Modelling the World Model (was Motivation (was Teleology))

[Martin_Taylor2]

[Martin Taylor 2005.12.07.09.52]

[From Bjorn Simonsen (2005.12.07,09:00 EUST)]

In response to [Marc Abrams (2005.12.06,1256)]

When I look at your graph there is an Output Signal going to the
environments and also to the input function. I tried to put these
information into Rick's Spreadsheet, and I lost the negative feedback.
Then the model was not able to perceive what it wished to perceive.

Since Marc's diagram was the same as mine, in most respects, I think your comment should be aimed at me, too. I can answer it on my behalf, anyway.

Marc says his is merely a metaphor. I say mine is an incomplete circuit diagram. It's incomplete in that the "World Model" is not specified, and the manner in which the output of the "World Model" contributes to the PIF is not specified.

You say you "put" the connection through the world model into Rick's spreadsheet, and lost the negative feedback. You don't say just what you mean by "put". I assume you just added it in, but I can't see how that would make you lose the negative feedback, so I'm probably wrong.

The most simple-minded conception I have of how I would make the connection is a weighted average, where the perceptual signal is formed by a weighted average of the normal (sensory data only) PIF output and the output of the World Model (which emits the imagined result of what would be perceived given the actual output of the control system's output function). If we call the normal PIF output Ps (Perception sensory) and the output of the world model Pi (Perception imaginary), then the perceptual signal would be k*Ps + (1-k)*Pi.

The "World Model" is supposed to behave the way the actual world would do if influenced by the same output. I imagine that in your modelling, both connections from the output were simple connections, with no lags or complicated functions. That's fine, but if that's what you did, I can't see how you lost the negative feedback. All that should have happened is that the loop gain correcting for disturbances should have been reduced by the factor k.

In a more realistic use of the model, there must be provision for determining the value of k, and for determining the structure of the World Model. In the real world, even at the lowest level, the output has an effect on the outer world, and thus on the perceptual signal, that is extended in time. The World Model ideally should duplicate that, without itself being able to influence the outer world. How can this be done?

One approach to this is to use Bill P's ideas about the "Artificial Cerebellum", which uses the ongoing time pattern of the error signal to derive what amounts to an inverse world model in the output function, so that the time pattern of the output balances out the temporal fluctuations inherent in the environmental feedback path. The difference is that the World Model imagination connection doesn't balance out the effects on the real world, but replicates those effects in its structure. What it would proesumably use to do that is to compare its outputs with the actual Ps, and use those to alter its structure over time.

This approach to evolving the structure of the World Model also provides a means of arriving at "k" in the simplistic "weighted average" approach to merging the Ps with the Pi. The closer Ps and Pi have been over recent time, the greater the weight placed on Pi in the merging process.

As I imagine it, there are other things that influence "k", especially in a system that contains many interacting control loops. Among them are whether the current output is actually switched to influence the real world, and whether the data required to create Ps are currently available (which requires a separate perceiving system to make that determination). This latter enables a system with a good World Model to maintain stability (though not resistance to disturbance) when the perceptual input is disconnected. It is needed, because if there is no such "data available" signal in a control system, the loss of perceptual input would be treated the same as a massive disturbance, and the output would be likely to go crazy.

Which brings up a third factor affecting "k". Assuming a well developed World Model, if perceptual input is indeed coming in, and the Ps signal differs substantially from Pi, either a significant disturbance has occurred, or the World Model no longer correctly represents the environmental feedback path. Something has changed in the environment. In either case, k should be very small.

So there are three factors that affect how the output of the World Model should be expected to merge with the sensory input to generate the perceptual signal: (1) The mid-term accuracy of the World Model when compared with the purely sense-based perception (which, of course, might include imaginary components from lower levels in the presumed hierarchy), (2) the state of the switches (whether the control system's output is actually going to the outer world or the perceptual signal is coming in from the outer world), and (3) whether there is current significant difference between Ps and Pi.

Anyway, that's more or less how I conceive of completing the circuit diagram that is sketched in my figure. If you want to model it, using the spreadsheet or anything else, I'd be delighted. If that's what you have already modelled, I'd like to get a copy of th spreadsheet, to se where my ideas went wrong.

Martin

[Bozo_the_Clown]

From [Marc Abrams (2005.12.07.1044)]

In a message dated 12/7/2005 10:37:20 A.M. Eastern Standard Time, mmt-csg@ROGERS.COM writes:

···

[Martin Taylor 2005.12.07.09.52]

Marc says his is merely a metaphor. I say mine is an incomplete
circuit diagram. It’s incomplete in that the “World Model” is not
specified, and the manner in which the output of the “World Model”
contributes to the PIF is not specified.

It seems that our ‘world models’ serve different purposes. My world model is a good deal more than simply our ‘cognitive’ maps, and imagination, and I would be very interested in having you take a look at what I have done.

The reason I believe they serve different purposes are because we each seem to be representing the control processes at different levels of abstraction. This of course will have to be confirmed when we share some meaningful information.

I can see why you might have had a problem collecting data for your ideas. I just don’t think we have the tools or understanding yet at this level.

Regards,

Marc

[Bj_rn_Simonsen]

[From Bjorn Simonsen (2005.12.09,16:20 EUST)]
[Martin Taylor 2005.12.07.09.52]

Since Marc's diagram was the same as mine,
in most respects, I think your comment
should be aimed at me, too. I can answer it on my
behalf, anyway.

Marc says his is merely a metaphor. I say mine is an incomplete
circuit diagram. It's incomplete in that the "World Model" is not
specified, and the manner in which the output of the "World Model"
contributes to the PIF is not specified.

Marc's diagram is a metaphor and he asks me to not treat it as a model.
You say your diagram is an incomplete circuit diagram, then we can test it
out.
You emphasized that

.. , the output goes either to the real world or to
the imagined representation of the real world, not
both, though I acknowledge that it might be more
useful if the connection to the world model were
always there, and only the output to the real
world were to be switched on or off.

