I do not like repeatedly moving the goal posts. However, I have to say that
I have come to the conclusion that Watson and Deep Learning are narrow. 40
or 50 years ago I thought that Encyclopedic Knowledge would have to follow
Human-Level Knowledge - and I still think it does. It would take human
level knowledge to be able to understand an encyclopedia more thoroughly.
However, some information can be expressed economically in a form which
strongly associates facts or derivations with certain types of questions or
computational problems. On the other hand human beings are not be able
to understand everything that can be read in an encyclopedia. And there is
no question that these kinds of programs are doing kinds of thinking. In
retrospect it makes sense that computers would achieve encyclopedic fact
searching accuracy before human level thinking so I guess I just did not
foresee the WWW in the old days (even though I did hear about such
possibilities a long time ago.)

>From the language used in the IBM websites about Watson there was very
little to suggest that Watson-Jeopardy used something that could be likened
to deep learning. Their reference to Deep NLP seems to refer to something
more of a search than a series of hierarchal applications of a network of
some sort. However this question is irrelevant to my topic and the argument
that Watson did not use Deep Learning in Jeopardy is irrelevant to my point
of view that I think the programmers could find things in Watson that
are directly comparable to Deep Learning. I don't think the boundaries of
what Deep Learning is have been completely set yet just because so many
types of network methods can be used in Deep Learning.

So I am pretty doubtful that Watson + Deep Learning is going to be able to
think outside of the box. I believe that my discovery that searching for
'cat in a box' in Google Images turned up some pretty impressive results
but that the search for 'cat getting out of a box' did not was symbolic for
the contemporary state of the art.



-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to