explicit. Compare groups only when the work, period and measurement are sufficiently alike. Look for evidence of complementary work as well as displacement. The result is a question that new data can answer, rather than a prediction produced by relabelling an exposure estimate.