I am not so sure that there is no upper limit to intelligence. I am pretty sharp and looking at myself and other people of comparable intelligence, this seems linked to the ways in which we are weird. People smarter than me are even more weird. Much more may be dysfunctional.
-
-
Replying to @miniver
I agree completely; I was putting words in the mouth of a generalized rationalist.
1 reply 0 retweets 0 likes -
-
Replying to @miniver @Meaningness
It’s funny, for a long time I imagined that one possible form for AI would be a sort of Reason Machine partner to human intelligence. A system which could do stuff like read every published scientific paper and check which ones held water. That is looking a lot less plausible.
1 reply 0 retweets 1 like -
Replying to @miniver @Meaningness
Machine learning is giving us systems whose logics we are LESS able to inspect than our own minds!
1 reply 0 retweets 4 likes -
Replying to @miniver
Yes although I think DL systems mostly are understandable. It’s that proponents don’t want to do that because it takes away the illusion of magic.
1 reply 0 retweets 0 likes -
Replying to @Meaningness
They are understandable like logic is understandable; they are not understandable like proofs are understandable.
1 reply 0 retweets 0 likes -
Replying to @miniver @Meaningness
There's been some work on explainability for deep learning systems, but AIUI it's still preliminary, and the field is *mostly* focused on "look, new shiny! Who cares how it works?" rather than understanding what's really going on.
2 replies 0 retweets 1 like -
Replying to @pozorvlak @miniver
> There's been some work on explainability for deep learning systems Mostly has the general defect of work in this area: it proceeds by analyzing the structure of the mechanism, rather than the interaction of the mechanism and the problem.
1 reply 0 retweets 5 likes -
“What extra hair can we add to the NN that will force it to explain itself” rather than “let’s look at what it’s actually doing, and then it will probably be obvious how it’s doing that.”
1 reply 0 retweets 0 likes
E.g. I tentatively concluded in January 2015 that image classification is mostly using texture, which has since turned out to be true. I could do that because I understood the problem space (having done machine vision in other styles years before).
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.