second-order and third-order beliefs

CSGnet Archive CSG2008
1

[From Bruce Nevin (2008.03.15 1647 EDT)]

http://www.eetimes.com/showArticle.jhtml?articleID=206903246

Yes, yes, I know this is brute force with massive computing power matching cognitive maps to inputs. The interesting things are (1) background knowledge and (2) second- and third-order beliefs.

We may presume that a lot of background knowledge is recorded as reference values for controlled perceptual variables. How strongly may we claim this?

Second-order beliefs are e.g. what I imagine that you imagine to be so. You don’t know that we moved the teddy bear to the closet, so I believe that you expect the bear to be in the cabinet still (if I’m more than about 4 years old).

Third-order beliefs must be what I imagine that you imagine about my beliefs. My “beliefs”, in their sense, I take to be my perceptions and references.

/Bruce Nevin
2

Re: second-order and third-order
beliefs
[Martin Taylor 2008. 03.15.17.10]

[From Bruce Nevin
(2008.03.15 1647 EDT)]
Second-order
beliefs are e.g. what I imagine that you imagine to be so. You don’t
know that we moved the teddy bear to the closet, so I believe that you
expect the bear to be in the cabinet still (if I’m more than about 4
years old).
Third-order beliefs
must be what I imagine that you imagine about my beliefs. My
“beliefs”, in their sense, I take to be my perceptions and
references.

You are talking about the basic elements of the Layered Protocol
Theory of dialogue. I don’t know whether the program to which you
linked is doing the same, but that’s the essence of the General
Protocol Grammar: when I believe that you have understood what I want
you to understand, and I believe that you believe that I believe that,
then the dialogue is satisfactorily completed (from my point of view).
There’s a complementary set of beliefs from your side, and the GPG is
devised to allow each of us to act so as to influence the other’s
perceptions that we do or don’t believe the relevant things.

Martin

3

Re: second-order and third-order beliefs
[From Bruce Nevin (2008.03.15 1824 EDT)]

Martin Taylor 2008. 03.15.17.10 –

You are talking about the basic elements of the Layered Protocol Theory of dialogue. I don’t know whether the program to which you linked is doing the same, but that’s the essence of the General Protocol Grammar: when I believe that you have understood what I want you to understand, and I believe that you believe that I believe that, then the dialogue is satisfactorily completed (from my point of view). There’s a complementary set of beliefs from your side, and the GPG is devised to allow each of us to act so as to influence the other’s perceptions that we do or don’t believe the relevant things.

Yes. It’s been a long time since I read your admirable LPT writings (you kindly sent me reprints sometime in the 1990s).

And yes, I suppose we should think of this program as controlling

beliefs

beliefs (belief)

beliefs (belief (belief))

for the sake of managing dialog in the Turing test. It very probably maintains “beliefs” as propositions using language or a language-like system grounded in natural language. I don’t know if LPT requires that.

I was proposing that humans actually maintain these “beliefs” (in all their echo-chamber layered complexity) as reference values for controlled non-linguistic perceptions, with the implication that articulation of them in sentences would be constructed de novo each time. I think it is the exception rather than the rule for us to recall a “credo” and to set our non-linguistic reference values according to it – what is called “acting on principle” or “principled behavior.” Put in other terms, a genuine “man (or woman) of principle” has non-verbal reference values which are consistent with his (or her) verbal articulation of principles.

To conduct the Turing Test is to interact with the other, looking to see if they fail to control some variable that a human would control, or if they fail to contol some variable in a manner in which I would expect a human to control it. So far, the only means of interacting are verbal, and so AI folks can imagine that it can all be stored as propositional knowledge. A proper Turing Test would be more broadly interactive.

/Bruce Nevin
···

From: Control Systems Group Network (CSGnet) [mailto:CSGNET@LISTSERV.UIUC.EDU] On Behalf Of Martin Taylor
Sent: Saturday, March 15, 2008 5:16 PM
To:
CSGNET@LISTSERV.UIUC.EDU
Subject: Re: second-order and third-order beliefs

[Martin Taylor 2008. 03.15.17.10]

[From Bruce Nevin (2008.03.15 1647 EDT)]
Second-order beliefs are e.g. what I imagine that you imagine to be so. You don't know that we moved the teddy bear to the closet, so I believe that you expect the bear to be in the cabinet still (if I'm more than about 4 years old).
Third-order beliefs must be what I imagine that you imagine about my beliefs. My "beliefs", in their sense, I take to be my perceptions and references.

You are talking about the basic elements of the Layered Protocol Theory of dialogue. I don’t know whether the program to which you linked is doing the same, but that’s the essence of the General Protocol Grammar: when I believe that you have understood what I want you to understand, and I believe that you believe that I believe that, then the dialogue is satisfactorily completed (from my point of view). There’s a complementary set of beliefs from your side, and the GPG is devised to allow each of us to act so as to influence the other’s perceptions that we do or don’t believe the relevant things.

Martin

4

Re: second-order and third-order
beliefs
[From Martin Taylor 2008.03.15.21.55]

[From Bruce Nevin
(2008.03.15 1824 EDT)]
Martin Taylor 2008. 03.15.17.10

You are talking about the basic elements
of the Layered Protocol Theory of dialogue. I don’t know whether the
program to which you linked is doing the same, but that’s the essence
of the General Protocol Grammar: when I believe that you have
understood what I want you to understand, and I believe that you
believe that I believe that, then the dialogue is satisfactorily
completed (from my point of view). There’s a complementary set of
beliefs from your side, and the GPG is devised to allow each of us to
act so as to influence the other’s perceptions that we do or don’t
believe the relevant things.

Yes. It’s
been a long time since I read your admirable LPT writings (you kindly
sent me reprints sometime in the 1990s).
And yes, I suppose
we should think of this program as controlling

beliefs

beliefs (belief)

beliefs (belief (belief))
for the sake of
managing dialog in the Turing test. It very probably maintains
“beliefs” as propositions using language or a language-like
system grounded in natural language. I don’t know if LPT requires
that.

