...because previously, the academics would use an army of undergrads to do the same data labeling that ML accomplishes.

(The dis-economy of scale hurts less if you're already starting from a point with the manual labor.)