Posting problem (was Uncertainty...)

CSGnet Archive CSG2008
1

[Martin Taylor 2008.04.04.10.00]

[From Rick Marken (2008.04.03.2240)]

Still no part 2 at 10:38 PDT. What are you putting in those arguments?
I think you are just demonstrating how uncertain perceptions can
be;-)

I'm quite mystified now.

Translate that sentence into slightly more technical language.

My perception of uncertainty about the cause of the problem is at a high level. I control for it to be at a low level, but the usual environmental feedback path for getting more information about the problem (reducing the uncertainty) has been failing to reduce the uncertainty, and indeed my uncertainty has been increasing. So I have been using different output methods, in good PCT form.

History:

Initially I perceived (believed, imagined) that my ISP blocked the message because of some bit or character pattern that its algorithm took to be spam, since that has happened in the past with messages I have sent that contained pictures and PDFs, which are coded into ASCII for transmission. At that point I was not controlling for uncertainty with any high gain, since my uncertainty level about (my perception of) the problem was fairly low. I was controlling for getting the message posted, and since my normal output action (send to the CSGnet mailer) didn't work, and in the past I had been able to send the pictures and PDFs from my gmail account, I assumed that if I posted.

I was, hoever, also controlling for the uncertainty as to whether it was really my ISP that blocked the original posting, so I sent it to Rick both through my ISP and through my gmail account. When Rick told me that he received both, but had been unable to get CSGnet to accept the message, my uncertainty about my ISP was eliminated, leaving the uncertainty about where in the message the problem existed. I now had low uncertainty about my perception that the problem was with the CSGnet mailer, but had fairly high uncertainty about aspects of the message, included whether the problem was local or distributed, and if local, where it lay.

I reduced my uncertainty about whether the problem was local or distributed, and gained one bit if information about its location when Part 1 was successfully posted. At that point my uncertainty about the problem was roughly as follows: I had very low uncertainty about whether the problem was local or distributed (category), one bit less uncertainty about the conditional probability distribution "location given local", and very high uncertainty about the nature of the problem (exactly what bit pattern or letter sequence would have to be changed).

At that point I had an output mechanism that I imagined would be effective in continuing to control for getting the message posted while reducing the uncertainty as to the location and nature of the problem. I would post successive halves of the not-accepted parts of the message (with overlaps to help readers fit the chunks together), until there remained a small enough unaccepted segment that I could rewrite it and at least bring the error in the "perceive this message to be posted" control sysem near zero.

As it happens, when I posted Part 2.1 and Part 2.2, neither was accepted by the mailer. Since my uncertainty about the "local vs distributed" category perception was low (I was perceiving "local" with high probability), I also had reduced my uncertainty about the location perception quite dramatically. I believed (perceived with less than 1.0 probability) that the problem lay in the two short paragraphs of overlap that were common to both parts.

My action to complete the posting of the original message (the underlying control loop for all this) was to excise the overlapped section and write a paraphrase for both the quoted paragraph from Bill and my response to it, and send the complete Part 2 as a single posting. The resulting revision of Part 2 was not accepted by the mailer. That brings us to the present.

Now my uncertainty is higher about whether the problem is local or distributed, very high about the conditionals "if distributed, what could be the nature of the problem" and "if local, whether there are two instances of the problem in separate places". Because in 15 years and several hundred postings I have never had a message rejected, I perceive it to be highly unlikely that a random pattern that causes rejection would occur twice in one half of one message, and also highly unlikely that a paraphrase of two short paragraphs would contain the same problem as existed in the original paragraphs, I am left with very high uncertainty as to what or where the problem is, and whether there is more than one instance of the problem in the message.

Translation: I'm quite mystified now.

According to most versions of PCT, when actions fail to control a perception, it is likely that different actions will be tried. I am controlling several perceptions here, not all hierarchically connected. This is a skeleton list of some. For a detailed analysis the list would probably be ten times longer, or more.

1. A perception that readers of CSGnet (including me) come to understand the way perceptions of uncertainty fit into some structure of perceptual control systems, whether that be HPCT or something else.

2. A perception that other readers of CSGnet understand the proposition as to the above that I examine in the message I am trying to post, among others.