There is still a difference between your diagram (Martin Taylor
2005.12.06.13.07) and Marc's metaphor. He has traced an arrow from World
Model to the Comparator. Your diagram is more like Bill's diagram where a
reference comes from above. The only difference I see between your (Martin
Taylor 2005.12.06.13.07) and Bill's model is your box called World Model.
Maybe I have overlooked something?
But you acknowledged that it might be more useful if the connection to the
world model were always there.

My comments so far.
I have tried to imagine a subject and tried to perceive it at the same time,
but I can't get any confirmation that so happens. My wife helped me and hold
her thumb up in front of my nose. Then I closed my eyes and
imagined/remembered the thumb. My wife either hold her thumb there or took
it away. Then I opened my eyes. I "saw"/imagined her thumb with closed eyes,
but I could not "see"/imagine the thumb when I opened my eyes and she had
taken it away.
I think it is impossible to get a confirmation if we perceive and imagine
the same object at the same time.

I think upon your diagram as a composite loop. Without thinking upon a
circuit loop, I think some of the same when I look at Marc's metaphor. Then
I understand the World Model as a reservoir of experiences. The same
reservoir as "all references at all levels". When we imagine we disconnect
the environment, the error signal starts up with a reference signal (?). If
there are no parameters in the World Model or in the Input function, the
perceptual signal is instantaneous quite like the reference signal and the
error is zero. Next time I imagine something else, the same happens and so
on each time I imagine something. Different signals including different
loops and levels result in different imaginations. You can therefore
metaphorical name the box The World Model. Am I far away?

When I look at Marc's metaphorical diagram he brings the references from the
World Model. And that is OK, but by this he doesn't give any suggestions
about different levels. I think he do that conscious.

You say you "put" the connection through the world model into Rick's
spreadsheet, and lost the negative feedback. You don't say just what
you mean by "put". I assume you just added it in, but I can't see how
that would make you lose the negative feedback, so I'm probably wrong.

I did it with the changing disturbances Rick has programmed in his hier.exl.

I have done it again. If you are able to read the enclosed spreadsheet, you
find 4 copies of Rick's spreadsheet. First I put all disturbances to 5. Then
I ran the simulation with 1 repetition. The simulation has just started and
you find an error. Then I run the simulation with 1000 repetitions. Rick's
model has no errors.
Then on the third part I have changed all Rick's formulas for the P values.
I did change them to Rick's formula + Qo (the value you find just below P).
I ran the simulation with one repetition. The error is great. The error is
greater the higher up you are in the system. Then I ran the simulation with
1000 repetitions. There are still great errors at the level you control, the
highest level.

The most simple-minded conception I have of how I would make the
connection is a weighted average, where the perceptual signal is
formed by a weighted average of the normal (sensory data only) PIF
output and the output of the World Model (which emits the imagined
result of what would be perceived given the actual output of the
control system's output function). If we call the normal PIF output
Ps (Perception sensory) and the output of the world model Pi
(Perception imaginary), then the perceptual signal would be k*Ps +
(1-k)*Pi.

Yes, I understand. I guess it is OK that k=1 ??
I think what you propose is what I have done?

The "World Model" is supposed to behave the way the actual world
would do if influenced by the same output. I imagine that in your
modelling, both connections from the output were simple connections,
with no lags or complicated functions. That's fine, but if that's
what you did, I can't see how you lost the negative feedback. All
that should have happened is that the loop gain correcting for
disturbances should have been reduced by the factor k.

Yes, simple connections.
Maybe another value for k (than k=1) would have given a minor error.

Maybe I have not considered the World Model thoroughly enough, but I
expected a greater error the higher in the system we are. First because the
formula qo + slow(gain*error-qo) is the formula that gives the output that
makes negative feedback. Then because a change of p at level 3 comes with an
qo effect 4 repetitions before the qo effect should come at p et level 3.
( sorry for bothersome expression)

In a more realistic use of the model, there must be provision for
determining the value of k, and for determining the structure of the
World Model. In the real world, even at the lowest level, the output
has an effect on the outer world, and thus on the perceptual signal,
that is extended in time. The World Model ideally should duplicate
that, without itself being able to influence the outer world. How can
this be done?

Yes, I don't see the greatest problem at the lowest level.

One approach to this is to use Bill P's ideas about the "Artificial
Cerebellum", which uses the ongoing time pattern of the error signal
to derive what amounts to an inverse world model in the output
function, so that the time pattern of the output balances out the
temporal fluctuations inherent in the environmental feedback path.
The difference is that the World Model imagination connection doesn't
balance out the effects on the real world, but replicates those
effects in its structure. What it would presumably use to do that is
to compare its outputs with the actual Ps, and use those to alter its
structure over time.

I will look at the "Artificial Cerebellum" later.

This approach to evolving the structure of the World Model also
provides a means of arriving at "k" in the simplistic "weighted
average" approach to merging the Ps with the Pi. The closer Ps and Pi
have been over recent time, the greater the weight placed on Pi in
the merging process.

I have not thought upon Bill's imagination mode with special parameters
explaining why we remember something better because we just have
perceived/controlled it. Maybe a box as your World Box is a place where
parameters can be put in. Is this why you created the World box?

As I imagine it, there are other things that influence "k",
especially in a system that contains many interacting control loops.
Among them are whether the current output is actually switched to
influence the real world, and whether the data required to create Ps
are currently available (which requires a separate perceiving system
to make that determination). This latter enables a system with a good
World Model to maintain stability (though not resistance to
disturbance) when the perceptual input is disconnected.

Maybe the ks change. Look at white circle on a black paper. Close your eyes
and you see the circle still some time. But it vanishes. After a minute or
less, it is hard to imagine the circle. But you can.

It is needed,
because if there is no such "data available" signal in a control
system, the loss of perceptual input would be treated the same as a
massive disturbance, and the output would be likely to go crazy.

I didn't get this. P = disturbance + feedback signal. or P = disturbance +
qo + feedback signal (maybe some ks). If the perceptual input is lost, it is
the same as if you are in an imagination mode?

Which brings up a third factor affecting "k". Assuming a well
developed World Model, if perceptual input is indeed coming in, and
the Ps signal differs substantially from Pi, either a significant
disturbance has occurred, or the World Model no longer correctly
represents the environmental feedback path. Something has changed in
the environment. In either case, k should be very small.