No it doesn’t. LPT makes no explicit statements about language.
According to LPT, multi-level dialogues could be conducted entirely
without words, though under most circumstances it’s much easier to use
words than not. How easy it is, relatively speaking, to use words or
to avoid them depends a great deal on what each dialogue partner knows
of the other and of the other’s knowledge of onesself. LPT was
originally invented as a design method for multimodal (speech,
keyboard, and gesture) interaction with computers, and the combined
use of verbal and non-verbal interaction is discussed as the central
point of at least one paper.

If I know no Chinese and my Chinese dialogue partner knows no
English, we may still be able to conduct a dialogue of sufficient
complexity to allow me to get a satisfactory meal or a hotel room,
though perhaps not to debate the finer points of Confucian
philosophy.

At the other end of the spectrum, with someone I know very well,
a turn of the head and a lifted eyebrow may suffice to get across some
non-trivial concept such as that one choice of food to be bought would
be preferable to another, or that it might be time to repaint a
room.

It is with people in the middle, people who shae a common
language and some background, that language becomes most
important.

I was proposing
that humans actually maintain these “beliefs” (in all their
echo-chamber layered complexity) as reference values for controlled
non-linguistic perceptions, with the implication that articulation of
them in sentences would be constructed de novo each
time.

Perhaps we think of beliefs differently. I think of “belief”
as equivalent to “perception” rather than as equivalent to
“reference”. The dialogue is initiated because some
controlled perception has a non-zero error value, and one mechanism to
influence that perception is to bring my perception of some state of
another person to some reference state different from its current
state – in equivalent words, I would like a window to be open and it
isn’t; I don’t believe the person near the window is currently about
to open it, and I wish to believe that he is.

I think it is
the exception rather than the rule for us to recall a “credo”
and to set our non-linguistic reference values according to it –
what is called “acting on principle” or “principled
behavior.” Put in other terms, a genuine “man (or woman) of
principle” has non-verbal reference values which are consistent
with his (or her) verbal articulation of
principles.

Yes, but I’m not sure how this relates to the recursion of belief
that started this thread and this message. It seems to belong to a
different thread.

To conduct the
Turing Test is to interact with the other, looking to see if they fail
to control some variable that a human would control, or if they fail
to contol some variable in a manner in which I would expect a human to
control it. So far, the only means of interacting are verbal, and so
AI folks can imagine that it can all be stored as propositional
knowledge.

That might be adequate, as it might be difficult to distinguish
the machine from a high-functioning autistic person who does have
reference values derived from verbal propositions painstakingly
learned.

A proper
Turing Test would be more broadly interactive.

But to what degree? Must the machine look like a human and
display appropriate facial expressions? If the Turing tester is to see
the machine in order to have a non-verbal interaction with it, then if
it doesn’t look human, it won’t be judged to be human, no matter how
human-like its behaviour.

I agree with your statement, but don’t quite see how one could
conduct a test based on interactive style and competence while
avoiding the problem of non-human appearance. Perhaps we might allow
interaction through Second-Life avatars, or something along those
lines.

Martin

5

Re: second-order and third-order beliefs
[From Bruce Nevin (2008.03.17 1543 EDT)]

Martin Taylor 2008. 03.15.17.10 –

Perhaps we think of beliefs differently.

Ah, yes, you’ve identified some equivocation. I distinguish belief from imagined perception of what the other perceives. By this standard, I’m misusing the term belief equivocally to refer to both. as you say

Martin Taylor 2008. 03.15.17.10 –

I’m not sure how this relates to the recursion of belief that started this thread and this message. It seems to belong to a different thread.

Perhaps they are two different threads, and perhaps the word “belief” should be used only in one of these threads. Let’s see.

What distinction am I making and is it valid? One could say that “imagined perception of what the other perceives” is a particular case of belief. Is it no more than that? The particular case of empathetic imagining of another’s perceptions, enabling communication (your preferred thread topic), is typically rather transient. Belief in the general case, excluding empathetic projections, is more static and more durative.

The Theory of Mind experimental setup shows that empathetic projection can be more durative. The child A (if old enough) imagines that B expects the bear to be where he (B) last saw it, in the cabinet, even though she (the child A) knows that she and C moved it to the closet while B was out of the room. This setup is concerned with

  • A’s perception (in imagination) of what B perceives
  • A’s perception (in imagination) of B’s perception (in imagination) of what A perceives
    This ToM scenario is not concerned with the converse (B’s perception of A, etc.), which is to say that it is in fact about concealment as contrasted with communication. Or we can think of concealment and deceit as a special case of communication. Only slightly perverse, that.

Martin Taylor 2008. 03.15.17.10 –

I think of “belief” as equivalent to “perception” rather than as equivalent to “reference”.

Are all perceptions beliefs, then? Wouldn’t you limit the term to controlled perceptions, at least? If so, then isn’t the preferred value the belief, and the range of possible other values not the belief? To believe something is to have an expectation that it is so.

But here I refer to beliefs in general, the other thread topic, and you have in mind the narrow case of dynamically constructing imagined perceptions of what the other is perceiving, your preferred thread topic. This is the equivocation noticed above.

Martin Taylor 2008. 03.15.17.10 –

I would like a window to be open and it isn’t; I don’t believe the person near the window is currently about to open it, and I wish to believe that he is.

I hope you will forgive me when I say that is a tortured use of the word “believe”. You don’t wish merely to believe that he is about to open the window. You could have abundant faith that he will open it, and whether or not your belief corresponds to reality may very likely remain unaffected by any of your behavior precisely because of that confident expectation. (Why should you do anything about it? You believe he’s going to open it.) It won’t do to say that you will control what are for you relevant variables until you believe that he will open it, because you are still only believing that he will. The anthropologists’ workhorse example of the cargo cult in the South Pacific illustrates the problem.

No, you perceive that he is currently not opening the window and you wish to perceive that he is opening the window. The instrumentality of communication to bring that wish to fulfillment depends upon your correctly imagining that he is controlling some variable(s) whose states you can influence, and then influencing the states of those variables by your control actions. It is not the particular actions (linguistic, gestural, or anything at all) that makes them communicative acts, rather, it is the successful alignment of your imagination of what he is controlling with what he is in fact controlling, and a corresponding alignment the other way as he (controlling disturbances due to your influencing of the states of some of those variables) comes to imagine in turn what you are controlling. Communication induces an interlock of the respective purposes of those communicating.

