# Learning, Control, Science, Tetris

**URL:** <http://discourse.iapct.org/t/learning-control-science-tetris/6292>\
**Category:** CSG1996\
**Created:** [March 6, 1996, 5:14pm UTC](http://discourse.iapct.org/t/learning-control-science-tetris/6292 "1996-03-06T17:14:42Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Richard\_Marken](http://discourse.iapct.org/letter_avatar_proxy/v4/letter/r/9fc348/32.png) [@Richard\_Marken](http://discourse.iapct.org/u/Richard_Marken)\
**Post date:** [March 6, 1996, 5:14pm UTC](http://discourse.iapct.org/t/learning-control-science-tetris/6292/1 "1996-03-06T17:14:42Z")

</div>

[From Rick Marken (960306.0900)]

Hans Blom (960306a) --

> If you have already screwed in a lot of things in your lifetime --  
> screws, spark plugs, cork screws, what have you -- you may remember  
> that most of those required a clockwise movement.

Stefan Balke (960306.1400 MEZ) --

> This is a somehow artificial situation. In almost all situations I don't  
> start from zero, because I can make use of my knowledge of solutions for  
> related problems.

I was trying to describe the problem that exists for a "learning" system that  
is essentially starting from scratch; one that is trying to build a control  
system for the first time.

Try to imagine that \_you\_ are this learning system. You are given the  
following pieces of a control system: a perceptual signal (p), a reference  
for that signal (r) and a set of output variables (o.i) that can be hooked  
up in some way to the error signal (r-p). That is, your job as a learning  
system is to select an output variable and a function connecting that  
variable to the error, o = f(r-p), so that p is kept nearly equal to r. That  
is, you have to pick an o.i and an f(r-p) so that p is kept under control.

Notice that you have no basis for selecting a particular o.i (rotating,  
pulling, pushing, etc) or a particular f() (clockwise, counterclockwise,  
gently, forcefully, etc). I think this is the situation we were in when we  
originally learned most of what we learned in our lives; we had no basis  
for selecting outputs and output functions that would keep certain  
perceptions under control. That's why PCT assumes (as a starting assumption)  
that reorganization is random.

As we develop we learn things that can be a basis for subsequent learning or  
control; we "learn to learn" to some extent. But just because we do have a  
basis for selection, it is not necessarily a helpful basis. As Hans said:

> on average, trying clockwise first will save effort.

But we don't control on average. If we turn a counter-clockwise light-bulb  
clockwise, it doesn't work; we don't control the perception of the light-bulb  
until we find the right output function. In this particular case, using an  
existing light-bulb control system output (clockwise rotation) didn't work;  
we are in the same position we would be in if we had no control system for  
rotating light bulbs; all that previous learning was for nought.

Hans asks:

> Do you see the relation with model-based control?

Not really. [I still think "model-based control" is an oxymoron; there  
is no control (protection of a physical variable from disturbance) when there  
is no perception and you are generating output based on a model or to control  
a modelled (imagined) perception.] When you rotate a light-bulb using an  
existing output function (clockwise rotation) you are simply controlling; if  
you then use this control system to screw in light bulb that happens to  
require a counter-clockwise rotation you are still controlling but, in this  
case, unsuccessfully. I don't see how "model- based control" is involved at  
all.

Me:

> You seem to have discovered that control is a real phenomenon, that  
> it is exhibited by organisms and not atoms or (most) molecules.

Hans Blom (960306b) --

> Rick, I talk about a "concept", whereas you talk about "real". It's  
> all perception is what I want to say, and I seem to have quite different  
> perceptions than you do. Mine serve me well; yours serve you well, I  
> assume. Let's leave it at that, shall we?

Whether or not control is phenomenon is a factual question, not a matter of  
opinion. The question is answered by data, such as that from simple tracking  
tasks, that show certain variables (like the position of a cursor) being  
protected from the effects of disturbances by the precise opposing effects of  
another system.

If we can't agree on the answer to the factual question about the existence  
of the phenomenon of control, then I don't see how we can make any progress  
towards a better understanding of human nature. If it's all just a matter of  
opinion ("perception"), then your model of human nature is as good as mine.  
This relativistic attitude is fine when it comes to art but I reject it  
completely when it comes to science. The idea that control is not "real" may  
work for you but I can show (rather easily) that you are wrong; your opinion  
(perception) of control may work for you, but it's an obstacle to scientific  
communication - as though you were persisting in the belief that the earth  
is flat, because that belief works for you.

I asked:

> Does the environment control behavior?

You say:

> Rick, I've answered this question several times in the past. Yet you  
> are not satisfied with my answers. What do you \_really\_ want from me?

I want you to participate in and contributed to the development of a science  
of purposive systems. If you don't accept the facts on which this science is  
built (particularly the fact of control, a phenomenon exhibited by living and  
a few non-living [control] systems) then you cannot possibly contribute fully  
to the development of this science or even discuss it meaningfully.

You are obviously a skillful modeller. But there were Ptolomaic Jesuits  
who were pretty skillful modellers too. What I \_really\_ want to see is  
evidence that you (unlike the Jesuits) are willing to allow your theoretical  
skills to by disciplined by empirical observation.

Stefan Balke (960306.1400 MEZ)--

> I'm planing to do a study, where tetris players are videotaped while  
> playing tetris the first times...What points should be considered in order  
> to study the reorganization? Would it be useful to ask the people to do loud  
> thinking?

Wonderful idea. I know tetris; cool game. Of course, the problem with  
studying games like this is that people can get pretty skilled at it but they  
are never completely in control (if they were, it would be no fun anymore).

I think what I would watch for first is where they end up; try to get an idea  
of what variables people control when they have become pretty good at the  
game. For example, do they control for leaving certain shapes open; do they  
tend to put the L shapes in a certain position, etc? Once you know where  
people end up you can go back to the begining of the tape and see what  
people were trying to control at the start; did they try to control certain  
things about the L shapes, did they try to put things in the middle vs edges.  
See if you can see any "evolution" of the successful strategies.

This is a pretty complex study but one that could be pretty informative;  
there are natural disturbances (the shapes and orientations of the pieces,  
for example) that vary in unpredictable ways. If your records are detailed  
enough you could do a version of The Test to see if certain variables are  
controlled; or if an attept is being made to control those variables. Keep us  
posted.

Best

Rick