Ps and Pi, representing the same object usually differ. Pi are always minor
Ps. You can't perceive an imagination as clear as you can perceive a
disturbance + feedback signal.
If you imagine the white circle on black paper, no ks make your imagination
as clear as when you see it with open eyes.

So there are three factors that affect how the output of the World
Model should be expected to merge with the sensory input to generate
the perceptual signal: (1) The mid-term accuracy of the World Model
when compared with the purely sense-based perception (which, of
course, might include imaginary components from lower levels in the
presumed hierarchy), (2) the state of the switches (whether the
control system's output is actually going to the outer world or the
perceptual signal is coming in from the outer world), and (3) whether
there is current significant difference between Ps and Pi.

I must spend more time after a comment from you. Maybe I am too far away?

Anyway, that's more or less how I conceive of completing the circuit
diagram that is sketched in my figure. If you want to model it, using
the spreadsheet or anything else, I'd be delighted. If that's what
you have already modeled, I'd like to get a copy of the spreadsheet,
to se where my ideas went wrong.

I will send you Rick's Excel witch I have changed later today.

Yes I will. If you give me some time I will make a loop viewed against the
background you have sketched here.

bjorn

hierMartin.exl (28.5 KB)

[Bj_rn_Simonsen]

[From Bjorn Simonsen (2005.12.09,16:21 EUST)]
[Martin Taylor 2005.12.07.09.52]

I enclose Rick's hier.exl. I have changed his formulas relative to my last
mail.

Bjorn

hierMartin.exl (28.5 KB)

[Bozo_the_Clown]

From [Marc Abrams (2005.12.09.1350)]

Bjorn, I hope you don’t mind me jumping in here but I think I can help shed some light on at least a few of the things.

I am, of course, not trying to represent Martin’s ideas. All of my comments are my ideas.

In a message dated 12/9/2005 10:40:38 A.M. Eastern Standard Time, bjornsi@BROADPARK.NO writes:

···

[From Bjorn Simonsen (2005.12.09,16:20 EUST)]

[Martin Taylor 2005.12.07.09.52]

Since Marc’s diagram was the same as mine,
in most respects, I think your comment
should be aimed at me, too. I can answer it on my
behalf, anyway.

Marc says his is merely a metaphor. I say mine is an incomplete
circuit diagram. It’s incomplete in that the “World Model” is not
specified, and the manner in which the output of the “World Model”
contributes to the PIF is not specified.

Marc’s diagram is a metaphor and he asks me to not treat it as a model.
You say your diagram is an incomplete circuit diagram, then we can test it
out.
You emphasized that

… , the output goes either to the real world or to
the imagined representation of the real world, not
both, though I acknowledge that it might be more
useful if the connection to the world model were
always there, and only the output to the real
world were to be switched on or off.

There is still a difference between your diagram (Martin Taylor
2005.12.06.13.07) and Marc’s metaphor. He has traced an arrow from World
Model to the Comparator. Your diagram is more like Bill’s diagram where a
reference comes from above. The only difference I see between your (Martin
Taylor 2005.12.06.13.07) and Bill’s model is your box called World Model.
Maybe I have overlooked something?

Yes, as I stated in

From [Marc Abrams (2005.12.07.1044)]

Martin’s diagram is at a much lower level of abstraction. What this means is that the components in my diagram represents many different functions and the interrelationships between them.

My diagram also talks about different things than Martins. Sort of like talking about the differences between cellular function and organ systems.

But you acknowledged that it might be more useful if the connection to the
world model were always there.

I think it is impossible to get a confirmation if we perceive and imagine
the same object at the same time.

I think you might be a bit confused here. What you ‘imagine’ and what you ‘remember’ come from the same exact place. Whatever you sense is recalled by your brain as being whatever you have remembered it to be. That memory of course is an interpretation of something you sensed in the environment and stored in memory.

All of your ‘perceptions’ are stored in memory and hence are ‘imagined’ as well.

Imagination encompasses all of our thinking, although popularly we like to distinguish between memory and imagination they are identical.

I think upon your diagram as a composite loop. Without thinking upon a
circuit loop, I think some of the same when I look at Marc’s metaphor. Then
I understand the World Model as a reservoir of experiences.

My world model encompasses more then just ‘experiences’.

It represents any number of ‘filters’ used to construct our perceptions, emotions, and it is where out ‘reference’ conditions ultimately come from.

When I look at Marc’s metaphorical diagram he brings the references from the
World Model. And that is OK, but by this he doesn’t give any suggestions
about different levels. I think he do that conscious.

Why not ask Marc? I don’t bite and I would love to talk about it, even privately if you feel uncomfortable about doing so on CSGnet. But why ask Martin to explain my ideas when I am available.

Yes, I do that consciously. My model as I currently envision it is a distributed network. My SD modeling should help clear up exactly what the functional structure looks like if I am successful.

Why does it matter what the structure is like? Are we interested in content or form?

From here on out I am no help as it all pertains to Martin and his ideas, there is nothing in the rest of this post that has anything to do with my model.

Regards,

Marc

[Bj_rn_Simonsen]

[From Bjorn Simonsen (2005.12.09,22:55 EUST)]

From [Marc Abrams (2005.12.09.1350)]

Bjorn, I hope you don't mind me jumping
in here but I think I can help shed some
light on at least a few of the things.

I appreciate when anybody shed some light on my way.

My diagram also talks about different things
than Martins. Sort of like talking about the
differences between cellular function and organ systems.

Yes I understand that.

I think you might be a bit confused here. What you
'imagine' and what you 'remember' come from the
same exact place. Whatever you sense is recalled
by your brain as being whatever you have remembered
it to be. That memory of course is an interpretation
of something you sensed in the environment and stored in memory.

I could have expressed the same, but I prefer to say that memory is a
representation of something we sensed earlier.

My world model encompasses more then just 'experiences'.

But I guess everything in your World Model have been experienced?

It represents any number of 'filters' used to construct
our perceptions, emotions, and it is where
out 'reference' conditions ultimately come from.

I must wait till you have explained what you mean by 'filters'.

