Amanda Askell · Anthropic · 2020-12-14
Askell drags AI ethics out of vibes-mode by reintroducing a basic distinction philosophers have used for centuries: pro tanto reasons versus all-things-considered judgments. Yes, a system causes harm X. So does the alternative. So does doing nothing. Without comparing the full set, you haven't actually said anything about what to do.
This is the antidote to the genre of AI ethics writing that catalogs harms and stops there. Short, surgical, and unusually direct about a failure mode endemic to the field. A learner who internalizes this argument will read every subsequent ethics paper with sharper eyes.
> These are all pro tanto harms... it just tells us about one particular harm the system causes.