3. A perception that a particular message has been posted to CSGnet.

4. A perception of my uncertainty about what pattern in a message prevents it from being posted by the CSGnet mailer.

5. A perception of my uncertainty about where in this particular message there might be a problematic pattern.

Planned action (yes, we do plan, even if we understand perceptual control -- it's called controlling in imagination): try to post successive small chunks of the problematic message, with no overlapping segments, under a new subject line (controlling for a perception that keeping the same subject line increases the error in (1) above by making it hard for readers to follow the thread). With luck there will be one segment that fails.

The segments will have the subject line of this message, with the addition of Part 2.1, Part 2.2, .... If you are interested in reading the whole message, please concatenate these parts (if any arrive) with the Part 1 that did manage to get through the mailer's defences.

Martin

···

sent from the gmail account and asked Rick to post it, it would be

2

[From Bill Powers (2008.04.04.0948 MDT)]

Martin Taylor 2008.04.04.10.00--

I am left with very high uncertainty as to what or where the problem is, and whether there is more than one instance of the problem in the message.

I think the state you refer to as "high uncertainty" is what I call "being out of ideas." For me, uncertainty requires that alternatives be available to be uncertain about (there must be some distribution of possibilities). When all the alternatives are gone, all I am left with is error: I want a result and I don't have it. I don't call that uncertainty. It's just error.

The next post from you contains this:

There are no visible characters in those fields, and I don't think any could have been introduced, since each posting has been done by copy-pasting the binary chunk into a new message.

Spaces are not visible characters.

Did you really mean "binary chunk?" Do you really mean that you're pasting binary segments into ASCII text? I'm grasping at straws here....

And how come your posts lamenting about the ones that don't get through are getting through with no problem?

Best,

Bill P.

3

[Martin Taylor 2008.04.04.12.30]

[From Bill Powers (2008.04.04.0948 MDT)]

Martin Taylor 2008.04.04.10.00--

There are no visible characters in those fields, and I don't think any could have been introduced, since each posting has been done by copy-pasting the binary chunk into a new message.

Spaces are not visible characters.

Did you really mean "binary chunk?" Do you really mean that you're pasting binary segments into ASCII text? I'm grasping at straws here....

No. Binary as in binary subdivision.

And how come your posts lamenting about the ones that don't get through are getting through with no problem?

In over 15 years of posting to CSGnet, this is the first message that has had a problem. I don't anticipate problems with messages generated independently, and if I were really not controlling for a reduction in the uncertainty of what and where the problem lies, I would probably simply rewtie my response to your post in the light of subsequent dialogue.

I don't have time to do any more until later this afternoon. I've got to finish rounding up my tax-relevant papers and get them to the accountant in the next couple of hours.

Martin

4

Part 2.1 of problematic message [Martin Taylor 2008.03.31.17.30]

This follows directly after Part 1, with no overlap. I plan to send Part 2 in three sections this time. The quoted paragraphs are from [Bill Powers (2008.03.30.1045 MDT)]

···

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

So suppose we have a perception p which varies over a set of samples as sin(n) + R, where R is a random variable. This gives us the following series of readings: p = (0.7,2.2.,-0.4. 0.01, 1.06). and on and on. Is there anything uncertain about this series of readings? Rick says no; Martin sort of agrees, and I agree too. That is exactly the set of readings (perceptions) that was obtained. So where is any uncertainty?

A better question is "about what is there uncertainty?". And the answer is that it could be about many things, such as ...

We might say it is in the hypothesis that this series of numbers come a sine function plus a random variable. We might feel quite uncertain about that hypothesis, since any series of numbers at all would fit it, unless we were a lot more specific about the random variable's distribution and mean. Any uncertainty would arise only at higher levels of analysis. And it could be mistaken: maybe the series of numbers is an exact representation of exactly how some real mechanism behaved.

5

Re: Posting problem (was
Uncertainty…)
Part 2.2 of problematic message [Martin Taylor
2008.03.31.17.30]

This follows directly after Part 2.1, with no overlap. I plan to
send Part 2 in three sections this time. The quoted paragraphs are
from [Bill Powers (2008.03.30.1045
MDT)]

···

In the case of the
smoothed random disturbances I use in tracking experiments, that is
exactly the case. The disturbance is a computer-generated pattern that
varies through time in a way that seems unpredictable (and hence
uncertain) more than a few tenths of a second into the future – yet
if the “randomize” function isn’t used, running the program
a second time will generate exactly the same pattern of
disturbances. It is completely determinate and not uncertain at
all.

