It’s especially easy to imagine happening implicitly: feed everything into an ML system and if it tags it with something like “(sex noises)” or “(moaning)” (which Hollywood subtitles and other things in someone’s training data probably have) that’s searchable without anyone explicitly setting out to build a system.
I'd assume they are more likely to get valuable data from people talking about their sex life, if companies are actively trying to monetize in that way
Oh, sure. The point is just that nobody has to put a business goal in to directly create that feature for it to be exposed by a general purpose classifier.