When I look at Marc's metaphorical diagram he brings the references from

the

World Model. And that is OK, but by this he doesn't give any suggestions
about different levels. I think he do that conscious.

Why not ask Marc? I don't bite and I would love
to talk about it, even privately if you feel uncomfortable
about doing so on CSGnet. But why ask Martin to
explain my ideas when I am available.

I don't remember I asked Martin to explain your ideas.

I know you don't care for a hierarchy, you care for e distributed network.
That's the reason you put an arrow from the World Model and not references
from higher levels. Am I correct. If not, please explain how a baby after
some trial and error is able to stretch her arm toward colored dingeldang.

Why does it matter what the structure is like?
Are we interested in content or form?

I think you are more interested in content than in form. But I can be wrong.
I am more interested in a form because I define the form to include the
content.

bjorn

[Martin_Taylor2]

[Martin Taylor 2005.12.09.17.22]

[From Bjorn Simonsen (2005.12.09,16:20 EUST)]
[Martin Taylor 2005.12.07.09.52]

.. , the output goes either to the real world or to
the imagined representation of the real world, not
both, though I acknowledge that it might be more
useful if the connection to the world model were
always there, and only the output to the real

>world were to be switched on or off.

The only difference I see between your (Martin
Taylor 2005.12.06.13.07) and Bill's model is your box called World Model.
Maybe I have overlooked something?

Bill calls it "the imagination loop", but his connection is a simple wire, which is OK, because the environmental feedback part of the control loop is also drawn as a simple wire. I make this part of the circuit have possible complexities rather than being a simple connection.

My comments so far.
I have tried to imagine a subject and tried to perceive it at the same time,
but I can't get any confirmation that so happens. My wife helped me and hold
her thumb up in front of my nose. Then I closed my eyes and
imagined/remembered the thumb. My wife either hold her thumb there or took
it away. Then I opened my eyes. I "saw"/imagined her thumb with closed eyes,
but I could not "see"/imagine the thumb when I opened my eyes and she had
taken it away.
I think it is impossible to get a confirmation if we perceive and imagine
the same object at the same time.

I'm not at all clear what you are driving at with this example. When I expanded on my notion of "completing the circuit diagram" in [Martin Taylor 2005.12.07.09.52], I said:

"Which brings up a third factor affecting "k". Assuming a well developed World Model, if perceptual input is indeed coming in, and the Ps signal differs substantially from Pi, either a significant disturbance has occurred, or the World Model no longer correctly represents the environmental feedback path. Something has changed in the environment. In either case, k should be very small."

What that means is that, In your example, the world model would contribute little or nothing to the perceptual signal (as you experienced).

The World Model has nothing to do with predicting disturbances such as your wife's removal of her thumb. What it does is to reproduce the effect the control system's output would have on its perceptual signal if the disturbance signal remains unchanged. (At least that's a crude statement of it, because if the disturbance signal is predictable, such as being rhythmic, the World Model's evolution would cause it to incorporate that rhythm into its structure, in the way that the same thing happens with Bill's Artificial Cerebellum).

If you were controlling the position of your wife's thumb by some mechanism that used only your vision, and you shut your eyes, then if she didn't alter the force with which she was resisting your control, the World Model would allow you to keep her thumb in the desired place, but if she changed her force direction or magnitude, that would be a disturbance that you couldn't deal with by using only the World Model.

Then
I understand the World Model as a reservoir of experiences.

That's a wrong way to look at it. It is a time-domain filter, the form of which is affected by experience, but no single experience is recoverable from it. All it does is to produce an effect on its output "Pi" when the control system produced an output "o". Pi = Pi(o), a function that is extended in time. If the World Model is 100% accurate, then Pi(o) _= Ps(o) (Sorry, I can't write the three-barred equivalence signs).

The same
reservoir as "all references at all levels". When we imagine we disconnect
the environment, the error signal starts up with a reference signal (?). If
there are no parameters in the World Model or in the Input function, the
perceptual signal is instantaneous quite like the reference signal and the
error is zero.

You are asserting that the World Model is a simple wire, but that's OK if the normal connection through the environment is also a simple wire from output to perceptual signal (allowing for the addition of a disturbance, of course). But that's the kind of situation where you really wouldn't expect to need a World Model at all!

Next time I imagine something else,

By "something else" you must mean "a different value for the reference signal".

the same happens and so
on each time I imagine something.

That's right. The perceptual signal through the "simple wire" World Model will each time come to match the value of the reference signal.

Different signals including different
loops and levels result in different imaginations.

You suddenly changed the situation, here. You started with only one loop, and that loop had its perceptual input disconnected from the environment. Where did all these others come from?

You can therefore
metaphorical name the box The World Model. Am I far away?

What box? The one in my circuit diagram called the World Model? I don't metaphorically name it "World Model". That's the name I give the filter. Nothing metaphoric about it. If you want to call it something else, such as "Internal Feedback Filter" you can, but that would lose the connotation that the filter is supposed to match the filter that the outer environment imposes between the output signal and the perceptula signal. "I think "World Model" catches that connotation.

If strict HPCT were true, then the set of all the World Models in all the control loops of the hierarchy would represent everything you know about how the world works (not the codified knowledge you can put into words, but the kind that lets you ride a bike).

>You say you "put" the connection through the world model into Rick's

spreadsheet, and lost the negative feedback. You don't say just what
you mean by "put". I assume you just added it in, but I can't see how
that would make you lose the negative feedback, so I'm probably wrong.

I did it with the changing disturbances Rick has programmed in his hier.exl.

I have done it again. If you are able to read the enclosed spreadsheet, you
find 4 copies of Rick's spreadsheet.

My version of Excel (Mac Office X from a couple of years ago) doesn't understand the "exl" extension. If, after this message, you still think you are representing the World Model correctly, could you send it as "xcl" or something an older Excel version can read?

The most simple-minded conception I have of how I would make the
connection is a weighted average, where the perceptual signal is
formed by a weighted average of the normal (sensory data only) PIF
output and the output of the World Model (which emits the imagined
result of what would be perceived given the actual output of the
control system's output function). If we call the normal PIF output
Ps (Perception sensory) and the output of the world model Pi
(Perception imaginary), then the perceptual signal would be k*Ps +
(1-k)*Pi.

