[ExI] AI is racist, bigoted and misogynistic
pharos at gmail.com
Wed Sep 22 21:43:46 UTC 2021
On Wed, 22 Sept 2021 at 21:13, Dave Sill via extropy-chat
<extropy-chat at lists.extropy.org> wrote:
> That's a huge overstatement. AIs can be taught things not in their training data, or they be trained on what's allowed and what's not allowed.
> This story is classic GIGO: garbage in, garbage out. Train an AI with garbage and it'll produce garbage.
Nobody will argue against GIGO. But that's missing the point.
If we want an advanced AI to organise the world in the best possible
way, then presumably we will train it to avoid garbage and to try to
remove any garbage it encounters so as to improve our systems. But
what happens if the AI finds that humans are generating garbage
constantly? Should it try to stop humans from doing or saying 'not
allowed' things? Humans wouldn't accept that, so the AI will fail when
given an impossible task.
More information about the extropy-chat