[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: Sun, 19 Apr 2015 00:11:36 +0000 (UTC)
Message-id: <1334251028.7133565.1429402296288.JavaMail.yahoo@xxxxxxxxxxxxxx>
John Sowa wrote:
1. Rule-based systems and connectionist systems have very different
    goals and solve very different problems.  They are complementary
    rather than competing.

I have to disagree. 

Jerry Fodor, of Language of Thought fame, believes connectionism to be a challenge to the idea that ideas are embodied as discrete representations in the brain, and that transformations (inferences, etc.)  are embodies as discrete processes in the brain. He believes that there is a "language of thought", and that this is required to explain both the compositionality and productivity of thought. Most work in cognitive science is still based on this paradigm.

Connectionism -- "neural net AI -- has progressed far beyond its roots in the 70s, and is presented by such supporters as Paul and Patricia Churchland as a correct (albeit currently primitive) account of how we think. To them, Jerry's LOT is a formalization of what they call "folk psychology", which they regard as equivalent to earlier beliefs in phlogiston -- a set of beliefs that will not be refuted, but will simply wither away as it becomes less and less plausible as connectionist science evolved (much, as Thomas Kuhn persuasivly argued, the geocentric theory of the solar system was never refuted, but instead became less and less plausible as time went on). (All of this oversimplified, of course.)

Both Fodor and the Churchlands regard the rules vs. connectionism approaches to be in direct competition with one another. A substantial body of literature exists which assumes that they are in conflict (e.g. are concepts represented locally in the brain, or in a distributed manner).

For anyone interested, I can provide references which will lead into this literature. Or, they can follow the references in the IEP (Internet Encyclopedia of Philosophy) article on Connectionism, and several of the articles in the SEP (Stanford Encyclopedia of Philosophy), such as the articles on Mental Representation and the Language of Thought.

On Saturday, April 18, 2015 2:28 AM, John F Sowa <sowa@xxxxxxxxxxx> wrote:

On 4/17/2015 6:58 AM, Thomas Johnston wrote:
> Is Watson rules-based AI or connectionist  (neural network) AI?

The IBM research project for the Jeopardy! challenge put together
a wide range of technologies for many different tasks that have
to work together.  For any AI paradigm X, if you ask "Is Watson X?",
the answer is either yes, some versions are, or it could be.

In any case, there are many issues involved:

  1. Rule-based systems and connectionist systems have very different
    goals and solve very different problems.  They are complementary
    rather than competing.

  2. Among other things, IBM has also designed a chip with 5.4 billion
    transistors that is specialized for implementing the so-called
    "neural networks" -- which aren't really a realistic model of
    actual neurons.  For a brief summary of the technology, see

  3. You can't implement a complete system with today's version of
    neural networks.  Even the IBM chip can only serve as one
    component of a much larger system.  If it were plugged into
    Watson, it could perform many functions quite well, but there
    are many other kinds of AI tasks that would be better handled
    by more conventional software.

I discussed some related issues in the following slides:

    Why has AI failed?  And how can it succeed?

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>