Yes, I understand. I guess it is OK that k=1 ??

If k = 1, then 1-k = 0, meaning no input from one or other (depending on whether you use k*Ps and (1-k)*Pi, as I did in th above paragraph, or the other way round, as I did later in my message, and in subsequent messages -- sorry for that confusion).

Maybe I have not considered the World Model thoroughly enough, but I
expected a greater error the higher in the system we are.

I wouldn't have expected any change other than a reduced effective gain.

First because the
formula qo + slow(gain*error-qo) is the formula that gives the output that
makes negative feedback.

OK, but remember that if you are including the World Model with "always on" connection to the perceptual signal, you have p = k*Pi + (1-k)*Ps. (I've reverted to giving "k" to the World-model's output).

I have not thought upon Bill's imagination mode with special parameters
explaining why we remember something better because we just have
perceived/controlled it. Maybe a box as your World Box is a place where
parameters can be put in. Is this why you created the World box?

No...

>It is needed,

because if there is no such "data available" signal in a control
system, the loss of perceptual input would be treated the same as a
massive disturbance, and the output would be likely to go crazy.

I didn't get this. P = disturbance + feedback signal. or P = disturbance +
qo + feedback signal (maybe some ks). If the perceptual input is lost, it is
the same as if you are in an imagination mode?

That's the idea. You can't resist disturbances, but if the disturbance doesn't change, you won't immediately go haywire.

Which brings up a third factor affecting "k". Assuming a well
developed World Model, if perceptual input is indeed coming in, and
the Ps signal differs substantially from Pi, either a significant
disturbance has occurred, or the World Model no longer correctly
represents the environmental feedback path. Something has changed in
the environment. In either case, k should be very small.

Ps and Pi, representing the same object usually differ. Pi are always minor
Ps. You can't perceive an imagination as clear as you can perceive a
disturbance + feedback signal.

Are you saying |Pi| < |Ps| always? I'm not at all clear what you mean here. Pi and Ps are both numbers. That's all they are.

I don't know whether this is helpful. I hope so.

Martin

[Bozo_the_Clown]

From [Marc Abrams (2005.12.09.2000)]

In a message dated 12/9/2005 4:58:23 P.M. Eastern Standard Time, bjornsi@BROADPARK.NO writes:

···

[From Bjorn Simonsen (2005.12.09,22:55 EUST)]

From [Marc Abrams (2005.12.09.1350)]

I could have expressed the same, but I prefer to say that memory is a
epresentation of something we sensed earlier.

Not only sensed Bjorn but interpreted. The interpretation is the important part. Both are needed for a perception. ‘Sensing’ only provides us with the ability to realize something is ‘out-there’.

It is how we interpret those sensations that dictate what we believe is out there.

Bill believes we ‘interpret’ our senses through his hierarchy. I disagree.

My world model encompasses more then just ‘experiences’.

But I guess everything in your World Model have been experienced?

Yes. Either prior or currently.

It represents any number of ‘filters’ used to construct
our perceptions, emotions, and it is where
out ‘reference’ conditions ultimately come from.

I must wait till you have explained what you mean by ‘filters’.

Good idea.

When I look at Marc’s metaphorical diagram he brings the references from
the
World Model. And that is OK, but by this he doesn’t give any suggestions
about different levels. I think he do that conscious.

Why not ask Marc? I don’t bite and I would love
to talk about it, even privately if you feel uncomfortable
about doing so on CSGnet. But why ask Martin to
explain my ideas when I am available.

I don’t remember I asked Martin to explain your ideas.

I responded to this;

When I look at Marc’s metaphorical diagram he brings the references from the
World Model. And that is OK, but by this he doesn’t give any suggestions
about different levels. I think he do that conscious.

It seemed to me you had a question as to whether I was or was not doing it ‘consciously’ and this post was addressed to Martin

I know you don’t care for a hierarchy, you care for e distributed network.
That’s the reason you put an arrow from the World Model and not references
from higher levels. Am I correct. If not, please explain how a baby after
some trial and error is able to stretch her arm toward colored dingeldang.

Yes, that is correct. But you must try to understand why. I am not trying to model what Bill is modeling. I am modeling different processes and at a different level of abstraction. I am attempting a top down view rather than bottom up.

Do you understand the differences between what I am attempting and what Bill or Martin is?

I don’t believe one is better than the other. I feel more comfortable and feel I can be more productive with my approach. Bill and I are ultimately after some of the same results and that is the acceptance of control as a necessary means of understanding human behavior.

Your preference might lie on the bottom as well and that is fine. I just don’t think we have the tools or understanding yet to move very far, but I think we can see the effects of control if we look hard enough at all the work that has already been done in studying human behavior.

Remember, if control is foundational, then everyone who has studied human behavior must have been studying part of the control process. The question then becomes; what aspect or effect of control have they been observing?

Why does it matter what the structure is like?
Are we interested in content or form?

I think you are more interested in content than in form. But I can be wrong.
I am more interested in a form because I define the form to include the
content.

Is this true? ‘Form’ to me is logic, and ‘content’ is truth. You can have valid form without valid content;

All celestial bodies are made of green cheese;

The moon is a celestial body;

so;

The moon is made out of green cheese.

This is a perfectly logical but false syllogism.

I have said this so often that I am tired of saying it but maybe it might be worth it to repeat at this time.

I believe what Bill Powers did 50 years ago was superb. But Bill was not the only one who had thoughts in this area.

To me who thought of what is really unimportant, what is important is the body of knowledge built up by some great minds. Bill being one of them. But all theories that have ever been devised by man are in fact "a work-in-progress’ or as Martin might say; the jury is still out.

I thought Bill understood this when he wrote in B:CP that this book was simply the opening salvo in our understanding of perceptual control and as Rick said yesterday was not intended to be taken as the last word.

But as Rick also said yesterday, somehow all this became written in stone and that has been very unfortunate for all involved.

There are many ways of viewing control processes and I happen to be taking a different one then Bill.

