Ben Goertzel wrote:
Sure, I know Pylyshyn's work ... and I know very few contemporary AI
scientists who adopt a strong symbol-manipulation-focused view of
cognition like Fodor, Pylyshyn and so forth. That perspective is
rather dated by now...
But when you say
"
Where computation is meant in the sense of abstract symbol
manipulation according to rules. 'Rules' means any logic or calculii
you'd care to cite, including any formally specified
probablistic/stochastic language. This is exactly what I mean by COMP.
"
then things get very very confusing to me. Do you include a formal
neural net model as computation? How about a cellular automaton
simulation of QED? Why is this cellular automaton model not "abstract
symbol manipulation"?
If you interpret COMP to mean "A human-level intelligence can be
implemented on a digital computer" or as "A human level intelligence
can be implemented on a digital computer connected to a robot body" or
even as "A human level intelligence, conscious in the same sense that
humans are, can be implemented on a digital computer connected to a
robot body" ... then I'll understand you.
We're really at cross-purposes here, aren't we?...this is a Colin/Ben
calibration process.... :-) OK.
By COMP I mean any abstract symbol manipulation at all in any context.
The important thing is that in COMP there's a model of some kind of
learning mechanism being run by a language of some kind or a "model of a
modelling process" implemented programmatically. In any event the
manipulations that are occuring are manipulations of abstract
representation of numbers according to the language and the model being
implemented by the computer language.
But when you start defining COMP in a fuzzy, nebulous way, dismissing
some dynamical systems as "too symbolic" for your taste (say,
probabilistic logic) and accepting others as "subsymbolic enough"
(say, CA simulations of QED) ... then I start to feel very confused...
I agree that Fodor and Pylyshyn's approaches, for instance, were too
focused on abstract reasoning and not enough on experiential learning
and grounding. But I don't think this makes their approaches **more
computational** than a CA model of QED ... it just makes them **bad
computational models of cognition** ...
Maybe a rather stark non-COMP example would help: I would term non-COMP
approach is /there is no 'model' of cognition being run by anything./
The electrodynamics of the matter itself /is the cognition/. Literally.
No imposed abstract model tells it how to learn. No imposed model is
populated with any imposed knowledge. No human involvement in any of it
except construction. Electrodynamic representational objects are being
manipulated by real natural electrodynamics... is all there is. The
'computation', if you can call it that, is literally maxwell's equations
(embedded on a QM substrate, of course) doing their natural dynamics
dance in real matter, not an abstraction of maxwell's equations being
run on a computer....
In my AGI I have no 'model' of anything. I have the actual thing. A bad
model of cognition, to me, is identical to a poor understanding of what
the brain is actually doing. With a good understanding of brain function
you then actually run the real thing, not a model of it. The trajectory
of a model of the electrodynamics cannot be the trajectory of the real
electrodynamics. for the fields inherit behavioural/dynamical properties
from the deep structure of matter, which are thrown away by the model of
the electrodynamics. The real electrodynamics is surrounded by the
matter it is situated in, and operates in accordance with it.
Remember: A scientific model of a natural process cuts a layer across
the matter hierarchy and throws away all the underlying structure. I am
putting the entire natural hierarchy back into the picture by using real
electrodyamics implemented in the fashion of a real brain, not a model
of the electrodynamics of a real brain or any other abstraction of
apparent brain operation.
Does that do it? It's very very different to a COMP approach.
cheers
colin
-------------------------------------------
agi
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/
Modify Your Subscription:
https://www.listbox.com/member/?member_id=8660244&id_secret=117534816-b15a34
Powered by Listbox: http://www.listbox.com
