Productivity Figures Hint at an AI Effect. Economists Urge Caution.

Every technology that promises to change the economy eventually meets the productivity statistics, and the meeting is usually disappointing. Computers were everywhere in the 1980s, as the economist Robert Solow observed, except in the productivity numbers. In this illustrative scenario, the latest figures gave AI's advocates something to point to and its skeptics plenty to question.

Output per hour in professional services and finance, the two sectors where surveys find the heaviest use of AI tools, grew faster over the past year than in the economy as a whole. The gap was large enough to be noticed and small enough to be explained away.

Three explanations, one data series

The first explanation is that AI is doing what its developers claim: letting the same workers produce more. The second is composition. If firms stop hiring junior staff, output per remaining hour rises even if nobody works differently. The third is timing. Both sectors had above-average productivity growth before the tools arrived.

The data cannot yet distinguish among these. Sector productivity figures are revised for years after first publication, and the revisions are often larger than the differences being discussed.

What would settle it

Economists point to two kinds of evidence that would be more persuasive. Firm-level data matching AI adoption to output over several years would separate the composition effect from genuine gains. And studies that count the time spent reviewing and correcting AI output, not only the time saved producing it, would show whether the gains survive contact with quality control. An independent group revised its own estimate downward this week after adding review time to its model.

Until then, the responsible reading of the figures is the boring one: something is happening, it is probably positive, and it is smaller and less certain than either side would like.