Martin Taylor 2008. 03.15.17.10 –

If the Turing tester is to see the machine in order to have a non-verbal interaction with it, then if it doesn’t look human, it won’t be judged to be human, no matter how human-like its behaviour.
Perhaps we might allow interaction through Second-Life avatars, or something along those lines.

Yes, that is what was reported in the article. I suppose nonverbal aspects of these cgi avatars might get pretty sophisticated in time. (But this is yet another thread topic.)

/Bruce
6

[From Bill Powers (2008.03.187.1509 MDT)]

Perhaps they are
two different threads, and perhaps the word “belief” should be
used only in one of these threads. Let’s see.

What distinction
am I making and is it valid? One could say that “imagined perception
of what the other perceives” is a particular case of belief. Is it
no more than that?

I’ll leave the main thrust of this thread to the experts, but the concept
of belief seems important, and important to define a little better if we
can.

Suppose I say, “I am seeing a blue ball.” Focus on the
“blue” term. That is a description of an experience that is
happening now. I am simply saying the word I use to refer to experiencing
that color.

Now suppose I say “I believe I am seeing a blue ball.” That is
not a description of an experience, but a comment about the experience.
In fact, I think that using the term “believe” in that context
introduces some uncertainty into the statement that is not in the
previous statement. Perhaps my meaning is something like “I know I
am experiencing the color blue [no uncertainty] but I don’t know if the
ball is actually blue.” Perhaps I think there may be some trick of
lighting making it look temporarily blue, or perhaps the color is near a
boundary of the set of colors for which I use the word “blue.”
Perhaps I think it may be over the boundary that someone else uses, or
maybe I think I could be having a minihallucination.

At any rate, it seems to me that saying I believe something introduces
the possibility of doubt. “He’s an honest person,” versus
“I believe he’s an honest person”. “It’s ten o’clock”
versus “I believe it’s ten o’clock.” To say I believe some
statement is to indicate that I think I could be wrong. It’s like saying
“She’s sincere” versus “She’s very sincere.” The
“very” weakens the assertion of sincerity, especially if you
add, “…I’m sure.” So does the idea of belief weaken any
assertion of fact. And to say “I believe X”, while weaker than
just saying X, is stronger than saying “I really believe
X.”

Best,

Bill P.

7

Re: second-order and third-order
beliefs
[From Martin Taylor 2008.03.17.18:00]

[From Bill Powers (2008.03.187.1509
MDT)]

Perhaps they are
two different threads, and perhaps the word “belief” should
be used only in one of these threads. Let’s see.

What distinction am
I making and is it valid? One could say that “imagined perception
of what the other perceives” is a particular case of belief. Is
it no more than that?

At any rate, it seems to me that saying I
believe something introduces the possibility of doubt. “He’s an
honest person,” versus “I believe he’s an honest
person”.

Yes, I agree. That’s the distinction I would make between the
kind of perception in “I see a chair in this room” and
“I believe there is a chair in the next room” and “I
believe that what I see through the fog is a chair”. They are all
perceptions, but the latter two have less supporting information –
the perceptual input function has ambiguous inputs along with the well
defined ones.

This suggests that all perceptions actually have two attributes:
the value and the uncertainty. Locally, we had some dealings a while
back with a company who seemed to get much more powerful results from
their neural nets by making all values complex, value being theta,
uncertainy being r. There may be something to that in the
psychological case, too.

Martin

8

Re: second-order and third-order
beliefs
[From Martin Taylor 2008.03.17.23.32]

[Bruce Nevin
(2008.03.17 1543 EDT)]
Martin Taylor 2008.
03.15.17.10 –
Perhaps we think of
beliefs differently.
Ah, yes, you’ve
identified some equivocation. I distinguish belief from imagined
perception of what the other perceives. By this standard, I’m misusing
the term belief equivocally to refer to both. as you
say
What distinction am
I making and is it valid? One could say that “imagined perception
of what the other perceives” is a particular case of
belief.

In my equation of belief with peception rather than with
reference, I don’t think of imagination. When I engage in dialogue I
perceive – perhaps foggily, perhaps non-veridically – what the other
perceives. I am not imagining. I am using what data I can get from
observation to determine the state of the other. Some other time and
place, I may imagine what the other might perceive if I were to say
something, but if I am truly engaged in a dialogue, I am controlling
my perception of what the other is perceiving.

Martin Taylor 2008.
03.15.17.10 –
I think of “belief” as
equivalent to “perception” rather than as equivalent to
“reference”.
Are all perceptions
beliefs, then? Wouldn’t you limit the term to controlled perceptions,
at least?

No. I have lots of beliefs about states of the world I am not
controlling, and often could not control. I believe there are over 6
billion people in the world, but I don’t control that by going around
murdering millions (my reference value for that belief is well under
one billion). But after reading Bill P’s contribution, I might limit
“belief” to perceptions that are less certain than
perceptions to which I would give the term “immediate”, such
as the perception of “blue”.

If so, then
isn’t the preferred value the belief, and the range of possible other
values not the belief? To believe something is to have an expectation
that it is so.

To have an expectation that X has the value x is not to have a
reference for X to have a value x. My preferred value for the name of
the President of the United States is not and never has been George W.
Bush, but my belief is that this is the name of the current
President.

But here I refer to
beliefs in general, the other thread topic, and you have in mind the
narrow case of dynamically constructing imagined perceptions of what
the other is perceiving,

Cut out “imagined” when we are talking about the second
and third order beliefs that occur in dialogue. And, as you see above,
I don’t accept your interpretation of “belief” as reference
value, even for beliefs in general.

Martin Taylor 2008.
03.15.17.10 –
I would like a window to be open and it
isn’t; I don’t believe the person near the window is currently about
to open it, and I wish to believe that he is.
I hope you will
forgive me when I say that is a tortured use of the word
“believe”. You don’t wish merely to believe that he is about
to open the window. You could have abundant faith that he will open
it,

Actually, in the cited example, I have no belief that he would
open it unless I disturbed some perception in such a way that his
action would result in him opening it. In fact, as stated, I believe
that he will NOT open it, but I wish to believe that he will. My
reference state for my belief differs from my current belief.

and whether
or not your belief corresponds to reality may very likely remain
unaffected by any of your behavior precisely because of that confident
expectation. (Why should you do anything about it? You believe he’s
going to open it.)