It is indeed uncertain, to the observer who
has not been told that the sequence will repeat and has not observed
it repeating often enough to have formed that hypothesis. It’s not
uncertain to the programmer. Uncertainty, like beauty, is in the mind
of the beholder.

So this tells us we
can be uncertain about a perception of a completely determinate
variable. The uncertainty arises not from the presumed thing being
perceived, but strictly from the observer’s inability to find any way
to predict it.

Yes, exactly! No uncertainty about
that:-)

6

Re: Posting problem (was
Uncertainty…)
Part 2.3 (final section) of problematic message [Martin Taylor
2008.03.31.17.30]

This follows directly after Part 2.2, with no overlap. I plan to
send Part 2 in three sections this time. The quoted paragraphs are
from [Bill Powers (2008.03.30.1045
MDT)]

···

How, then, would we
be able to generate a complex number with one number representing the
perceived value of the external variable, and the second number
representing the uncertainty in that external variable? We couldn’t,
because there is no uncertainty in the external variable to
detect.

No. The uncertainty is in the mechanism
that perceives the external variable. For example, in low light, the
statistics of the few photons creates appreciable variabiity in the
contrasts across an edge, and the location of the edge is uncertain
for this reason alone. The same is true for brief presentations. Given
an exact replay of the photon pattern and timing, the pattern and
timing of impulses from the rods and cones would probably not repeat.
Uncertainty is inherent in all measuring mechanisms, not just those
that depend on stochastic impulse trains – which should answer Rick’s
comment [Rick Marken (2008.03.31.0920)]

"For example, I have no idea how to
deal the the diagram in Martin’s “Postscript” post (Martin
Taylor 2008.03.30.16.27). Perhaps Martin could explain how this
would apply to a compensatory tracking task, for example. Where in the
actual tracking task is the distortion and noise that is shown
to

enter the PIF after the cursor position
(o+d) is displayed?"

The possibilty for perceiving uncertainties
are propagated and introduced at every level of processing, which is
not to say that uncertainties are actually perceived at all levels.
That’s the caveat I mentioned about whether uncertainties are
propagated along with the peceptual signals that are
uncertain.

Martin

7

[Martin Taylor 2008.04.05.10.51]

I have just sent Part 2 of the problematic message in three short sections with no overlaps. They have been constructed by copy-paste from the failed original message. I am waiting to see which, if any, survive teh CSGnet mailer. If two get through, I will paraphrase the missing one. If less than two get through, the problem becomes even more mysterious.

Martin

8

[Martin Taylor 2008.04.05.11.04]

[Martin Taylor 2008.04.05.10.51]

I have just sent Part 2 of the problematic message in three short sections with no overlaps. They have been constructed by copy-paste from the failed original message. I am waiting to see which, if any, survive teh CSGnet mailer. If two get through, I will paraphrase the missing one. If less than two get through, the problem becomes even more mysterious.

This get's ever weirder. All three parts got through, when all of the following failed to be posted:

The original message
Part 2 entire
Part 2.1 and Part 2.2 with an overlap of 2 short paragraphs, one quoted, one a response
Part 2 entire with the overlapped response para rewritten
Part 2 entire with the whole overlap rewritten

As a further test, I have tried reposting the original Part 2 entire, despite that you naw have it complete in three chunks. If it succeeds, something must have changed in the mailer. If it fails, there must be some non-local problem about the message itself.

.... an hour later ...

The full Part 2 message apparently still can't be posted, even though all its parts can. A real puzzle. But I'm not going to pursue it any further, since all of the material in the original message has not been sent.

Please, if you respond to te original message, do it as a single item, and don't respond to the four parts separately.

Martin

9

[From Bill Powers (2008.04.05.1050 MDT)]

Part 2.2 of problematic message [Martin Taylor 2008.03.31.17.30]

running the program a second
time will generate exactly the same pattern of disturbances. It is
completely determinate and not uncertain at all.

It is indeed uncertain, to the observer who has not been told that the
sequence will repeat and has not observed it repeating often enough to
have formed that hypothesis.

