This is a repost of an article I sent out around the time we lost our
Internet feed. I don't know if it was successfully distributed or not,
but since I haven't seen any replies, I'll assume the worst and repost
it. If anyone has replied, it did not make it to DCIEM. My apologies if
you receive two copies of this.
-- Allan Randall
···
----------------------------------------------
Hi all,
Although there have been delays, I am happy to say that I am now working
full-time on Martin Taylor's control project. I will be working in
conjunction with Chris Love and Jeff Hunter.
Anyway, my first order of business is to get more properly grounded in
control theory, so I have just finished giving "Behaviour: The Control Of
Perception" a thorough reading (BTW, my formal background is in computing
and AI).
I am intrigued by the model of memory presented in Chapter 15. Bill, is
this still your working model, or has it changed? (I know its been a while
since you wrote the book). Do you still see the basic function of memory
as providing reference signals? Do you see the imagination loop as
operating through memory as it does in the book? Or have the intervening
years changed your conception of the problem? My conception of the diagram
on page 221 is that the memory unit would take references from the
higher level ECSs and remember how the world is supposed to act, so the
memory is basically a little toy world inside the ECS (a toy world, that
is, for all the ECSs above it that feed it references). This is a neat
model, I think, especially with the switches that give the four
different modes of behaviour. But I also realize there are probably many
other spcific ways of implementing memory that do not conflict with the
basic HPCT scheme, so I am wondering how you feel about this model now,
twenty years later.
I had previously thought of imagination mode as one ECS telling upper
levels that they are controlling, even though they aren't. However,
talking with Martin recently, he pointed out the need to imagine that a
reference has NOT been satisfied, as well as imagining that it had, I
realized this model can do this as well. If the memory box is acting as
a little toy world, then it can respond in imagination mode that
"no, this won't work." -- the higher level ECS will experience prolonged
error in imagination.
I also found it interesting that you used the RNA model of memory. Have
you dropped this idea since the RNA theory has fallen out of favour?
Allan Randall, randall@dciem.dciem.dnd.ca
NTT Systems, Inc.
Toronto, ON