If I believed he would be about to open it, I would probably do
nothing, although I might act to give him data that would allow him to
perceive that I would be pleased by that action. I’m presuming that
one of his controlled perceptions (beliefs) is my state of happiness,
and I would help his control of that perception by giving him data on
its current state.

No, you perceive
that he is currently not opening the window and you wish to perceive
that he is opening the window.

You mean “Yes”, not “No.”

The
instrumentality of communication to bring that wish to fulfillment
depends upon your correctly imagining that he is controlling some
variable(s) whose states you can influence, and then influencing the
states of those variables by your control actions.

Exactly. One possible control action might be to say “Would
you mind opening the window”. Another might be to say “Isn’t
it awfully hot in here”. Another might be to say nothing but to
keep wiping my brow. All of these (I would hope) would disturb his
controlled perception of my happiness and give him a clue as to what
action he could take to bring that perception to what I hope is its
reference value of perceiving me to be happy with the situation. The
main point of using langauge is that it eases his job of determining
what action would make me happy. Language tends to be more specific
than gesture in most circumstances.

If those don’t work, just as with any other control at moderately
high levels, I would try something else. Depending on whether the
perception I am controlling is simply to have the window open, or to
have that person open the window (controlling my perception of my
ability to control him), the alternate actions might differ
appreciably.

It is not the
particular actions (linguistic, gestural, or anything at all) that
makes them communicative acts, rather, it is the successful alignment
of your imagination of what he is controlling with what he is in fact
controlling, and a corresponding alignment the other way as he
(controlling disturbances due to your influencing of the states of
some of those variables) comes to imagine in turn what you are
controlling.

Yep. I might even tolerate your use of “imagination” in
line 3, even though I think it is more a perception based on real data
gathered over some probably long period of time. If you know the
person, it is based on your prior interactions, and if you don’t (here
we impinge on 'social control") it is based on past interactions
with others of a character that this person seems to have.

Communication
induces an interlock of the respective purposes of those
communicating.

Yes. That’s the core of LPT.

Martin Taylor 2008.
03.15.17.10 –
If the Turing tester is to see the
machine in order to have a non-verbal interaction with it, then if it
doesn’t look human, it won’t be judged to be human, no matter how
human-like its behaviour. Perhaps we might allow interaction through
Second-Life avatars, or something along those lines.
Yes, that is what
was reported in the article. I suppose nonverbal aspects of these cgi
avatars might get pretty sophisticated in time. (But this is yet
another thread topic.)

Sorry, I missed that bit.

Martin

9

[From Rick Marken (2008.03.17.2200)]

Martin Taylor (2008.03.17.18:00)

This suggests that all perceptions actually have two attributes: the value
and the uncertainty.

Hey, what's going on here? I thought this was St. Patrick's Day, not
April Fools Day! :wink:

First Dag says that different athers behave differently in the same
situation because they perceive it differently, implying that
different behavior is the result of different perceptions. And now you
suggest that perceptions come equipped with values and uncertainly
that cause belief. That seems to be what you seem to be saying here,
anyway:

That's the distinction I would make between the kind of perception in "I see
a chair in this room" and "I believe there is a chair in the next

room" and "I believe

that what I see through the fog is a chair". They are all perceptions, but the latter
two have less supporting information -- the perceptual input function has
ambiguous inputs along with the well defined ones.

It seems to me that it is your references for things like "being
right" and "being embarrassed" and the like that determines whether
you say, based on your perceptions, that "there is" a chair in the
other room or in the fog or, instead, that you "believe there is" a
chair in the other room or in the fog. I've never had to include
"supporting information" along with the perception inputs in my
models of control. Controlling works well by just assuming is that
what is controlled is the scalar output of a perceptual function.
Maybe if you showed me a working model of what you have in mind it
would make more sense to me.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

10

[Martin Taylor 2008.03.18.01.31]

[From Rick Marken (2008.03.17.2200)]

Martin Taylor (2008.03.17.18:00)

This suggests that all perceptions actually have two attributes: the value
and the uncertainty.

Hey, what's going on here? I thought this was St. Patrick's Day, not
April Fools Day! :wink:

And there never were any snakes in Ireland, or so I believe.

First Dag says that different athers behave differently in the same
situation because they perceive it differently, implying that
different behavior is the result of different perceptions.

I believe Dag is right, though behaviour is not the result of perceptions; that's an S-R claim that he didn't make. I know I see things differently at one time as compared to another, even though the physical situation may not have changed. It certainly happens after a period of trying to find particular weeds in a hedge! If that happens, who am I to claim that all people perceive a given phsyical sitaution the same?

And now you
suggest that perceptions come equipped with values and uncertainly
that cause belief.

Not that cause belief. I'm agreeing with Bill P [Bill Powers (2008.03.187.1509 MDT)]: "At any rate, it seems to me that saying I believe something introduces the possibility of doubt." I'm labelling an uncertain perception as a belief.

Yes, I know we perceive what we perceive. But of some (many) things, we perceive them and at the same time we perceive that we are uncertain about them. We can control for achieving a reference value of that perception of uncertainty, too, by acting to get more information.

Are you suggesting the we never are uncertain?

That seems to be what you seem to be saying here,
anyway:

That's the distinction I would make between the kind of perception in "I see
a chair in this room" and "I believe there is a chair in the next

room" and "I believe

that what I see through the fog is a chair". They are all perceptions, but the latter
two have less supporting information -- the perceptual input function has
ambiguous inputs along with the well defined ones.

It seems to me that it is your references for things like "being
right" and "being embarrassed" and the like that determines whether
you say, based on your perceptions, that "there is" a chair in the
other room or in the fog or, instead, that you "believe there is" a
chair in the other room or in the fog.

It has nothing to do with interactions with other people. Of some things I am more assured than I am of other things.

I've never had to include
"supporting information" along with the perception inputs in my
models of control.

No, I appreciate tha, but then you are certain about a lot of things :slight_smile:

  Controlling works well by just assuming is that
what is controlled is the scalar output of a perceptual function.
Maybe if you showed me a working model of what you have in mind it
would make more sense to me.