You left out a qualifier, so it seems that you are saying that the
pattern of disturbances, even though it repeats exactly, “is indeed
uncertain.” It is not in the least uncertain. It is not even
uncertain “To the observer who has not been told that the sequence
will repeat”. It is not uncertain at all. The missing qualifier
would add that the observer “believes that the uncertainty is in the
pattern rather than himself.” If the qualifier were not true, the
observer would say not “It is uncertain” but “I am
uncertain.”

It’s not uncertain to the
programmer. Uncertainty, like beauty, is in the mind of the
beholder.

Yes, and nowhere else. Language affords us the ability to say many things
contrary to fact, such as “It is uncertain to the observer,”
and thus conceal what is actually meant, “The observer is uncertain
about it.” In the present context, that distinction is
essential. “The programmer is not uncertain about the disturbance
pattern,” though he might be uncertain about whether the computer is
about to crash.

So this tells us we can be
uncertain about a perception of a completely determinate variable. The
uncertainty arises not from the presumed thing being perceived, but
strictly from the observer’s inability to find any way to predict
it.

Yes, exactly! No uncertainty about that:-)

And now you agree with me. I am confused. Do you think there is such a
thing as uncertainty that can be perceived, or do you think that
uncertainty IS a perception?

I remember a thread from a few years ago that contains echoes of the
present problem. I think it was Bruce Nevin who proposed (with some
support) that there is not a category level of perception and control,
but that category perceptions occur at every level, as if in a column
running parallel to the hierarchy and conducting category signals upward
along with all the other perceptual signals. Thus each perceptual signal
would be accompanied by a second signal indicating the category to which
that perceptual signal belongs.

Now you are proposing that for every perceptual signal at every level in
the hierarchy, there is a second signal accompanying it which carries
information about the uncertainty in the perceptual signal. I don’t like
this proposal any more than I liked Bruce’s. If you simply pause and ask
how you would design such an arrangement, I think you would see what is
wrong with it.

Suppose you want to generate a signal indicating the uncertainty in an
intensity signal from the retina. You would need an input function that
can receive a copy of the intensity signal, and compare that signal with
a signal representing the actual intensity of light falling on the
retina. By comparing the two signals this function could determine the
relationship between the intensity signal and the actual intensity. If
the intensity signal varied exactly as the actual intensity varied, the
uncertainty signal would be zero. As the signal’s variations departed
more and more from the actual intensity variations, the uncertainty
signal would become larger. I won’t even ask how the necessary
statistical calculations are carried out, or where. I’ll stipulate that
there are some cells in the retina that can do an equivalent analog
computation, though I’ll leave it to someone else to point them out.

The uncertainty signals from these cells would then run up the optic
nerve, paired with the corresponding intensity signals.

The real problem is locating the mechanism by which this
uncertainty-perceiver measures the actual intensity of the light falling
on the retina. We can assume that the intensity of light is detected only
by rods and cones (unless you want to propose some other kinds), so the
uncertainty-detector must be using some spare rods and cones. These,
however, would have to be different from the ones that generate intensity
perception signals, because they would have to generate signals that are
accurate measures of the light intensity at all brightness levels, down
to counts of individual photons. Otherwise we would just have another set
of intensity signals having unknown uncertainty.

Perhaps you have a better way of detecting the uncertainty in an
intensity signal. If so, I’d like to hear the details of your design,
because mine obviously doesn’t work. Since my argument can be applied to
the generation of uncertainty signals at any level in the hierarchy, I
can’t see how the uncertainty in ANY neural signal can actually be
measured directly. The only way I can think of doing this is to use
computing processes at the logic level, where we do mathematics and other
rule-driven things, and measure uncertainty by comparing perceptions of
different kinds from lower levels. If we then view all of experience from
this level, we will be able to perceive the amount of uncertainty in any
set of lower-level signals – but the perceptual signals indicating
uncertainty would then exist only at the logic level. And none of the
measurements could depend on knowing the actual state of whatever the
perceptual signal represents.

Best,

Bill P.

10

Re: Posting problem (was
Uncertainty…)
[Martin Taylor 2008.04.05.14.15]

I think I must be writing postings that are too long for the
message in them to get across. So this will be shooooort, consisting
almost entirely of quotes.

