Re:Google’s new AI ethics board is already falling apart
by naasking on Thursday April 04, 2019 @09:25AM
(#19571984)
Exactly. Moreover, we could never be sufficiently confident in an argument of a position until it's been thoroughly excoriated by people who vehemently disagree with each step or even the basic premises. That's how you get watertight proofs, and that's what we would need for AI ethics.
An ethics board isn't a technical committee working to patch AI when they become unethical. They are a mechanism to flag ethical concerns _before_ there is concrete evidence of a problem. That means they will have to function more like a debating panel for management and interested observers than a technocratic mechanism.
1) A part of what is being done when discussing ethics is modeling what the arguments are to deal with bad positions. That is much easier to do if someone on the panel takes up a bad position. Watching a bad argument get dismantled is more useful then hearing 3 respectable people agree on an obvious point.
2) Even if an idea is wrong, being radically different is often enough to trigger some exploration of very different perspectives on a problem, some of which will be valid. People aren't usually wrong because they are wrong about everything they've ever thought or done ever. People are usually wrong because they've got a few key parts of a larger picture wrong, but many other details are still correct.
3) There is always a risk that, no matter how obvious the situation is, a group has been contaminated by groupthink. As I like to say, intelligence is no defense against groupthink; it just means that the individuals involved are better at rationalising it.