Interesting idea. I actually didn't have anything in mind other than my and Bill's observation of the phenomenon that we do perceive our own uncertainty.

To make this into a useful observation certainly does require a working model. I wonder how a one would look. I mentioned the company (whose name I am blocking on) who used complex-valued representations in perceptron-like neural nets. The perceptual side of an HPCT structure is exactly a perceptron, so maybe it would work better if the input functions took complex arguments and provided a complex perceptual signal. It would be quite interesting to follow that through. I wonder if there should be one complex comparator or whether only the value component would go to the comparator, while the uncertainty component might modulate the output function...Hmmm...goes away imagining whether this makes any sense...

Martin

11

[From Rick Marken (2008.03.18.1200)]

Martin Taylor (2008.03.18.01.31)_-

> Rick Marken (2008.03.17.2200)--

>First Dag says that different athers behave differently in the same
>situation because they perceive it differently, implying that
>different behavior is the result of different perceptions.

I believe Dag is right,

Me too.. But when one says that fathers behave differently (some
answering the cry and some not) because they perceive the cry
differently, the implication is that people behave differently because
they perceive things differently. Of course, control theory says that
this is true, because different output actions may be needed to
control different perceptions of the same environmental variable. So
saying that people behave differently when they perceive things
differently is certainly consistent with PCT. It's just that this
would come as no surprise to a conventional psychologist either, most
of whom believe that we behave differently because we perceive
different aspects of the same situation. This, as you know, was a
basic component of stimulus sampling theory; it is certainly
consistent with information processing theories, etc.

What distinguishes PCT from conventional psychologies, I think, is the
idea that the same perception can lead to quite different behavior
depending on the reference for that perception. In the crying baby
example, it seems very likely that all fathers with normal hearing
perceive a baby's cry in pretty much the same way, just as subjects in
a tracking task probably perceive the position of the cursor in pretty
much the same way. The main reason for any difference in their
behavior can probably be explained by a difference in reference for
the state of the cry. What makes PCT unique (and uniquely unacceptable
to psychologists, I'm afraid) is that it shows that people can want
the same perception in different states. While it's true that people
may perceive the same situation differently, I think it's important to
explain that any observed difference in the way people behave is due
to the fact that different actions are needed to control different
perceptions, just as different actions are needed to control the same
perception relative to different references.

Perhaps the PCT point I am trying to make is that a difference in
behavior is not diagnostic of a difference in perception. Indeed,
behavior (observed actions) in itself don't tell you much about what a
person is doing; what is important is to determine what perceptions
are being controlled. In the case of the fathers who behave
differently, understanding this difference is a matter of determining
what perceptions each is controlling (perceptions that, in one case,
are disturbed by crying and in the other case not). It may be that the
fathers are controlling different perceptions of crying; or it may be
that both are controlling for a perception of crying, but at a
different reference level. The important thing to know about the
fathers is what perception that are controlling -- and at what
reference level.

though behaviour is not the result of
perceptions; that's an S-R claim that he didn't make.

Right, he didn't it explicitly. But when you say that behavior differs
because perception differs, people who are not familiar with PCT could
assume that you are making something like this claim: that perception
causes behavior, because this is what most people (and virtually all
psychologists) think is the case. I think it's always best to talk
about PCT in terms of _controlled_ perceptions.

> And now you
>suggest that perceptions come equipped with values and uncertainly
>that cause belief.

Not that cause belief. I'm agreeing with Bill P [Bill Powers
(2008.03.187.1509 MDT)]: "At any rate, it seems to me that saying I
believe something introduces the possibility of doubt." I'm labelling
an uncertain perception as a belief.

I'm sorry. You said:"This suggests that all perceptions actually have
two attributes: the value and the uncertainty." This sounded to me
like a perceptual variable, such as my perception of my distance from
the computer screen, comes along with a value and an uncertainty. I
think of the distance perception as just that; a perception of
distance. I can also perceive the value of this perception as
_another_ perception (it's not very valuable to me); same for the
uncertainly (it's very certain).

We can control for achieving a reference value
of that perception of uncertainty, too, by acting to get more
information.

Yes, I agree. But now you are controlling for another perception: certainty.

Are you suggesting the we never are uncertain?

No. I'm suggesting that certainly is a perceptual variable itself,
separate from other perceptions, like my perception of the letters in
this sentence.

To make this into a useful observation certainly does require a
working model.

Before you make the model I think you should describe the phenomenon
to be explained by the model. Perhaps control of a probabilistic
variable; you move a handle to make an event more or less "certain".
See if a person can control for a reference amount of certainly (or
probability of occurrence). The disturbance would be an independent
influence on the certainty of the event. The subject would be asked to
control the event at a reference level of certainty.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

12

[From Bill Powers (2008.03.18.0215 MDGT)]

Martin Taylor 2008.03.17.18:00 --

At any rate, it seems to me that saying I believe something introduces the possibility of doubt. "He's an honest person," versus "I believe he's an honest person".

This suggests that all perceptions actually have two attributes: the value and the uncertainty. Locally, we had some dealings a while back with a company who seemed to get much more powerful results from their neural nets by making all values complex, value being theta, uncertainy being r. There may be something to that in the psychological case, too.

To speak of the uncertainty of a perception by itself is to assume that we have a way of checking to see if it's correct. But the only way we have of doing that is to compare one perception with another one. I can say that perception A is uncertain relative to perception B in the sense that the one can be used to predict the other with some long-term degree of success, so perhaps we can classify perceptions in this way, finding some set that we trust more than others. But once we are down to the "most trusted set of perceptions," we can't measure uncertainty any more. We trust them or we don't. There's no absolute measure of uncertainty, is there? To assume that there is implies that we can be certain about something.

On a related subject, there's a big difference between prediction and control. To predict that something will happen always carries the assumption that one has no way of influencing the outcome. You throw the dice and wait for them to settle down, and you know the odds of making your point. But if we allow the player to influence the outcome, the odds change drastically -- the uncertainty in the outcome, or perhaps I should say the long-term variability in the outcomes, shrinks very close to zero. If I want to throw a seven, I toss the dice onto the table, then reach out and turn them until they show a three and a four, or a five and a two, or a six and a one. Essentially every time I throw them (but the last time), they will end up showing seven. The other players, of course, will all have disappeared, or I will find myself coming to rest with my nose in the gutter somewhere, before the last "influence" can be exerted.