[From Bill Powers (2008.04.05.1050
MDT)]

Part 2.2 of problematic message [Martin Taylor 2008.03.31.17.30]

running the program a second time will
generate exactly the same pattern of disturbances. It is
completely determinate and not uncertain at all.

It is indeed uncertain, to the observer who has not been told that the
sequence will repeat and has not observed it repeating often enough to
have formed that hypothesis.

You left out a qualifier, so it seems that you are saying that the
pattern of disturbances, even though it repeats exactly, “is
indeed uncertain.” It is not in the least uncertain.

How do you know, if not from your perception of uncertainty about
it? You write as though the sentence “It is not in the least
uncertain” made some sense. You can’t (legitimately) take a
Newtonian go-like view of the world and at the same time assert that
all we know is our perceptions.

It is not even uncertain “To the
observer who has not been told that the sequence will repeat”. It
is not uncertain at all. The missing qualifier would add that the
observer “believes that the uncertainty is in the pattern rather
than himself.” If the qualifier were not true, the observer would
say not “It is uncertain” but “I am
uncertain.”

It’s not uncertain to the
programmer. Uncertainty, like beauty, is in the mind of the
beholder.

Yes, and nowhere else.

Yes, YES, YES, YES !!!

The rest of your message therefore needs little comment. We are
talking mechanism, nothing else, if you accept that at least sometimes
one does perceive uncertainty.

Let us at least TRY to remember that there IS no absolute
uncertainty in an objective sense. Never.

By the way, if I say “X is uncertain to me” how do you
perceive that as different from my saying “I am uncertain about
X”? If you can see any difference other than the formal register,
your use of language differs significantly from mine. I have a tiny
suspicion that you are hanging onto straws to show that what I say
must be wrong.

Martin

11

[From Bill Powers (2008.04.05.1512 MDT)]

Martin Taylor 2008.04.05.14.15 –

Part 2.2 of problematic message
[Martin Taylor 2008.03.31.17.30]

running the program a second
time will generate exactly the same pattern of disturbances. It is
completely determinate and not uncertain at all.

It is indeed uncertain, to the observer who has not been told that the
sequence will repeat and has not observed it repeating often enough to
have formed that hypothesis.

You left out a qualifier, so it seems that you are saying that the
pattern of disturbances, even though it repeats exactly, “is indeed
uncertain.” It is not in the least uncertain.

How do you know, if not from your perception of uncertainty about
it?

I do not doubt that we have perceptions that we call
uncertainty.

You write as though the
sentence “It is not in the least uncertain” made some sense.
You can’t (legitimately) take a Newtonian go-like view of the world and
at the same time assert that all we know is our
perceptions.

If I know that a disturbance pattern is generated twice in the same way,
I am justified in calling it deterministic and not uncertain. That is, of
course, a report on a perception. What is not? I could have said,
“Having written the program, I am not in the least uncertain about
that pattern of disturbances.” Perhaps, in that context, I should
have put it that way.

But all this is beginning to seem moot, since you are apparently far more
of my opinion about uncertainty than I thought you were. The problem is
with the layered protocols; what I believe that you believe, at least,
and perhaps one more half-cycle beyond that.

Your previous writings have seemed to propose that for every perceptual
signal in the hierarchy, there is another signal accompanying it that
carries uncertainty information about it. That implies that somewhere
down at the bottom, or at several stages of the process, there is an
uncertainty sensor of some kind. And that implies that there is some
uncertainty to sense at levels of perception where one would not expect
to find statistical analysis going on.

The point farther above has to do with language and the way we use it to
equivocate. Viz:

By the way, if I say “X is
uncertain to me” how do you perceive that as different from my
saying “I am uncertain about X”? If you can see any difference
other than the formal register, your use of language differs
significantly from mine. I have a tiny suspicion that you are hanging
onto straws to show that what I say must be wrong.

I do see them as different; the first is a way of implying that there is
something about X that tells me it is uncertain; the second says that
whatever X is, it is something in me that is generating the uncertainty.
I’m sure these nuances of language are not lost on you. If I say,
“He doesn’t seem trustworthy to me,” I am dutifully
acknowledging that I am interpreting appearances, but I am also asserting
that the appearances are there, and inviting you to see them, too. That
puts more of an onus on the object of this comment than, for example,
saying “I’m feeling distrustful of everyone today, including
him.” The “to me” is not a sufficient disclaimer to negate
the assertion “X is uncertain.”