Regards,

Marc

[Richard_Marken2]

[From Rick Marken (2005.12.10.1010)

Martin Taylor (2005.12.09.17.22)

My version of Excel (Mac Office X from a couple of years ago) doesn't understand the "exl" extension. If, after this message, you still think you are representing the World Model correctly, could you send it as "xcl" or something an older Excel version can read?

Just change the file extension manually from "exl" to "xls".

Best

Rick

···

---
Richard S. Marken Consulting
marken@mindreadings.com
Home 310 474-0313
Cell 310 729-1400

[Martin_Taylor2]

[Martin Taylor 2005.12.10.13.53]

[From Rick Marken (2005.12.10.1010)

Martin Taylor (2005.12.09.17.22)

My version of Excel (Mac Office X from a couple of years ago) doesn't understand the "exl" extension. If, after this message, you still think you are representing the World Model correctly, could you send it as "xcl" or something an older Excel version can read?

Just change the file extension manually from "exl" to "xls".

Thanks. That worked. Except that Bjorn said there were four copies of the spreadsheet in the first version he sent, and I see only one. I don't know which one that is, but a clue might be that P2,1 (cell D8) has an added "D9" (a link directly to the corresponding output). And I can't find the difference between the two versions he sent. They have the same numbers, and in the several cells I checked, they seem to have the same formulae.

I can't see where Bk�rn implemented the World Model unless it is only done at level 2 and 3, and I do see a couple of things that look formally strange to me.

The first strangeness would have no effect, since the relevant numbers are identical, but in Row 5, the column A multipliers all refer to row 9 instead of 5.

The othe strangeness is that in row 12, one of the places where I would have expected to see the World Model, the perceptual signal is made by summing the CV (row 15) and a row marked "System" (row 14), which I assume is intended for the addition of system biases in perception. Rick's version doesn't have this system bias, but it doesn't matter, as they are all null. Perhaps Bjorn meant to add in the corresponding output (Row 13), instead of row 14.

I hate to admit this, but I can't find in the Excel menus or the spreadsheets (Rick's or Bjorn's) how to run it from scratch, or to single-step it. Both Rick and Bjorn obviously know how to do this, so could you give an infrequent Excel user a hint?

Thanks.

Martin

[Richard_Marken2]

[From Rick Marken (2005.12.10.1410)]

Martin Taylor (2005.12.10.13.53) --

I hate to admit this, but I can't find in the Excel menus or the spreadsheets (Rick's or Bjorn's) how to run it from scratch, or to single-step it. Both Rick and Bjorn obviously know how to do this, so could you give an infrequent Excel user a hint?

Go to the "Excel" pull down menu (one the top left, just to the right of the apple) and select "Preferences" then "Calculation". In the lower part of the "Calculation" menu is "Iteration" (which is checked) and below that "Maximum Iterations". Type in 1 for "Maximum Iterations" and close up the menu. Now each time you press F9 (or CMD =) to "run" the spreadsheet it will perform only one calculation (iteration).

Best

Rick

···

---
Richard S. Marken Consulting
marken@mindreadings.com
Home 310 474-0313
Cell 310 729-1400

[Bj_rn_Simonsen]

[From Bjorn Simonsen (2005.12.11,20:30EUST)]
[Martin Taylor 2005.12.09.17.22]

I am sorry for answering so late. It has been a busy weekend.

Bill calls it "the imagination loop", but his
connection is a simple wire, which is OK,
because the environmental feedback part
of the control loop is also drawn as a
simple wire. I make this part of the
circuit have possible complexities rather
than being a simple connection.

Very well.

The World Model has nothing to do
with predicting disturbances such
as your wife's removal of her thumb.

My motive with the thumb experiment was not to predict a disturbance. When I
opened my eyes, my wife hold her thumb before me or not. If she held it, I
perceived the thumb. In this case, there was a disturbance and if the World
Model is correct, there also were a World Model signal connecting the
perceptual signal. If my wife didn't hold her thumb before my nose, there
was no disturbance. Sometimes I imagined her thumb, sometimes not. If the
World Model is correct, sometimes there were a World Model signal and
sometimes the k made it like zero or very small.

Bill's model doesn't explain why I sometimes imagined the thumb and
sometimes not. He has no k.

What it does is to reproduce the
effect the control system's output
would have on its perceptual
signal if the disturbance signal
remains unchanged. (At least that's
a crude statement of it, because if
the disturbance signal is predictable,
such as being rhythmic, the World
Model's evolution would cause it
to incorporate that rhythm into its
structure, in the way that the same
thing happens with Bill's Artificial Cerebellum).

Do I understand you correct if I read your first sentence in this way: I
open my eyes and look at the thumb for 10 seconds (the disturbance variable
remains unchanged). In these 10 seconds there is an output signal feed back
to the Input function (in the retina, I think). At the same time this output
signal influence the perceptual signal through the World Model.

If you were controlling the position of your
wife's thumb by some mechanism that used
only your vision, and you shut your eyes,
then if she didn't alter the force with which
she was resisting your control, the World
Model would allow you to keep her thumb
in the desired place, but if she changed her
force direction or magnitude, that would be
a disturbance that you couldn't deal with
by using only the World Model.

Do I understand you correct if I read the above as: I controlled the sight
of my wife's thumb using only my sight. Then I closed my eyes. With my eyes
closed, I could imagine the sight of my wife's thumb. This is explainable
from both Bill's imagination/remembering model and your World Model.
Now to your subordinate clause ",then if she didn't alter the force with
witch she was resisting my control, the World Model would allow me to keep
her thumb I the desired place,". What could she de to resist my control when
my eyes were closed? With more ordinary words; What could she do to resist
my wish to imagine her thumb? Yes, what could she do? Detonate a bomb?
Now to your last part of your sentence. ", but if she changed her force or
magnitude,". I am too simple to understand how she could change her force
direction or magnitude when I could not perceive it because my eyes were
closed".
If she detonated a bomb, I am sure the World Model couldn't help me to
imagine her thumb. I am sorry. Here is something I don't understand.

Then
I understand the World Model as a reservoir of experiences.

That's a wrong way to look at it. It is a time-domain filter, the
form of which is affected by experience, but no single experience is
recoverable from it. All it does is to produce an effect on its
output "Pi" when the control system produced an output "o". Pi =
Pi(o), a function that is extended in time. If the World Model is
100% accurate, then Pi(o) _= Ps(o) (Sorry, I can't write the
three-barred equivalence signs).

