Benjamin Goertzel wrote:
But each of these things has a huge raft of assumptions built into it:
-- hierarchical clustering ... OF WHAT KIND OF SYMBOLS?
-- hash coding ... OF WHAT KIND OF SYMBOLS?
-- sequence completion ... OF WHAT KIND OF SYMBOLS?
In each case, Granger's answer is that the symbols are vaguely
behaviorist units playing an incredibly simplistic role in a simplistic
system.
If we take his claims at face value, he has found library functions that
operate on junk that cannot possibly be "symbols" at a cognitive level.
If he had simply said that he had found hiererchical clustering of
neural signals, or hash coding of neural signals, or sequence completion
circuits at the neural signal level, I would say good luck to him and
keep banging the rocks together.
But he did not: he made claims about the cognitive level, and the only
way those claims could be meaningful and useful would be in a cognitive
level system that is manifestly broken.
Well, I don't fully agree with your final paragraph...
Suppose we take Greenfield's hypothesis that a fundamental role in
cognition, perception and action is played by transient neural assemblies,
that form opportunistically based on circumstance, but that are centered
around "cores" that are tightly-interconnected neural subnets ...
Potentially, Granger's primitive mechanisms could act on sets of neural
signals coding for these "cores", which then indirectly drive the cognitive
activity that occurs mainly on the level of the transient assemblies that
the cores induce...
Fair enough: but much depends on specifics from Greenfield.
This is *not* what Granger says, but it seems generally plausible to me...
BTW I am curious to hear something about what you think might be
a correct cognitive theory ;-)
There is a sketch of some aspects of it in the Loosemore & Harley paper.
I'll take your expression of interest to be sufficient excuse for me
to send you a copy of it.
I am sure you would like to know more, but I am keeping further details
behind a proprietary screen for the moment.
Richard Loosemore
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=8660244&id_secret=56378838-51313d
