> From: Linketscher, Nadja <nbl196@soton.ac.uk>
>
> Cognitive science proposes three different approaches to explain the
> mind: computation, analogy and neural nets. Both computation and
> analogy are based on the assumption that there are mental
> representations, which consist of symbols or images.
Close, but not quite:
It's not analogy but "analog processes." These include the proximal
images cast on your retina by distal objects, plus all the further
analog processing that goes on in your brain, plus any neural nets that
might be operating on those analog images.
> According to my
> understanding, they differ in that computation manipulates symbols
> using algorithm (operates on shapes, meaning is irrelevant),
Any resemblance between the shape of the symbols and the shape of the
things the symbols are about: that's what is irrelevant.
> whereas
> the latter modifies images and related to it their meaning.
No meaning yet with analog structures and processes. (Is there
any meaning in a sun dial? Yet its shadow is an analog of the position
of the sun, hence of the time of day.
For meaning you need to put all this (and more, surely) together.
No one has done that yet!
> Neural
> nets, on the other hand, operate on the basis of back propagation.
Yes some nets do use backpropagation, but that is not relevant to
the analog/symbolic distinction: Neural nets can operate on analog or
symbolic input.
> This
> concept suggests that people learn to act intelligently on the world by
> trial and error and feedback (operant conditioning). All three models
> can be simulated using a computer. Is that correct?
Nets and analogs can be simulated computationally. Computation itself
need not be, because it is already computational!
No, we certainly don't learn EVERYTHING by trial and error and feedback.
Some of it is innate, and some of it we learn from words (symbols)
rather than sensorimotor trial and error -- We can, for example
learn "zebra" from words alone -- as long as the words are grounded
are ultimately grounded bottom-up.
This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:50 GMT