If Marc reads this sentence, I am asking him if his World Model is a
reservoir of experiences?

Now to you Martin. Yes I understand that a "reservoir of experiences" is a
wrong way to look at it. The reservoir of experiences is for your model as
for Bill's model all existing, or all aroused references are the reservoir
of experiences.
When you say the the time-domain filter, The World Model, is effected by
experience, I understand that the time-domain filter is affected by Qo.
This time-domain filter produces a certain effect on Qo in a certain time
lag. In another time lag it produces another effect on Qo.
When you say that Pi = Pi(o), a function that is extended in time, I see a
parallel to your no-linear influence on the reference signal.
When you say "If the World Model is 100% accurate, then Pi(o) _= Ps(o)" (I
read not =), and that is because of the k in the World Model, a parameter we
don't find in the Input function (nor Feedback function).

You are asserting that the World Model is a simple wire, but that's
OK if the normal connection through the environment is also a simple
wire from output to perceptual signal (allowing for the addition of a
disturbance, of course). But that's the kind of situation where you
really wouldn't expect to need a World Model at all!

I think I understand. We need a World Model in situations where k _= 1.

Next time I imagine something else,

By "something else" you must mean "a
different value for the reference signal".

Yes and in different loops.

the same happens and so
in each time I imagine something.

That's right. The perceptual signal through the "simple wire" World
Model will each time come to match the value of the reference signal.

Different signals including different
loops and levels result in different imaginations.

You suddenly changed the situation, here. You started with only one
loop, and that loop had its perceptual input disconnected from the
environment. Where did all these others come from?

Yes, I think it was a way to explain what I meant with the reservoir of
experiences.

You can therefore
metaphorical name the box The World Model. Am I far away?

What box? The one in my circuit diagram called the World Model? I
don't metaphorically name it "World Model". That's the name I give
the filter. Nothing metaphoric about it. If you want to call it
something else, such as "Internal Feedback Filter" you can, but that
would lose the connotation that the filter is supposed to match the
filter that the outer environment imposes between the output signal
and the perceptual signal. "I think "World Model" catches that
connotation.

OK, I will remember calling it The World Model. It will have a "great" k(?).

If strict HPCT were true, then the set
of all the World Models in all the
control loops of the hierarchy would
represent everything you know about
how the world works (not the codified
knowledge you can put into words,
but the kind that lets you ride a bike).

Above you said the World Model was a time-domain filter and you rejected my
reservoir of experiences concept. Here when you say "..would represent
everything you know about how the world works, ..", you reminds me about the
reservoir of experiences again. But I think upon the World Model as a
time-domain filter.
A side track. What is the difference between the codified knowledge that is
put into muscles in the croup when I say "juice" or a philosophical
statement and the "knowledge" I put into muscles when I ride a Bike or when
I dance the Swan Lake?

My version of Excel (Mac Office X from a
couple of years ago) doesn't understand the
"exl" extension. If, after this message, you still
think you are representing the World Model
correctly, could you send it as "xcl" or
something an older Excel version can read?

I try to translate it to an older version of excel. I send an enclosure to
you private. Give me a word if I shall do it in another way.

Ps and Pi, representing the same object usually differ. Pi are always

minor

Ps. You can't perceive an imagination as clear as you can perceive a
disturbance + feedback signal.

Are you saying |Pi| < |Ps| always? I'm not at all clear what you mean
here. Pi and Ps are both numbers. That's all they are.

No, I am not saying that |Pi| < |Ps| (always). The value of Pi and Ps are
dependent on the reference values. I guess.
How shall we explain in Bill's model or in the World Model how imagined
things are more indistinct than perceptions with disturbances from the world
out there?

I don't know whether this is helpful. I hope so.

I hope so too. We'll see.
bjorn

[Martin_Taylor2]

[Martin Taylor 2005.12.11.15.48]

[From Bjorn Simonsen (2005.12.11,20:30EUST)]
[Martin Taylor 2005.12.09.17.22]

I am sorry for answering so late. It has been a busy weekend.

The World Model has nothing to do
with predicting disturbances such
as your wife's removal of her thumb.

My motive with the thumb experiment was not to predict a disturbance. When I
opened my eyes, my wife hold her thumb before me or not. If she held it, I
perceived the thumb. In this case, there was a disturbance and if the World
Model is correct, there also were a World Model signal connecting the
perceptual signal.

If you aren't controlling a perceptual signal, then there won't be any World Model connection, will there? You just see what you see. All the World Model does is to try to match the effect your output would have on the controlled perception of that one loop.

>What it does is to reproduce the

effect the control system's output
would have on its perceptual
signal if the disturbance signal
remains unchanged. (At least that's
a crude statement of it, because if
the disturbance signal is predictable,
such as being rhythmic, the World
Model's evolution would cause it
to incorporate that rhythm into its
structure, in the way that the same
thing happens with Bill's Artificial Cerebellum).

Do I understand you correct if I read your first sentence in this way: I
open my eyes and look at the thumb for 10 seconds (the disturbance variable
remains unchanged). In these 10 seconds there is an output signal feed back
to the Input function (in the retina, I think).

What output signal? What are you doing to influence the position of your wife's thumb? Whatever it is that you are doing, that's the output signal that would be fed back through the World Model.

At the same time this output
signal influence the perceptual signal through the World Model.

Correct, but as far as I understand your demonstration situation, it's your wife who is determining where her thumb is, and you are just observing, so there isn't any output signal to go thrugh any World Model.

>If you were controlling the position of your

wife's thumb by some mechanism that used
only your vision, and you shut your eyes,
then if she didn't alter the force with which
she was resisting your control, the World
Model would allow you to keep her thumb
in the desired place, but if she changed her
force direction or magnitude, that would be
a disturbance that you couldn't deal with
by using only the World Model.

Do I understand you correct if I read the above as: I controlled the sight
of my wife's thumb using only my sight.

You controlled the _POSITION_ of your wife's thumb using only your sight to guide the actions you were using to put the thumb where you wanted it.

