Philosophical foundations

[Gerry Wolff]

Freddy Offenga wrote:
> 
> Gerry Wolff wrote:
> >Regarding "symbolic" and "connectionist" approaches to AI etc, I feel
> >that much of the confusion arises because neither term is very precise.
> 
> With "symbolic" I refer to the symbolic computational view of cognition.
> In this classic approach the symbols have a certain meaning. They are
> about something in the external world.
> In the "connectionist" view the knowledge is represented in the weights
> of the network. The 'symbols' have a distributed representation
> (although there are also localist networks, where the units have
> meaning). Anyhow, the weights are the most important properties for
> connectionism while symbols are the most important for symbolicism.

My use of the term 'symbol' is probably a bit non-standard: in the SP
scheme, it is simply a 'mark' that can be matched in yes/no manner with
other symbols and has no hidden meaning. As I said in another message,
any meaning which an SP symbol has must take the form of other symbols
which occur next to or close to the given symbol in the same way that a
dictionary definition of a word is written next to the word it defines.

...
> >...
> >
> >* A neural net might very well provide a foundation for an SP machine.
> >This could be an interesting and fruitful exercise which might help to
> >remedy some of the weaknesses of the current generation of neural nets
> >in performing some kinds of 'symbolic' task.
> >
> >...
> 
> I agree with the 'exercise' part, but I'm not sure I understand what
> you mean with the first sentence ('A neural net...').
> Do you mean that a neural net could be the right way to implement
> the SP system? If this isn't necessary, why do it? I was thinking
> about a system with other elementary processes. These processes
> should include the basic properties of CasC.
> Ofcourse the question about what the best implementation is remains.
> 

A possible reason for implementing an SP machine as a neural net is
that, with the right high-parallel hardware, a neural net could be a
very fast and effective way to search. There are probably other
high-parallel methods of searching and choosing between them and a
neural-net style is not something we can sensibly do without fairly
detailed proposals.

Best wishes,

Gerry