Perhaps I want what someone else considers "trash" or "shock content." Certainly, forty years ago that would have included John Waters' works.

I remember when other people were deciding that a given work was indecent, or unChristian, or obscene. I wasn't a fan of that. Sub in "fascism" or "doubleplus ungood," I don't care. I'll just rely on the Orwell when I repeat his bit how "The word Fascism has now no meaning except in so far as it signifies 'something not desirable' ..." by, well, someone else.

Person-to-person works to the extent that you know a lot of people who have a good understanding of your tastes. I remember the bad old days where you had to know someone who was aware you were into some particular thing and could direct you to a 'zine. I'm told that I am fairly good at recommending horror films once I understand someone's taste, but not a lot of people can do that for me.

Are a lot of algorithms currently crap? Hell yeah! But if you care to stare at the painful whys, they emerge in each case. Netflix doesn't have an recommendation algo problem, it has a "catalog too small and packed with Netflix originals and if we just slight of hand hard enough you won't notice" problem. Amazon has a similar problem to eBay: a skillion shady resellers. Music's hard and probably always will be hard. The Music Genome Project, if it were automatable, would be a lot better than Spotify pushing a mixture of what it has that it wants to trend and what someone else who liked that track also liked.

These are hard problems, because with art, you're often looking for something like that last thing you had, but with some kind of variation. It's not like hamburgers, where you can build entire chains on the dependability of a hamburger experience. Still, we can get somewhere decent, but for other factors distorting the results.