Back in ~2004, when I was looking for a job for the industrial placement year of my degree, one of the options was a nut packing factory using computer vision systems. I was intrigued, but went for the job processing satellite images instead.

Even well before that, ML is very closely related to statistics, so early practical applications would have been as simple as gathering data points on widget production and doing the kinds of analyses that are now backed into free spreadsheet software.