Allowing influence of the outcome assures that the desired outcome will occur nearly every time; the uncertainty all but disappears. However, this is only the uncertainty of achieving the reference condition. Achieving the reference condition for a perception does not tell us what we have done to the outside world in the process of controlling the perception. That remains unknown, though we can still speak of side-effects on other perceptions. In terms of layered protocol theory, we can cause our perception of the other person's action to approach and remain near a reference condition even with no knowlege of what the other person's internal state actually is. We simply keep varying our action until our perception matches our reference. It doesn't matter whether our theory about what the other believes is correct or incorrect. We don't even have to have a theory, though having a theory with some relation to reality might make control faster and more precise.

Finally, all this implies that we need a probabilistic measure of perception, as you suggest, but mainly in order to test theories and only secondarily to achieve reference conditions. If we have a clear reference condition for a perception-- that a certain window be closed, for example -- it is almost certain that we will achieve it. How we will achieve it, however, is only roughly predictable, as Rick Marken has pointed out, because if the environment does not allow one way of achieving it (asking another person to close the window) we will find another way (asking a different person or closing it ourselves). The uncertainty of the outcome is inversely related to how much it matters to us. That is true of control, but not true of prediction.

Given a measure of the uncertainty of a prediction, we can then calculate the uncertainty of a chain of dependent predictions (where the probability of any statement depends on the probability that a previous statement is true). If the statements are about beliefs, and each belief proves to be true in terms of predicting some other perceptions eight times out of ten, then the probability that a belief about a belief about a belief will predict correctly is 51.2%, about the same as a coin-toss. But that doesn't matter, because if we are controlling an important outcome that depends on the third order of belief, we will alter our actions until the outcome is 99.9% sure to happen.

Best,

Bill P.

···

--
No virus found in this outgoing message.
Checked by AVG. Version: 7.5.519 / Virus Database: 269.21.7/1332 - Release Date: 3/17/2008 10:48 AM

13

[From Martin Taylor 2008.03.18.09.58]

[From Rick Marken (2008.03.18.1200)]

with a note to [Bill Powers (2008.03.18.0215 MDGT)] in the middle

> Martin Taylor (2008.03.18.01.31)_-
> We can control for achieving a reference value

  of that perception of uncertainty, too, by acting to get more
  information.

Yes, I agree. But now you are controlling for another perception: certainty.

Could you suggest what might be the inputs to an input function that has certainty (or uncertainty) as its output, and how that perception of uncertainty is kept linked with the perception that is certain/uncertain?

> Are you suggesting the we never are uncertain?

No. I'm suggesting that certainly is a perceptual variable itself,
separate from other perceptions, like my perception of the letters in
this sentence.

Ah, so you have a general perception of uncertainty, not uncertainty about anything in particular, do you? That avoids the question of keepingthe uncertainty linked to any particular uncertain perception. Even so, one must ask what would be the inputs to the input function of its control system.

> To make this into a useful observation certainly does require a

  working model.

Before you make the model I think you should describe the phenomenon
to be explained by the model. Perhaps control of a probabilistic
variable; you move a handle to make an event more or less "certain".
See if a person can control for a reference amount of certainly (or
probability of occurrence). The disturbance would be an independent
influence on the certainty of the event. The subject would be asked to
control the event at a reference level of certainty.

At an everyday level, we do this all the time. One of the big problems of road design is that people seem to drive to a perceived level of risk. If you make the physical design of the road seem safer, people drive faster. The design problem is to make the road seem riskier than its physical design should make it, so that the objective risk becomes less than the perceived risk, as a kind of illusion. When we go to vote, we don't go into every detail of each candidate's pluses and minuses; we learn enough that we can decide on the preferred candidate. When we choose a fruit to eat, we look carefully enough at it to see that it isn't obviously rotten and is sufficiently ripe; we don't take it to the lab and perform a thousand biological tests for hidden infections. Controlling risk by acquiring just enough information is a part of normal life.

As for doing an experiment, it's the fuzziness rather than the probability that I find more interesting. Setting the focus is an analogy or metaphor for it. The question is one's uncertainty about what is, rather than the question of prediction. Prediction is a different issue (this is partly in response to [Bill Powers (2008.03.18.0215 MDGT)], as is accuracy. Accuracy requires an independent measure, whereas uncertainty is an internal perception requiring no calibration from outside, although there is also an external measure of uncertainty, just as there is an external measure of length to go along with the perception of length.

There used to be an interesting demonstration by Bouma at the Evoluon in Eindhoven (which unfortunately no longer exists, I believe). My wife and I used it in our 1983 book on the Psychology of Reading. Bouma arranged that a piece of handwritten text was displayed so out of focus that you could barely see the separation of lines. When the visitor pushed a button, it came slowly into better focus, until at some point the text could be easily read. But then the focus continued to improve, and the visitor could see the details of the letters -- but it turned out that there were no letters, only sinusoidal wiggles with lines projecting up and down at appropriate places. In that case, the uncertainty of the text perception went from high to very low, and then back to high, as the focus improved.

You get a similar effect with the classic "Lincoln Portrait" shown with very large pixels. If the picture is very blurred, it's all just a fuzzy blob. With slightly better focus, it's clearly Lincoln, but if the focus is good, you can't see a face. You see only the blocks of the big pixels.

In both those cases, the higher-level perception is built from uncertain values of lower-level perceptions. Values appropriate to a meaningful higher-level perception are within the range of uncertainty of the lower-level perceptions when that uncertainty is moderate, but not when the loer-level uncertainties are low, and when the lower-level uncertainties are too high, there are so many higher-level possibilities that none are seen as being THE perception; one strains to see what is there by gathering more data if possible.

I guess that's one knd of experiment. I never have done it experimentally, but I've often tried blurring my vision to see something clearly that is obscured by fine noise, or that is pixellated.

None of that is equivalent to making a model, which is what should be done before trying an experiment with humans.

