ontolog-forum
[Top] [All Lists]

Re: [ontolog-forum] Watch out Watson: Here comes Amazon Machine Learning

To: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: Thomas Johnston <tmj44p@xxxxxxx>
Date: Fri, 17 Apr 2015 10:58:19 +0000 (UTC)
Message-id: <1298560556.6048447.1429268299598.JavaMail.yahoo@xxxxxxxxxxxxxx>
John,

Is Watson rules-based AI or connectionist  (neural network) AI? 

As you probably know, connectionism has come a long way since the Rumelhart and McClellan days. I've been especially interested in what I've read (very little, actually) about SOFM - self-organizing feature maps, which learn without back-propagation. 

I may have heard that Tuevo Kohonen, one of the originators of SOFM, is now employed by Google. If Watson is primarily rules-based AI, I would be very interested to see what a connectionist Watson could do. Maybe, one of these days, Google will show us!

Tom



On Thursday, April 16, 2015 7:28 AM, John F Sowa <sowa@xxxxxxxxxxx> wrote:


Peter,

I'm sending this response to Ontolog Forum.

> Came across this article [on April 10]:
> http://www.zdnet.com/article/watch-out-watson-here-comes-aws-machine-learning/

Some comments:

  1. It's important to distinguish the research system called Watson,
    which was used to play Jeopardy!, from the commercial products
    that IBM sells under the Watson umbrella.

  2. Machine learning (of various kinds) is important for both.

  3. But the research system also used a wide variety of AI techniques,
    and many of them do not seem to be used in the commercial products
    from IBM (at least not yet).

(But I left IBM many years ago, and I have no information about the
details, other than the publicly available reports.)

  4. Generic observation:  There are many kinds of ML methods, and
    some AI systems use them as integral parts of other methods.

  5. The cost of "10 cents per 1,000 predictions" may seem cheap for
    "big" predictions, but it might be prohibitively expensive for
    a system that integrates learning with certain kinds of NLP
    methods or reasoning methods.  Such systems might be learning
    and predicting with every word of a text or other data item.

John

_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J




_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J    (01)

<Prev in Thread] Current Thread [Next in Thread>