I will be interested to see what you say about my comments about the kind
of design needed to provide the “complex” perceptual
signal.

···

==============================

Regarding your mysterious posting problems, I suggest the following. (1).
Write down all the steps you have taken and the hypotheses behind
them.

(2) CROSS THEM ALL OUT. Whatever the problem or the solution, it was none
of those.

Best,

Bill P.

12

[Martin Taylor 2008.04.07.17.47]

[From Bill Powers (2008.04.05.1512 MDT)]

If I know that a disturbance pattern is generated twice in the same way, I am justified in calling it deterministic and not uncertain.

Yes, and equally, if you don't know that, you are justified in being uncertain about it. Even if you do know that it is generated algorithmically, hoever, unless you independently run the algorithm, you my be uncertain about the value of successive samples.

That is, of course, a report on a perception. What is not? I could have said, "Having written the program, I am not in the least uncertain about that pattern of disturbances." Perhaps, in that context, I should have put it that way.

I would have demurred if you had. My preference would have been: "Having written the program, I am not the least bit uncertain that I could rederive the pattern of disturbances if I wanted to, but since I have not rederived it, I am uncertain about what this next sample is going to be."

Your previous writings have seemed to propose that for every perceptual signal in the hierarchy, there is another signal accompanying it that carries uncertainty information about it.

I have put that forward as a possibiity to be examined, yes. It's not a worked out theory, but a spontaneous notion that came from two sources: (1) your initial comment about belief being uncertain perception, which I saw as applying aequally at all levels of perception, and (2) my memory of the considerable imporvement that including uncertainty along with value in a complex number representation made in the functioning of perceptron-like structures (the same structure as the perceptual side of the HPCT hierarchy).

Even in that guise, I have not thought of it as the only way uncertainty about a particular perception could be derived. In my mind, if such a co-transmission exists, the uncertainty perception at one level would be just another input to the PIF that would compute the uncertainty at the next level.

That implies that somewhere down at the bottom, or at several stages of the process, there is an uncertainty sensor of some kind. And that implies that there is some uncertainty to sense at levels of perception where one would not expect to find statistical analysis going on.

Is there such a level? Surely statistical analysis MUST be going on at the level of rods, cones, and hair cells, must it not? At least it must if firing rates are significant, though perhaps not if it is only timings relative the the firings of neighbouring sensors. Even then, some statistics mus be involved. At what level in teh nervous system could statistics be dispensed with? Perhaps at category level and above? Even there I think there is a problem when the inputs suggest a perceptual analogue value near a category boundary.

The point farther above has to do with language and the way we use it to equivocate. Viz:

By the way, if I say "X is uncertain to me" how do you perceive that as different from my saying "I am uncertain about X"? If you can see any difference other than the formal register, your use of language differs significantly from mine. I have a tiny suspicion that you are hanging onto straws to show that what I say must be wrong.

I do see them as different; the first is a way of implying that there is something about X that tells me it is uncertain; the second says that whatever X is, it is something in me that is generating the uncertainty. I'm sure these nuances of language are not lost on you.

I am not sure about that, but I am sure I wouldn't have used those wordings to signify that kind of distinction.

If I say, "He doesn't seem trustworthy to me," I am dutifully acknowledging that I am interpreting appearances, but I am also asserting that the appearances are there, and inviting you to see them, too. That puts more of an onus on the object of this comment than, for example, saying "I'm feeling distrustful of everyone today, including him." The "to me" is not a sufficient disclaimer to negate the assertion "X is uncertain."

Those two statements are indeed different. My problem is seeing the difference between "He doesn't seem trustworthy to me" and "I see him as untrustworthy".

I will be interested to see what you say about my comments about the kind of design needed to provide the "complex" perceptual signal.

I take it you refer to [Bill Powers (2008.04.05.1050 MDT)]. I was trying to take some moments between trying to prepare for a meeting in Oslo next week to answer much of the same in response to [Tracy B. Harms (2008 4 3 8 30)]. There's a problem, but I don't think you address it, at least not in the paragraph that starts: "Suppose you want to generate a signal indicating the uncertainty in an intensity signal from the retina. You would need an input function that can receive a copy of the intensity signal, and compare that signal with a signal representing the actual intensity of light falling on the retina."

