Become a fan of Slashdot on Facebook

 



Forgot your password?
typodupeerror
×

Comment Re:Stop calling it AI. (Score 1) 78

The formal logic approach is still the only one that has a theoretical possibility of creating some aspects of true intelligence.

Sort of. We actually do have fairly robust theorem solvers written in prolog, but thats not enough. Intuitively, "true AI" works like extracting formal logic theorems out of huge set of before/after data fed to a blackbox.

Just like humans do something intuitively at first, with some degree of success, but when they find rational backgrounds (with help of formal logic rigor) behind that intuition, it gives significant accuracy boost. The two work in tandem - formal rigor is toothless when facing the totally unknown, but can explain it after intuitive models are trained and it can feed its hypotheses into them in lieu of farmed data.

Trouble is that layered NN camp ("intuitive") and formal logic camps are still too separated. But corporate interests will force merger to a degree.

This is most visible in speech recognition, and more recently, vision where formal grammar models sit above low level intuition NN, or better said, directs training of layers so it can work with less data and reason about unknown inputs because it actually "understands" what's going on on a formal level.

Comment Re:They reall don't mean this (Score 1) 78

Yes and no. Fun stuff is still done, especially in terms of computer vision and hearing - there are numerous stakeholders.

However zero interest in actual research, it's all engineering.

They do not want code monkeys, as you still do need a PhD-level knownledge to even design this kind of stuff, just like computers in the 60s. Thats why companies are bought and people are hired only if they have something useful to show already in applicable field.

Comment Re:I'll believe it when... (Score 1) 78

It's not that bad, but the gist is similiar.

1) Megacorp allows to run your models on their data, and you can write paper about it
2) But your results are not reproducible, because well, private corporate data

Solution indeed is to put the training sets in the public domain.

Comment Re:Stop calling it AI. (Score 1) 78

Quite the opposite, strong AI was the previous boom/bust.
This time we're using "sheer brute force" to train NNs in a bid to get emergent intelligence-like patterns, instead of using formal logic constraint solvers.

Comment Re:Price won't come down (Score 1) 317

Don't forget about valence electrons which affect net ion charge you can transport in a chemical bond - it's 1 Li vs 3 Alu. The reason why alu sucks weight wise is first and foremost unfavourable chemistry for cathode counterpart, not atomic number.
Alu does have interesting properties, though:
http://www.nature.com/nature/j...
While it indeed have a magnitude less specific capacity, these cells can serve as interesting interim in place of ultra capacitors (regenertive braking etc). Li reactivity is both blessing and curse.

Comment Re:Keeps the Russians out (Score 1) 164

Pretty much. Combined with Snowden, US has come out of its closet and said "fuck this shit about moral facade". While nobody likes russians in the region (as historically their imperialism was still far worse), europeans are now between two grinding stones - one completing an encirclement, and second having a good pretense to "feel threatened". Interesting times ahead.

Remains to be seen if NATO members signed the faustian deal with lesser of the evils.

Slashdot Top Deals

The moon is made of green cheese. -- John Heywood

Working...