What’s Dragging Down A.I. Efficiency? The ‘Verification Tax.’ - The New York Times

The article describes that work with three terms: a "verification tax," "botsitting" and "rework." The article presents the three terms as descriptions of the same work. According to the piece, this checking function puts a very human bottleneck on AI automation. The article states that the task is often arduous.

The article cites Torsten Slok, chief economist at Apollo, who said in a review of San Francisco Fed data this week that "the productivity payoff from A.I. remains a forecast rather than an observation." The review of San Francisco Fed data is described as having taken place this week. The article places Slok's statement alongside its description of the verification work, rather than in a separate section. The excerpt does not give a figure for the number of people performing this checking work, or the time it consumes.

The article's framing is that automating office tasks does not remove the need for human review of AI output. That review, per the article, covers whether the output is accurate, safe and usable. The piece describes the checking work as a distinct category of labour that sits alongside the automated tasks. Slok's assessment frames the productivity gains from AI as forecast rather than observed. The article names no sources other than Slok and the San Francisco Fed data it cites, and provides no figure for the volume of verification work.