News for hackers. Stuff that matters.
Re:The Well-Poisoning Machine by mnemotronic on Thursday March 23, 2023 @09:35PM (#35283960)
Ok. I'm dense. I can't understand what this is about. Is this a threat?
Re:The Well-Poisoning Machine by jacquesm on Thursday March 23, 2023 @09:37PM (#35283968)
I read it as a way to express the frustration about the speed with which AI tools are unleashed without proper consideration for the consequences 'just because we can'. We can't allow a well poisoning machine gap to come into existence.
Re:The Well-Poisoning Machine by evouga on Thursday March 23, 2023 @10:11PM (#35284199)
Seems like a flagrant strawman to characterize AI as a “well-poisoning machine.”
Re:The Well-Poisoning Machine by jacquesm on Thursday March 23, 2023 @11:21PM (#35284685)
Let's wait and see. I can see plenty of ways in which that could come true.
Re:The Well-Poisoning Machine by 1attice on Friday March 24, 2023 @12:35AM (#35285135)
It's not a strawman, it's an analogy.

In order to be a man of straw, it would have to claim that AI "just is" a well-poisoning machine (in the literal sense of a machine that inserts poison into a hole in the ground containing H2O).

Now, that alone does not mean that it goes through; the analogy works as an intuition pump, and those can be flakey (see https://en.wikipedia.org/wiki/Intuition_pump)

But nor does it mean that a fallacy has been committed

Re:The Well-Poisoning Machine by tpoacher on Friday March 24, 2023 @09:44AM (#35288988)
Not AI as a whole. Particular activities to do with AI which would generally be deemed not only dangerous, but also not having any clear non-malicious purpose.

So if there was a strawman argument here at all, it's literally the one you just made!

Re:The Well-Poisoning Machine by 1attice on Thursday March 23, 2023 @09:40PM (#35283993)
This is about Sam Altman types rushing to win the ML race. We can reasonably expect civ ('the village') to be disrupted ('poisoned'), but instead of that counting as a reason to not do it, it counts --- somehow --- as a reason to do it harder, so we do it first.
Re:The Well-Poisoning Machine by dgacmu on Thursday March 23, 2023 @10:28PM (#35284323)
OpenAI is, seemingly, trying to have it both ways: on one hand, they use warnings about the dangers of AI to justify their existence and practices such as keeping models closed, etc. And on the other hand, they're rushing towards commercialization in order to beat out their rivals.

It's a bit of a BS act. Either it's bad and we should slow development way down until our frameworks for dealing with it catch up, or it's not going to turn us all into paperclips.

(I'm one of those boring folks who think there's a lot of risk but not in the paperclip sense but in the "when we use tools we understand poorly, we are likely to get outcomes we may not desire in terms of fairness, trustworthiness, etc.")