[From: C. Love (920916.9:30)]
[To: B.Powers (920914.15:30)]
I have been thinking about your imagination model more
Bill and now feel that I better understand what your after.
This came about after a brainstorming session with a fellow
contractor here at DCIEM, with whom is also interested in
both my work and PCT, etc. Your imagination model involves
three levels. The top level (n+1) is the trigger level. It
tells it lower level (n) neighbors(s) to begin imagining
because it hasn't been *happy* for sometime! (This could
mean persitently bad error, who knows?) Anyways, in turn,
these level (n) ECS' (how many - one, two, all of them???)
recieve a signal (how is it received - details???) saying
that it should begin imagining. Ok, so now these level (n)
ECS' tell all the level (n-1) ECS' to short-circuit their
error signal (output) into their percept output, i.e., the
signals that are treated as percepts by the level (n) ECS'.
This is clear to me now, and I hope to everyone who reads
this.
Now what happens. We get this closed loop involving three
levels. THe objective: AS I see it, it is to satisfy the
ECS's that requested imagination to occur - thhe level n+1
ECS's. Ok, what I want to happen here is for something
better to occur as a result of this process. How can that
happen unless those percept weights are changed, right?
Which percept weights you ask? The percept weights connected
to the ECS' in the (n+1) level. Why these? Because it is
these weights that are causing the problems since we must
assume that the reference signals are correct, therefore only
our perceptions are wrong, and since we haven't been able to
*learn* the right perception to see, it must appear that
something like conflict exists. This means that the only
OTHER way of resolving this is to bias the percept weights.
The solution is to change the weights, either more negative
or more positive. By the way when I referred above to the
reference signals I mean the ones entering into the (n+1)
level ECS(s). The refernce signals must be relied on as being
correct or this entire argument falls apart.
We can't say that we must change one vble (percepts)
according to the references if the references are just
as unreliable - we'll end up going in circles and never
get anywheres. At the highest level, at least, the
references are those conceptual objectives, "match
tarhet to finger", which we KNOW to be correct. As such
we therefore must assume that all lower level ECS'
will also be receiving like objectives in their own
way. It is the percepts that must be changed to
perceive the right signals to match the references.
Bill, about his almighty ECS. I think you misunderstood
me. I don't wish to place lungs into my ECS model. THe ECS
model may have all these capacities but the model is set
up in a manner that the ECS will only use those methods
that it should know about. Remember that I'm doing object
orientated programming and all these *methods* are held
within my version of the *ECS* object. Many different
methods may be there, but may not always be used. I
think of it as a person having the ability to do anything
but that person only does a subset of those *possible* tasks
due to their environmental restrictions, whereas if another
person (object instance) is created and placed in a different
environment, this person will be able to do a different
set of tasks. Do you follow me?
About he Hebbian stuff? This is more just shots in the dark
right now. I'm not going to defend any of this just yet.
I'm floundering around as much as anyone else in search
of good methods to achieve control in such a model.
Suggestions/criticsim is always welcome because if you
learn from it, it is worth listening to.
Best,
Chris. (DCIEM)