You tangentially touch on the problem, which is that to assess uncertainty you need more than one scalar data stream, but to say "you need X" is to dismiss unexamined any other possibility. There are several. Control itself provides one, spatially neighbouring sensors provide another.

==============================
Regarding your mysterious posting problems, I suggest the following. (1). Write down all the steps you have taken and the hypotheses behind them.
(2) CROSS THEM ALL OUT. Whatever the problem or the solution, it was none of those.

I have given up on that problem, but if I ever have evidence that other messages have been lost equally mysteriously, perhaps the postmaster at CSGnet will be able to do some serious investigation. It seems to me that from time to time a message appears on CSGnet along the lines of "I posted but it still hasn't appeared". I've always assumed some finger problem at the source, but maybe that hasn't always been the problem. Anyway, to quote one of Sir Humphrey's colleagues, "YP" (if I were talking to the postmaster).

Martin

13

[From Bill Powers (2008.04.08.0909 MDT)]

Martin Taylor 2008.04.07.17.47 --

[From Bill Powers (2008.04.05.1512 MDT)]

If I know that a disturbance pattern is generated twice in the same way, I am justified in calling it deterministic and not uncertain.

Yes, and equally, if you don't know that, you are justified in being uncertain about it. Even if you do know that it is generated algorithmically, hoever, unless you independently run the algorithm, you my be uncertain about the value of successive samples.

I'm not arguing with you: I agree that we can perceive uncertainty in ourselves. If we have any mismatches, it concerned attributing that uncertainty to a cause. But I'm not going to argue that we don't do that, either.

That is, of course, a report on a perception. What is not? I could have said, "Having written the program, I am not in the least uncertain about that pattern of disturbances." Perhaps, in that context, I should have put it that way.

I would have demurred if you had. My preference would have been: "Having written the program, I am not the least bit uncertain that I could rederive the pattern of disturbances if I wanted to, but since I have not rederived it, I am uncertain about what this next sample is going to be."

Now you're making uncertainty equivalent to the ability to predict something. I didn't say that I could predict the random pattern in detail; I just said I didn't feel any sense of uncertainty about it, since I know that I can recreate it at any time just by running the program again. I think your argument here leans toward Rick's position, which is that we don't feel uncertainty about anything until we try or are required to base some action on it (such as predicting its future value).

Your previous writings have seemed to propose that for every perceptual signal in the hierarchy, there is another signal accompanying it that carries uncertainty information about it.

I have put that forward as a possibiity to be examined, yes. It's not a worked out theory, but a spontaneous notion that came from two sources: (1) your initial comment about belief being uncertain perception, which I saw as applying equally at all levels of perception,

What this overlooks is that a perceptual input function of any order can receive perceptual signals from any lower level, so we can have, for example, a sequence of light intensities (heliograph). All that is required is that at some high level there be an ability to compute uncertainty and represent its degree as a perceptual signal, and this automatically allows perceiving the uncertainty aspect of any perceptual signals of lower level. It's not necessary to generate a new uncertainty signal at each lower level. Uncertainty is first of all the name of a category, and at higher levels a kind of computation taking place according to rules of logic and mathematics.

In line with the ideas behind MOL, if your awareness resides mostly at the logic level, then you would perceive the uncertainty computation not as a computation, but as an aspect of whatever part of the lower-order world you attend to. You project the perception at the level with which awareness is current identified into the lower-order world, seeing it as "out there" instead of "in here." If you're looking at configurations from the relationship level, you don't see the relationship as an interpretation that you, the looker, are contructing, but as a feature of the configurations: you see that one configuration IS, for example, larger than another one. You have to step back from that level, somehow, to see that "larger than" is a perception in itself, not part of the configurations or itself a configuration. This stepping-back is the process by which I identified -- very, very slowly -- the levels that I have defined.

The same, I propose, goes for uncertainty, though I'm uncertain about the level where we should put this perception. When you look at something or listen to something under high-noise conditions, you do not normally say "I am seeing a snowy picture (or hearing a white-noise masked word) and making a judgment about how much uncertain I feel about it." You say "That picture (or word) is hard to identify," as if "hard to identify" describes a property of the picture or word, instead of a property of the observing agent.

