Original title: The Measurement Revolution? Credible Measurement and Inference in the Age of AI
Melissa Dell, Ashesh Rambachan · August 2026
Economists often study things that are hard to turn into numbers, such as ideas in text or details in images. Today, artificial intelligence can convert this messy data into structured variables cheaply and at scale. However, this creates a new challenge. Researchers no longer struggle to find a single workable metric. Instead, they must choose between many plausible options that can lead to completely different conclusions. The authors explain three stages where AI enters the measurement pipeline: discovery, constructThe abstract idea a researcher wants to measure, like 'consumer sentiment'. definition, and observation. To keep findings credible, researchers must validate their measures using explicit criteria rather than informal guesses. Crucially, the authors show that properly designed validationChecking whether a measurement tool produces correct numbers by comparing it to known examples. samples allow for trustworthy conclusions even if AI predictions suffer from arbitrary bias.
You want to measure how clean a room is, but checking every spot is impossible. So you ask a few friends to each rate the room from their quick glance. They give different scores, and you don't know whom to trust. In this paper, AI models are like those friends, turning messy observations into ratings, and validating them is like having a trusted inspector check a few rooms to see which friend's ratings are most reliable.
Economists can now convert messy data like tweets or satellite images into clean numbers at low cost. But this creates a new problem: many plausible measures exist, and they can lead to conflicting conclusions. This review lays out a framework for validating AI-generated measurements so that the resulting economic claims hold up to scrutiny.
This is a review, so it does not provide new empirical results or specific validation algorithms. It offers conceptual guidance but acknowledges that the problem is more difficult when a random validation sample is unavailable, without fully resolving that case.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
The Measurement Revolution? Credible Measurement and Inference in the Age of AI — Melissa Dell, Ashesh Rambachan
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.