Designers of ML-based systems often aspire to reach “human-level” performance on complex tasks. It’s a useful aspiration in some cases, but a helpful alternative framing is to identify tasks which normally feel like {tedious drudgery} to a human, and to ask what happens if {you make those “free.”}
Transcribing an interview is drudgery. If an ML system can do a mostly-good-enough job for me, I might be much more inclined to edit that interview, to make connections between interviews, etc.
Sorting out junk mail is plain drudgery. If an ML system can do it for me, I no longer need to worry about “protecting” my email address, which makes me more willing to use online services, unlocking new types of business.
This is a consistent theme in the history of technology. For example, the washing machine and the refrigerator were powerful enabling forces for women, in unexpected ways.
Q. What’s an example of a tedious task an ML system has made “free”?
A. e.g. classifying junk mail
Conversation with David Holz, 2020-05-09