Martin

14

[From Bill Powers (2008.03.18.1030 MDT)]

Martin Taylor 2008.03.18.09.58--

[replying to Marken]
Ah, so you have a general perception of uncertainty, not uncertainty about anything in particular, do you? That avoids the question of keepingthe uncertainty linked to any particular uncertain perception. Even so, one must ask what would be the inputs to the input function of its control system.

Rick has a point. There's a difference between having an uncertain perception and perceiving that it is uncertain. It's like finally realizing that the sun has set and it's time to turn on the reading lamp because the print is getting too hard to see. We tend to go on struggling to get the signal out of the noise without being conscious that the problem is fuzziness. All we're aware of is the increase in effort, not the reason for it.

Best,

Bill P.

···

--
No virus found in this outgoing message.
Checked by AVG. Version: 7.5.519 / Virus Database: 269.21.7/1333 - Release Date: 3/18/2008 8:10 AM

15

[From Rick Marken (2008.03.18.1030)]

Martin Taylor (2008.03.18.09.58) --

Could you suggest what might be the inputs to an input function that
has certainty (or uncertainty) as its output, and how that perception
of uncertainty is kept linked with the perception that is
certain/uncertain?

See Bill's post [Bill Powers (2008.03.18.1030 MDT)] which is
completely consistent with my view of "certainty" perception. My own
answer to you question would be, first, it depends on what you think
the perception of certainty is. My point is just that I think of
certainty as a perception that is _about_ other perceptions. If
certainty is like probability then a perceptual function that
perceives the certainty of a perception p is one that computes an
estimate of the probability of p, possibly by continuously counting
the frequency of a particular value of p relative to the frequency of
all values of p integrated over some interval. But determining what
"certainty" is is why I think you should first show what the
_phenomenon_ of certainty is; that will help me understand what you
mean by "certainty" and then we can start measuring it and determining
a person's ability to control it.

Ah, so you have a general perception of uncertainty, not uncertainty
about anything in particular, do you?

No. I think of uncertainty as being a perception about other
perceptions. So I can perceive some future perceptions (like the
perception of me entering grades) as very certain and others (like me
receiving a check for $1 million) as very uncertain.

At an everyday level, we do this all the time. One of the big
problems of road design is that people seem to drive to a perceived
level of risk.

They seem to but do they really? You have to test this before you can
say it's true. I don't believe it myself. I don't control for "risk"
when I drive; I control for speed, being in my lane, getting to my
destination at a particular time, etc.

If you make the physical design of the road seem
safer, people drive faster.

This is speculation. I won't believe it until I see some convincing
tests to show that people control for "risk".

Controlling risk by acquiring just enough information is a part of normal life.

So you say. How about some data. Almost everything I do could be
_described_ as controlling for risk; I wash my hands to reduce the
risk of disease, I eat to avoid the risk of starvation, I work to
avoid the risk of poverty. I am nice to people and voted liberally to
avoid the risk of god's judgment. I'm not sure I'm really controlling
for risk in all (or possibly any) these cases.

In both those cases, the higher-level perception is built from
uncertain values of lower-level perceptions.

The ability to perceive the block Lincoln can be explained quite well
in terms of differential spatial frequency perception. I don't see
what uncertainly has to do with it?

None of that is equivalent to making a model, which is what should be
done before trying an experiment with humans.

That's where we differ. As you know, my motto is "Phenomena Phirst".
The attraction of PCT for me has always been that the model is
directly tied to -- and accurately explains -- the phenomenon of
control in living systems. I suppose you could do the modeling first,
but it seems to me that when people do it that way the necessary
observations (of the phenomena) are rarely made.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

16

[Martin Taylor 2008.03.21.23.45]

[From Bill Powers (2008.03.18.1030 MDT)]

Martin Taylor 2008.03.18.09.58--

[replying to Marken]
Ah, so you have a general perception of uncertainty, not uncertainty about anything in particular, do you? That avoids the question of keepingthe uncertainty linked to any particular uncertain perception. Even so, one must ask what would be the inputs to the input function of its control system.

Rick has a point. There's a difference between having an uncertain perception and perceiving that it is uncertain. It's like finally realizing that the sun has set and it's time to turn on the reading lamp because the print is getting too hard to see. We tend to go on struggling to get the signal out of the noise without being conscious that the problem is fuzziness. All we're aware of is the increase in effort, not the reason for it.

Actually, I think that was MY point. In your example, you perceive that a particular perception is uncertain. My question is how the uncertainty of a perception is linked to that perception. Rick seemed to be saying that he had a global feeling of uncertainty, not linked to any particular perception.

I don't think you can even have an "uncertain perception". A perception you have is whatever it is. But you can perceive that a perception is uncertain. "It's like finally realizing that the sun has set and it's time to turn on the reading lamp because the print is getting too hard to see."

However, I don't follow why we would go turn on the light if we were not aware that the print was getting fuzzy -- "We tend to go on struggling to get the signal out of the noise without being conscious that the problem is fuzziness."

Martin

17

[From Bjorn Simonsen (2008.03.22,09:05 EUST)]
Martin Taylor 2008.03.21.23.45

Actually, I think that was MY point. In your example, you perceive
that a particular perception is uncertain. My question is how the
uncertainty of a perception is linked to that perception. Rick seemed
to be saying that he had a global feeling of uncertainty, not linked
to any particular perception.

When I experience a conflict, I at the same time perceive a feeling.
That is the (idea or thought) perception of my body when it is in the
certain conflict state.
Maybe it is the same feeling I perceive when I am uncertain.

The difference between beeing uncertain and having an uncertain
perception is where in the environment the disturbance comes from. When
the disturbance comes from outside the body, the feeling is an uncertain
perception. When the disturbance comes from inside the body (emotion),
the perception is uncertain.

What about these thoughts?

I don't think you can even have an "uncertain perception".
When I look at the word "finally", I as a foreigner have an uncertain
perception. Shall the word be written "finally" or "finaly".

Maybe Rick's global feeling of uncertainty is the same "global" feeling
I often have when I perceive a conflict.

bjorn

18

[From Bill Powers (2008.03.22.0832 MDT)]

[Martin Taylor 2008.03.21.23.45]
I don't think you can even have an "uncertain perception". A perception you have is whatever it is.