This phenomenon is the BASIS of the method of levels. The "disruptions" that we pick up on are moments when something about the current object of awareness stands out as separate from the focal subject matter; we notice for a moment that the favoritism we see being shown a co-worker is actually a feeling of jealousy on our part, and this puts a pause into the conversation, which normally we manage to patch over. In the method of levels, the guide says "What was that?" and keeps that momentary stepping-back from disappearing again.

and (2) my memory of the considerable imporvement that including uncertainty along with value in a complex number representation made in the functioning of perceptron-like structures (the same structure as the perceptual side of the HPCT hierarchy).

Well, they do that same sort of thing in "Modern control theory" using Kalman Filters, and manage to overlook the fact that there is more to random disturbances than a mean and standard deviation. A real control system can counteract all the effects of random disturbances up to some upper limit of the bandwidth of control, but the MCT approach just has to accept the noise as irreducible, and adjust the loop gain accordingly. I don't know how this works with perceptrons, but they seem to be used strictly as a means of categorizing perceptions, which is just one of eleven levels of perception and control that we are interested in.

Even in that guise, I have not thought of it as the only way uncertainty about a particular perception could be derived. In my mind, if such a co-transmission exists, the uncertainty perception at one level would be just another input to the PIF that would compute the uncertainty at the next level.

So what are the other means of "detecting" uncertainty?

That implies that somewhere down at the bottom, or at several stages of the process, there is an uncertainty sensor of some kind. And that implies that there is some uncertainty to sense at levels of perception where one would not expect to find statistical analysis going on.

Is there such a level? Surely statistical analysis MUST be going on at the level of rods, cones, and hair cells, must it not?

No, of course not. How would the standard deviation be detected, and the shape of the distribution of the variations, and the probability of a prediction being correct? About the only statistical calculation I might see at this level is the perception of the mean value, which can be obtained through smoothing. All the other statistical variables are calculated using symbols and analytical mathematical rules, at a much higher level of organization.

At least it must if firing rates are significant, though perhaps not if it is only timings relative the the firings of neighbouring sensors.

So are you proposing that these timings are themselves separately represented as neural signals entering neural computing networks?

Even then, some statistics mus be involved. At what level in teh nervous system could statistics be dispensed with? Perhaps at category level and above?

No, at all the analog levels below. The analog levels are not made to work by analytical computations, but by physical laws that determine firing rates, gradients of excitation, and so on. They do not calculate probabilities or distributions or any such things, and I can't imagine how they could give rise to perceptual signals that represent such things. You, perhaps, can imagine THAT they do these things, but I doubt that you can imagine HOW they do these things. "And then a miracle occurs"?

By the way, if I say "X is uncertain to me" how do you perceive that as different from my saying "I am uncertain about X"? If you can see any difference other than the formal register, your use of language differs significantly from mine. I have a tiny suspicion that you are hanging onto straws to show that what I say must be wrong.

I do see them as different; the first is a way of implying that there is something about X that tells me it is uncertain; the second says that whatever X is, it is something in me that is generating the uncertainty. I'm sure these nuances of language are not lost on you.

I am not sure about that, but I am sure I wouldn't have used those wordings to signify that kind of distinction.

Then indeed we do perceive language very differently.

Those two statements are indeed different. My problem is seeing the difference between "He doesn't seem trustworthy to me" and "I see him as untrustworthy".

The first part attributes the seeming (or not-seeming) to him; the second attributes it to the "I", the observer. This is called "spin," adjusting the nuances of seemingly neutral communications to leave desired impressions.

You tangentially touch on the problem, which is that to assess uncertainty you need more than one scalar data stream, but to say "you need X" is to dismiss unexamined any other possibility. There are several. Control itself provides one, spatially neighbouring sensors provide another.

It's the required computations that I am concerned with, not so much that which provides data to them. Perceptions generated at a low level can't take advantage of any computational abilities that exist at higher levels. At the sensation level, for example, the most complex analog computation is weighted summation of intensity magnitudes, perhaps smoothed. Perhaps you wish to propose other computations, but if so you had better say what they are, or might be. Otherwise I will continue to see your proposals as attributing very high-level properties to low-level systems.

Best,

Bill P.