Teaching system concepts; Brooks & PCT

[From Bill Powers (920629.0930)]

Hank Folson (920629) --

Your post on various ways of handling knowledge/ignorance about control is
most thoughtful. It might point toward some educational reforms (as long as
everyone's talking about them again, for the election year).

The jurors are all Category 1 people, unaware we are all control >systems

who will control our perceptions. They have no understanding of >levels,
especially those involved here, which include Category, >Principle and
System Concept. So they have no awareness that they are >choosing the way
they perceive the police and the drug dealer/victim. >If they did, they
might be able to separate their own System Concept >from a more generalized
and higher concept of how their society should >handle this situation, and
choose which way to go. It is interesting >how open they all were in
describing what they were controlling for.

It isn't really necessary to educate everyone about control theory. If HPCT
is right, everyone has the necessary levels of control and perception.
What's needed is a set of experiences in which they can see the basis for
justice and constitutional rights. I can imagine role-playing in school
situations where students are shown how ANYONE can be falsely accused and
suffer prejudice. You don't have to be black or poor to be treated
according to prejudices about a group to which you belong. This can happen
to salespeople, policemen, rich people, Polish people, actors, garbagemen,
psychologists, computer programmers, teachers, and students. Statistical
treatment of test results formalizes this kind of prejudice, but it happens
to everyone, or could.

What's important is to recognize that any person can suffer unjust
accusation and bad treatment when they're identified as members of a
population instead of being treated as individuals. The shared system
concepts under which we agree to live have to be designed to work no matter
which side of an accusation you find yourself on. If teaching this idea
were made the core of courses in civics, maybe people wouldn't grow up to
be jurors who don't get the point.

Had these jurors been raised in an environment in which truth and >justice

were very important aspects of their System Concepts, their >perceptions of
the evidence would have been different.

AMEN.

···

-----------------------------------------------------------------------

Technically Sweet (920628) -- (what were your parents thinking of?)

Welcome to CSGnet, TS. As Rick Marken said, we're aware of Brooks'
subsumption architecture, but he doesn't seem to think there's anything
interesting for him in control theory.

An addendum to Rick's remarks. The main thing missing from Brooks' approach
is an appreciation of what any control problem looks like from inside the
controlling system. It's all very cute to build a robot that will go around
collecting pop cans (although I understand that descriptions of this
device's behavior are extremely overblown). But to do this, there has to be
a designer who knows a hell of a lot about the properties of pop cans,
environments, physics, and so on. When you see behavior as the
accomplishment of a bunch of objective "tasks," you can give the behaving
system a lot of help -- laser range-finders, for example. You can place its
sensors so they are just right for responding to things you know are
critical to the task, and build into its computers all the knowledge it
will need about properties of its environment. Most of what makes these
little robots at all successful resides outside the robots, in the PhDs of
the designers.

If you put yourself inside a robot, everything disappears except what your
sensors tell you. That's the only world there is. Suddenly there aren't any
objects, or people, or motions, or even spatial attributes of reality. All
you have are sensor signals. Whatever you accomplish has to be defined in
terms of those signals, not in terms of their implications in a world that
a human observer can experience. The effects of your outputs are known only
in terms of the way the set of sensor signals changes. There's nobody to
tell you what's really happening outside you when you generate an output.
All you know of the consequences of acting is that your sensor signals
change. If you're going to make any sense of anything, it can be done only
in terms of the effects of your output signals on your input signals. What
goes on out there between output and input is hidden from you.

The problem with the task-oriented approach is that it's all about what
happens OUTSIDE the robot, as seen by a sophisticated human observer and
not as seen by the robot. The human designer can set up the robot so that
when it does things inside its own perceptual world, something of interest
to the designer happens outside the robot's ken, as a side-effect. I think
this is a misconception of what robot design is really about -- it's just a
way for a human being to accomplish something simple, of significance to
the human being, the hard way. Nothing has really been learned about the
design of autonomous systems.

Perhaps you will agree with this: what Brooks and others following similar
paths don't realize is that they themselves ALREADY live in a Virtual
Reality (VR). They are inside a brain that knows only what its sensors tell
it -- the outputs of all those sensors ARE the world, as far as they or
anyone else can tell. They only way they know anything about the physical
effects of their own actions (or anyone else's) is to examine their own
perceptions. Failure to realize this is the reason behind their rejection
of (or failure to grasp) the PCT principle of behavioral modeling. And it's
the reason behind their task-oriented approach to modeling behaving
systems.