When you see a snowy TV picture, your perception of the program content is uncertain because of the snow. You may think one thing is happening when, unbeknownst to you, you would see something quite different without the interference, or see details that you are missing.

However, I don't follow why we would go turn on the light if we were not aware that the print was getting fuzzy -- "We tend to go on struggling to get the signal out of the noise without being conscious that the problem is fuzziness."

It's perception of the struggle that finally alerts us to the fact that the perception of the print has degraded. Of course we might notice anyway if for any reason we back off or go up a level. Someone passing by might say, "How can you read in that light?"

Best,

Bill P.

···

--
No virus found in this outgoing message.
Checked by AVG. Version: 7.5.519 / Virus Database: 269.21.8/1338 - Release Date: 3/21/2008 5:52 PM

19

[From Rick Marken (2008.03.22.0940)]

I forgot to wish happy birthday to Bach (JS, of course, the only one
that matters) yesterday. I hope everyone played a two or three part
invention in his honor.

Martin Taylor (2008.03.21.23.45)

Rick seemed to be saying that he had a global feeling of uncertainty, not linked
to any particular perception.

Bjorn Simonsen (2008.03.22,09:05 EUST)

Maybe Rick's global feeling of uncertainty is the same "global" feeling
I often have when I perceive a conflict.

Before this turns into a "Rick claims to have invented the internet"
moment, I will just say that all I have said about "uncertainty" was
that the perception of uncertainty, whatever that is, is probably a
function of (ie, is linked to) other perceptions (which is basically
what Bill Powers is also saying).

Actually, I still have no idea what the perception of uncertainly is.
I was hoping Martin would give me an example of an experiment to
demonstrate control of uncertainly. I guess I have experienced
uncertainty in situations where I wasn't sure about whether I was
seeing (or hearing) one thing or another. In this case, the perception
of uncertainty seems like what Bill (and Bjorn) says it is; a
perception of the conflict about whether to say what is out there is
one thing or another. What I call the perception of uncertainty is a
perception that is a function of other perceptions (the perceptions of
the efforts involved in trying to say one thing or the other). I do
experience a "global" feeling of uncertainly but it is certainty but
it is definitely linked to (is a function of) other perceptions.

Best

Rick

···

--
Richard S. Marken PhD
rsmarken@gmail.com

20

[From Martin Taylor 2008.03.22.14.22]

[Rick Marken (2008.03.22.0940)]

Before this turns into a "Rick claims to have invented the internet"
moment, I will just say that all I have said about "uncertainty" was
that the perception of uncertainty, whatever that is, is probably a
function of (ie, is linked to) other perceptions (which is basically
what Bill Powers is also saying).

OK, I misinterpreted you, but that is what I am saying, too. My question all along has been how that linkage is made, within the HPCT hierarchy.

Actually, I still have no idea what the perception of uncertainly is.
I was hoping Martin would give me an example of an experiment to
demonstrate control of uncertainly.

I guess Bekesy audiometry is the standard one. The subject is asked to press a button that makes a sound go softer when pressed and louder when not pressed. The experiment starts with an audible sound, and the subject is told to press the button until he is just can't be sure he is hearing it, and then to press again when he is just sure he can hear it. The frequency of the tone is slowly increased as he is doing this. The result is a zig-zag trace covering the range of uncertainty about hearing the tone, between an upper limit of just sure he can hear it and just sure he can't hear it.

  I guess I have experienced
uncertainty in situations where I wasn't sure about whether I was
seeing (or hearing) one thing or another. In this case, the perception
of uncertainty seems like what Bill (and Bjorn) says it is; a
perception of the conflict about whether to say what is out there is
one thing or another.

Why would you SAY anything? To whom would you say it? The conflict is not between possible actions, but between possible category perceptions consistent with the data. Some uncertainties are of that kind: "am I perceiving A, B, ...Z". Others are of the kind "Is this door wide enough to take this chair?" That's a continuum kind of uncertainty.

What I call the perception of uncertainty is a
perception that is a function of other perceptions (the perceptions of
the efforts involved in trying to say one thing or the other).

And where does this perception of "effort" fit in the hierarchy? How is it linked to the data that could be consistent with the perceptions in conflict, or to the identities of those perceptions?

I do
experience a "global" feeling of uncertainly but it is certainty but
it is definitely linked to (is a function of) other perceptions.

I think you did an edit here, which didn't come otu right. I am uncertain as to what you meant.

Now, as to uncertainties induced by means like the Bouma fuzzy handwriting or the Lincoln portrait, we know very well the mechanisms that induce the uncertainties. Rick expressed the mechanism for the Lincoln portrait thus [Rick Marken (2008.03.18.1030)]: "The ability to perceive the block Lincoln can be explained quite well in terms of differential spatial frequency perception." That's a very bare statement, which needs fleshing out.

What happens in the Lincoln portrait is that a good clear portrait of Lincoln has a well defined spatial frequency spectrum. If you change the amplitude or, more importantly, the phase spectrum, the picture changes. When you blur a picture, you are simply reducing the amplitude of the higher frequencies without changing anything about the phase spectrum. If you blur enough, the low frequencies that are left are consistent with the original being a picture of almost anything -- very high uncertainty. If you sharpen it up a bit, the low frequencies add some not so low frequencies, and they are consistent with that range of frequencies from portraits, but from not much that isn't a portrait -- less uncertainty, but still uncertainty about whose portrait it is.

The interesting things happen when you continue to sharpen the image, bringing up higher and higher spatial frequencies, always without changing the phase spectrum. At some point, the frequencies that are left are consistent with the portrait being of Lincoln, but not consistent with a portrait of anyone else you know. You see Lincoln with little or no uncertainty.

But what happens next as you continue to sharpen the image made of very large pixels? The sharp edges of the pixels have lots of high spectral frequencies that are not present in a good portrait, and the even intensity across each pixel does not have the high-frequency variation that a clear portrait would have. As you further sharpen the image, the spatial sectrum becomes progressively less consistent with that of a portrait of Lincoln, and at some point it becomes so inconsistent that you can't see it as Lincoln any more.

The same thing happens with Bouma's fuzzy handwriting.

Martin