A regular computer program is written by someone. It goes wrong, you blame the coders. "Machine learning" is written by an omnipotent blank
Conversation
Replying to
pretty right - generally actions get harder to attribute the less legible the path between you and the results is
Quote Tweet
good luck I'm behind seven moral agents responding to incentives
1
1
2
a free moral agent in your chain introduces a lot of illegibility, or alternatively accepts responsibility (they "could've prevented it")
1
1
our machines are not quite full moral agents yet but they're getting there
There's moral philosophy being done in this direction - see the idea of "moral luck" en.wikipedia.org/wiki/Moral_luck
1


