Daniel Rocha <danieldi...@gmail.com> wrote:

I am aware of these tidbits. But,Language is mechanical and have universal
> rules that are quite strict.If you feed it with massive quantities of data,
> it will eventually get it right . . .
>

I disagree. Google was not making much progress with that approach. The
ambiguity problem is unlikely to be solved with that approach alone. But
even if that is true, the multi-level neural network approach is not
narrow. It is being applied to wide range of problems with outstanding
success. Especially compared to previous AI. It beat the world champion in
go, which the experts thought would not happen for many years. It does a
better job of pattern recognition even with fewer sample patterns to start
with. As I mentioned with the cat example, it can abstract the essence of
shapes and synthesize a recognizable image that a human recognizes as a
cat. It can synthesize a graphic image that summarizes the distinguishing
charactoristics of the object. That is amazing, and far more complicated
than just distinguishing one object from another.

This article shows the abstract image of a cat generated by a Google
program. It is eerie:

https://www.eetimes.com/document.asp?doc_id=1266579

http://www.nytimes.com/2012/06/26/technology/in-a-big-network-of-computers-evidence-of-machine-learning.html

Some aspects of AI are probably still far distant, such as a sentient
computer. But there is no denying that rapid progress is being made, and
the main reason now is the multi-level neural network. The previous boost
was from massive databases. In real world applications the two approaches
are blended together, and other techniques are being used.

- Jed

Reply via email to