As a person interested in VR, you probably see very clearly that behavior
has to be organized around making perceptions behave in particular ways,
not around "objective" effects of actions. When you set up a virtual
reality and display it on a person's retinas, you're simply creating a
different link between the person's actions and the person's (visual)
perceptions. You can put any properties you like into this link, and
eventually a person will learn to control the resulting perceptions by
generating the required actions. It's only a tiny step to realizing that
this is how natural behavior works, too. People know no more about the real
physical links between actions and natural perceptions than they know about
your computer programs for creating apparent links in a Virtual Reality.

By the way, the use of a computer running Lisp for controlling simple
behaviors is overkill of the most extreme degree. A lot of very interesting
behavior could be accomplished with a control system consisting of two or
three operational amplifiers costing about 50 cents apiece. The world has
gone digital and has apparently forgotten that there's a much easier (and
far faster) way to accomplish analog tasks.
----------------------------------------------------------------------
Best to all,

Bill P.

[From Oded Maler 920630]

[From Bill Powers (920629.0930)]
..

Perhaps you will agree with this: what Brooks and others following similar,
paths don't realize is that they themselves ALREADY live in a Virtual
Reality (VR). They are inside a brain that knows only what its sensors tell
it -- the outputs of all those sensors ARE the world, as far as they or
anyone else can tell. They only way they know anything about the physical
effects of their own actions (or anyone else's) is to examine their own
perceptions. Failure to realize this is the reason behind their rejection
of (or failure to grasp) the PCT principle of behavioral modeling. And it's
the reason behind their task-oriented approach to modeling behaving
systems.

I think you misinterpret Brooks a little bit. As a roboticist he is
surely aware that the robot's world is just the perceptual space of
its sensors. I think all his work in the last years drove AI-based
robotics much closer to the lines of HPCT. I *really* recommend reading
his "Intelligence without Reason" [IJCAI91, available thru ftp] in
order to see the context. (Of course, in the modern science game, hype
is inevitable, but such things exist in other forms in other small
communities not belonging to the scientific establishment..).

In particular, in section 6.2 when he describes the robot that picks
the soda can he says:

".. The hand has a grasp reflex that operated whenever something broke
an infra-red beam between the fingers. When the arm located a soda can
with its local sensors, it simply drove the hand so that the two fingers
lined up on on either side of the can. The hand then independently
grasped the can. ... As soon as it was grasped, the arm retracted -
it didn't matter whether it was a soda can that was intensionally grasped
or one that magically appeared."

[Please don't pick on the word "intensionally" in the last
sentence, it just means that the soda can which initially
"caused" the hand to approach the table, was not necessarily
the one whose perceived grasping "caused" the hand to retract.]

[[And don't pick on the word "cause", etc.]]

I think an experienced reader can rephrase all this in terms of
reference signals and controlled perceptions if he/she/it finds it
useful.

Best regards

--Oded

[From Rick Marken (920630.0900)]

Oded Maler (920630) says:

I think you misinterpret Brooks a little bit. As a roboticist he is
surely aware that the robot's world is just the perceptual space of
its sensors. I think all his work in the last years drove AI-based
robotics much closer to the lines of HPCT.

Funny how psychologists, biologists, roboticists, etc have been "moving
toward" HPCT for years. Randall Beer was supposedly close to PCT. Brooks
is claimed to be close. All kinds of psychologists are pointed to as
having PCT-like theories. Why don't they just adopt PCT, period. There
is a reason -- it's because "close" is VERY FAR same "getting" PCT and
getting PCT means kissing all the cause-effect crap goodby and recognizing
that behavior is the control of perception. The difference between "moving
toward" PCT and getting PCT is like the difference between "moving toward"
being pregnant and being pregnant. Even the stupid, confused Supreme
Court can apparently tell the difference between those two states. I
just think Brooks, Beer, Carver and Scheier, Lord, and all the other
people who are supposedly "moving toward" PCT just don't want to get
knocked up (if they came to a CSG meeting we could arrange a nice marriage
ceremony for them, however).

Best regards

Rick

···

**************************************************************

Richard S. Marken USMail: 10459 Holman Ave
The Aerospace Corporation Los Angeles, CA 90024
E-mail: marken@aero.org
(310) 336-6214 (day)
(310) 474-0313 (evening)