The Debate Over the Future of Labor Has an Evidence Problem

At the center of the debate over whether AI will replace jobs lies the concept of exposure, which indicates how much a given occupational group is affected by AI. Currently, many policymakers cite figures from the 2023 paper 'GPTs are GPTs,' but this involves serious data inconsistency issues.

The metric has three major limitations:

• A gap with the pace of technological progress: The scores were calculated based on early-2023 GPT-4 level capabilities, but AI technology today has advanced far beyond that.
• Regional bias: Since it is based on the U.S. Department of Labor's occupational classification system, it is difficult to apply directly to labor markets in other countries.
• Oversimplification of work units: By breaking down jobs into individual task units for measurement, it fails to capture the complex context of actual work.

As these limitations of the metric accumulate, a gap emerges between AI's actual impact and policy responses. Researchers and policymakers need to treat workers as partners rather than mere objects of analysis, and build more dynamic, actionable evidence.