Then I closed my eyes. With my eyes
closed, I could imagine the sight of my wife's thumb.

..could imagine the POSITION of your wife's thumb, which you were attempting to keep in some specific position.

Now to your subordinate clause ",then if she didn't alter the force with
witch she was resisting my control, the World Model would allow me to keep
her thumb I the desired place,". What could she de to resist my control when
my eyes were closed?

She could alter her arm muscle tensions.

With more ordinary words; What could she do to resist
my wish to imagine her thumb?

...to resist your wish that her thumb be where you wanted it to be.

If you are controlling for imagining that you see her thumb, that's an entirely different situation, and I think it's probably one that requires some structure other than HPCT to model it. You need another control system whose controlled perception is "whether you are imagining the thumb".

I'm dealing with only one basic control loop that contains a world model for that loop. The effect on the perception that the loop is controlling is what the World Model feeds back. In the present demonstration, the controlled perception is the position of your wife's thumb, and what gets fed back is the effect your control output should be expected to have on that position.

>> Then

I understand the World Model as a reservoir of experiences.

That's a wrong way to look at it. It is a time-domain filter, the
form of which is affected by experience, but no single experience is
recoverable from it. All it does is to produce an effect on its
output "Pi" when the control system produced an output "o". Pi =
Pi(o), a function that is extended in time. If the World Model is
100% accurate, then Pi(o) _= Ps(o) (Sorry, I can't write the

>three-barred equivalence signs).

Yes I understand that a "reservoir of experiences" is a
wrong way to look at it. The reservoir of experiences is for your model as
for Bill's model all existing, or all aroused references are the reservoir
of experiences.

Absolutely NOT!!!!!! References don't come into it at all!

When you say the the time-domain filter, The World Model, is effected by
experience, I understand that the time-domain filter is affected by Qo.

In a way, yes. The form of the filter is affected by the way changes in Qo have affected Ps, over its history. That's how it evolves to model the world.

This time-domain filter produces a certain effect on Qo in a certain time
lag. In another time lag it produces another effect on Qo.

Of Qo on Pi, not "on Qo".

When you say that Pi = Pi(o), a function that is extended in time, I see a
parallel to your no-linear influence on the reference signal.
When you say "If the World Model is 100% accurate, then Pi(o) _= Ps(o)" (I
read not =), and that is because of the k in the World Model, a parameter we
don't find in the Input function (nor Feedback function).

It's in the environmental feedback function, since we assert that the total effect of Ps and Pi is unity. "k" of it comes from the world model, (1-k) comes from the environmental feedback path, which is why I said that the only effect should be a reduction in the degree of control (caveat: that's true only when the disturbance has a "white" spectrum).

>You are asserting that the World Model is a simple wire, but that's

OK if the normal connection through the environment is also a simple
wire from output to perceptual signal (allowing for the addition of a
disturbance, of course). But that's the kind of situation where you
really wouldn't expect to need a World Model at all!

I think I understand. We need a World Model in situations where k _= 1.

It hasn't anything to do with k. It has to do with the time-complexity of the environmental feedback path.

>>Next time I imagine something else,

By "something else" you must mean "a
different value for the reference signal".

Yes and in different loops.

Which had different World Models of their own.

>> the same happens and so

in each time I imagine something.

That's right. The perceptual signal through the "simple wire" World
Model will each time come to match the value of the reference signal.

Different signals including different
loops and levels result in different imaginations.

You suddenly changed the situation, here. You started with only one
loop, and that loop had its perceptual input disconnected from the
environment. Where did all these others come from?

Yes, I think it was a way to explain what I meant with the reservoir of
experiences.

I'm afraid I'm getting quite lost, here. Could we go back to the simple World Model of a single loop? Dealing with a lot of them all at once, and at the same time dealing with the episodic memories that are supposed to contribute to reference signals, is a bit mind-boggling.

>If strict HPCT were true, then the set

of all the World Models in all the
control loops of the hierarchy would
represent everything you know about
how the world works (not the codified
knowledge you can put into words,
but the kind that lets you ride a bike).

Above you said the World Model was a time-domain filter and you rejected my
reservoir of experiences concept. Here when you say "..would represent
everything you know about how the world works, ..", you reminds me about the
reservoir of experiences again.

If, in your "reservoir" you have them all mixed up so that you can never extract a memory of any single experience, that's OK. But your previous references to the "reservoir" seemed to suggest that you might be able to pull one out and re-imagine it. That's a different area of imagination, called "episodic memory", as opposed to the effect of the World Model

What is the difference between the codified knowledge that is
put into muscles in the croup when I say "juice" or a philosophical
statement and the "knowledge" I put into muscles when I ride a Bike or when
I dance the Swan Lake?

OK. There are three different kinds of knowledge involved. One is the memory for specific events and circumstances. That's the kind Bill uses to reconstitute reference values, if I understand correctly. A second uses the way bits an pieces of sensory input hang together (correlate, at low levels). That "knowledge" is manifest in the structures of the various Perceptual Input Functions. The third kind is about the way the world works -- in other words, what happens when you act on it in different ways. This third kind is what is represented both in the Artificial Cerebellum and in the World Model.

If you are really acting in the real world, you are probably using all three. Explicit memory for, say, places, allows you to set a reference for where to bicycle, well-learned memory for what it feels like when you are riding straight, or turning, (or falling off), allows you to control your bike riding, and the Artifial Cerebellum and/or the World Model allow you to produce output that has an effect on what you will perceive the bike to be doing a short time in the future.

If you are planning an action, the World Model provides an estimate of the effect that action ought to have on the controlled perception if you did in the real world.

How shall we explain in Bill's model or in the World Model how imagined
things are more indistinct than perceptions with disturbances from the world
out there?

I don't have any thoughts on that, at least none that are precise enough to be put into words. The direction I would go with the idea is that the data of the real world are precise (apart from neural noise), whereas the imagination has historically never matched the reality when the two ought to have been the same, so the system ssomehow takes imagined scenes "with a grain of salt". Obviously that's not a scientific explanation, nor really a testable one (which amounts to the same thing). But it's probably the starting point from which I would seek a scientific explanation.

Martin