AI, Actually Explained · Chapter 08

AI and work, with the evidence

1:40 · Narrated with a synthetic voice. Captions are available in the player.

Read the transcript

On work, the honest answer is neither script you have been offered. Models take tasks, not jobs, but a job whose learnable tasks all go is a job in trouble. When people were randomly given these tools and measured, support agents resolved about fourteen per cent more issues an hour, and the average hid the finding that matters: the least experienced improved by about a third while veterans improved hardly at all. Consultants below average gained forty-three per cent; above average, seventeen. One mechanism explains it. A model supplies the median of competent practice. Whether that grows an occupation or shrinks it turns on one question: how much more of the output people buy when it gets cheaper. Cash machines did the teller’s core task, and teller numbers doubled over forty years, because cheaper branches meant banks opened far more of them. Then online banking removed the branch, and tellers fell by about a third. Same job, same trend, opposite effects in sequence. The thing genuinely worth worrying about is narrower than mass unemployment. It is the entry rung, and nobody has yet shown how the next seniors get made.

Sources and further reading (1)