Replying to
@Eeveecraft@dragonscave.space @cwebber@social.coop @_elena@mastodon.social @timnitGebru@dair-community.social @emilymbender@dair-community.social @alexhanna @DAIR@dair-community.social @jaredwhite@indieweb.social @tante@tldr.nettime.org
That 'help detect cancer better' doesn’t actually work all that well though, as it worsens doctors’ ability to detect/confirm cancer, gets worse over time as it can’t adapt to subtle outside changes, and has huge issues with biases. So while it might be slightly better for the standard-case of an average white man, it’s worse for everyone else.
Also at least your 'remove pause from audio' example sounds way more suited for relatively simple algorithms than statistical models to me.
(I do agree though that there are some cases where things that would be called 'AI' nowadays can actually be good)