[From Bill Powers (930909.0915 MDT)]
Michael Fehling (930909.0731) --
I am suggesting to accept Newell's challenge to biuld a UTC
(unified theory of cognition). I.e., show how the PCT
community _is_ capable of constructing (at least through
several prototypes) a model of a "PCT agent." This PCT agent
model should embody most, if not all, the control levels, the
same model should be able to perform a wide range of
experimental and real-life tasks (i.e,. from target tracking,
through memory performance, to problem solving and decision
making).
I'm not sure the PCT community can do this any better than Newell
or others can (could have). We don't have any better mental
equipment than they do, and our resources are a lot smaller. Of
course you, with your experience, resources, and background of
developments, might get a lot farther than any of us could, given
a basic grasp of the PCT model and a handle on the kinds of
phenomena it defines.
As Rick pointed out, in Rick's way, we have to start with the
phenomenon that control theory explains: control. This means
finding out what people actually control, starting as near to
scratch as necessary. I'm not satisfied that the levels as now
defined are stated correctly; we haven't even proven
experimentally that such levels exist, particularly at the higher
levels. For example, I've defined the "program" level as a
network of choice points. That's a long way from explaining human
thought or how it works in ongoing control processes. It's clear
that if we model this level as a digital computer, we can then
propose any sort of program we wish to run in this computer, so
"explaining" cognition then just becomes a matter of writing
programs that will solve problems, make decisions, etc., in the
same manner that a hardware computer would do these things. We
just say that the same programs are running in the brain.
This doesn't explain how we are able to write these programs,
either for a brain's computer or for a mainframe. It simply
demonstrates some of the activities that might be carried out by
such a computer, if indeed such a computer exists in the brain
and once the programming is done. I have great faith in and
admiration of clever programmers; I assume they can write a
program that will accomplish any task that we can define clearly,
three different ways. But that still doesn't explain to me how
they can be doing what they're doing. And it doesn't
automatically explain what it is they're trying to accomplish by
doing it. How do they conceive of the task before they've started
writing the program? That, it seems to me, is where we would
begin to get into what the brain, rather than a program running
in a brain, is doing.
I am, of course, very tempted to join in the challenge you
propose making. But I think that all these Grand Universal
projects that people want to get into assume that we know a lot
more than we do about the simple facts of behavior. This is what
got old-fashioned AI into a bind. Instead of starting by
observing what people do and redefining the problem, they wanted
to get busy with the fun part and start explaining things -- even
things that don't happen. I feel very insecure about going that
route. I'd like to have some justifications for my premises, and
I don't think they have been worked out yet.
Maybe this is just a matter of temperament or age. Who am I to
tell anyone else not to go for the gold? They might get lucky,
who can say?
On the other hand, a lot of people have tried it, and there are a
lot of abandoned projects along the road from 1945. It just seems
to me that science doesn't progress that way, by trying to jump
way ahead and guess how it's all going to come out. We have a new
tool and a new concept of behavior. In simple applications we
have established some pretty solid facts that aren't going to be
overturned easily, for a long time. We're gradually extending the
scope of what we know -- from one-person tracking to two, three,
and four-person tracking. We've explored bits and pieces of
control activities that seem to involve what look like higher
levels of control, but only piecemeal. Every control experiment
we've tried so far works and gives good data.
But the foundation is building up very slowly. I don't blame
anyone for wishing that we could say SHAZAM! and expand the
hierarchy to encompass all of human behavior in the same solid
way, but we can't. It's especially frustrating when we can see
all those other guys with their millions of dollars for hardware,
with people lining up to lend a hand, with all their flashy
claims and ambitions, with their confidence that they're way out
in front in some sort of race. We're just plodding along doing
one thing at a time, we hope like the tortoise and not like the
sloth. But I'll bet -- I am betting -- that we'll still be
plodding along with our little pile of high-quality facts
continually growing long after all those hares have forgotten
where they were going.
Of course our work would be going along far faster if as many
people were working at it as are working in AI and other fields.
The world is full of wonderfully clever people who are good at
modeling and thinking up experiments, who could rip off three
good experiments a week and add ten new positively-identified
controlled variables to the list every month. We're in need of a
data base that can only be built up in that way, a data base in
which all the things that are needed for a PCT analysis have been
noticed, written down, and reduced to parameter measurements.
We'll have it, eventually, but it would be nice to have it while
I'm still here to admire it.
Best,
Bill P.
ยทยทยท
----------------------------------------------------------------
[From Mary Powers 9309.09]
To Lewis Larue (9309.07)
Thinking about Solomon's decision - whoever was the real mother,
he seems to have decided that a woman willing to cut a child in
two was likely to be a less fit mother than one who would let the
child go so that it would live. Whoever the real mother was.
This contrasts with the recent case in which the little girl was
returned to her birth mother. Apparently cutting the child in two
emotionally was not a consideration in the law, nor a concern of
the birth mother, so that case went the other way. (I realize it
was dragged out for two years; if it had been settled promptly,
it would not have been such an anti-Solomonic decision).
As for bowels yearning vs wombs stirring: part of a new mother's
discovery that she loves her baby so much it hurts is the fact
that it indeed does hurt, since the uterus contracts when she
nurses. This can be painful sometimes, but before the point of
pain it feels very good, as it should, since it's restoring one's
female plumbing to a reference state. The control systems
involved in having this hooked up to nursing must be pretty wild.
Anyway, the combined impact of the sensations of bowels yearning
and womb stirring adds up to an emotion call mother love, which
is generally considered to be the top kind.
Mary Powers
P.S. to Ed Ford. The picture came. Marvellous! Thanks!!