3 ms·
We should stop calling deep learning models AI, it gives unnecessary hype and misunderstanding of the capacities of the things, as well as a feeling that resear
by make3 6y ago
We should stop calling deep learning models AI, it gives unnecessary hype and misunderstanding of the capacities of the things, as well as a feeling that researchers aren't being honest about the capacities of the things they are developping
- whatshisface 6y agoAI hasn't meant human-level general intelligence for decades now. I think that ship has sailed, if it was ever at port.
- qsort 6y agoML is part of AI, even if you are going by purely academic definitions. I absolutely agree that universal approximator != intelligence, but we're just arguing semantics, here.
- bopbeepboop 6y agoI’m curious what you mean. Wouldn’t a class of universal approximators necessarily include some intelligent ones — for any measure of intelligence? There’s obvious scaling issues, though.
- qsort 6y agoThe entire question is predicated on how you define "intelligence". Personally I subscribe to the view that we should avoid using the word altogether when referring to computers. Sorry to disappoint, but those questions are bottomless pits of debating semantics.
- bopbeepboop 6y agoYou personally believe we shouldn’t use the word “intelligence” in these contexts? Okay — then why even make the comment in the first place? You clearly meant something by “intelligence != universal approximator”, but you’re unwilling to provide any semantics to back that up. Go ahead and define it: how do you mean that statement? I think you’re just too embarrassed to admit you said a personal view that’s wrong.
- qsort 6y agoI'm going to apply the HN rulebook and assume you are in good faith even though I'm almost certain that you are not. I'm using "intelligence" in the sense of "thing that a layman who doesn't know what compact convergence or lebesgue integrable means would recognize as intelligent". A layman would not call such an agent, by itself, intelligent. I'm advocating avoiding the use of "intelligence" precisely to avoid this kind of conversation.
- bopbeepboop 6y agoI am replying in good faith. Can you give an example of a "thing that a layman who doesn't know what compact convergence or lebesgue integrable means would recognize as intelligent" which can’t be modeled by universal approximators? Not just due to scaling. I think you’re advocating avoiding the use of “intelligence” as a duck of your comment about intelligence and approximators just being wrong.