Tar baby, audiometry

[From Rick Marken (980216.0930)]

Me:

PCT shows that the I/O relationships measured in standard
psychophysical studies are actually O/I relationships (that's
the behavioral illusion).

Martin Taylor (980215 18:50) --

Wrong, as usual.

What is wrong about it? Please give a reason.

But it would be easier to dispute such claims if they were
accompanied by reasons rather than simple re-asserted.

We have been discussing the reasons for the last several days.
Briefly, the reason is that the stimulus (I) in psychophysical
experiemts is a disturbance to some perception. The obsrverd
output (O) probably also has an effect on the same perception.
Any observed relationship between O and I thus reflects (at best)
the feedback connection between O and I (the O/I relationship),
not the forward equation relating controlled perception (a
function of both O and I) to output, O.

Me:

Also, the "I" variable manipulated in these experiments is not
necessarily the same as the perceptual input to the system

Martin:

The input to any function is not the same as its output. So
what's your point?

That wasn't my point. My point was simply that the perception
is a function of many variables: p = f(I,O, I2, ...).

Me:

(this was Bill's point)

Martin:

And what was his point?

Just what I said. In concrete terms, if p is a function of
intensity and frequency and you manipulate only intensity (I)
then the observed I/O relationship is not only subject to the
behavioral illusion, I is not even a correct analog of the
presumed perceptual "cause" of the output.

"The system"? What "system?"

The control system involved in producing the behavior involved in
psychophysical experiments.

I still have great difficulty in understanding what it would benefit
PCT theory or practice to discover that the subject was controlling
for a self-image as a competent subject (or for satisfying the
experimenter, or ...).

The Test is not done to "benefit" PCT theory. It is done because
we know from control theory that _if_ the subject in a psychophysical
experiment is a control system, then the only way you can make sense
of the results of such experiments is by determining, as best as you
can, what perceptions the subject is controlling.

Me:

Maybe this is the bee in Oded's bonnet about the Test. Is this it
Oded? Do you think that I am claiming that the Test allows the
Tester to experience the same thing as the subject under study?
If so, let me reassure you that I know that the Test doesn't let
me do this.

Oded Maler (980216)-1

the claim resembles the point I was trying to make. While you
seem to accept it in principle, you seem not to be bothered by
it so much (at least not as much as you are bothered by
deficiencies of other experimental methoids).

What's to be bothered by? I am interested in determining which
ones of my own perceptions correspond to your controlled
perceptions. If I find that you control what I perceive as color
then that's enough for me; you may _actually_ experience "color" as
what I cexperience as "pitch", but who cares?

I see no reason to reject psychophysical experiments that give
some crude estimations of perceptual capabilites.

We don't reject psychophysical experiments becuase they give
"crude" estimates of perceptual capabilities. We reject
taking for granted that the I/O relationships observed in these
experiments correspond to the actual I/O characteristics of the
system. We do this because we know that, _if_ the subjects in
these experiments are control systems, then the observed I/O
relationships are _misleading_. All we are saying is that we
should _redo_ experiments like these in a way that takes into
consideration the possibility that the subjects are input
control systems.

I don't "reject" the results of conventional psychophysical (or
other conventional) experiments because I don't like them. What
may seem to you like a hostility to conventional research on
my part is really nothing more than an earnest desire to see
conventional behavioral researchers seriously consider the
possibility that the subjects of their investigations are input
control systems and that they try to understand the implications
of this possibility for how they go about studying these
investigations.

Bill Powers (980216.0310 MST) --

Nice!

Best

Rick

ยทยทยท

--

Richard S. Marken Phone or Fax: 310 474-0313
Life Learning Associates e-mail: rmarken@earthlink.net
http://home.earthlink.net/~rmarken/

(From Rick Marken (980216.0930)
> Oded Maler (980216)-1

> I see no reason to reject psychophysical experiments that give
> some crude estimations of perceptual capabilites.

We don't reject psychophysical experiments becuase they give
"crude" estimates of perceptual capabilities. We reject
taking for granted that the I/O relationships observed in these
experiments correspond to the actual I/O characteristics of the
system. We do this because we know that, _if_ the subjects in
these experiments are control systems, then the observed I/O
relationships are _misleading_. All we are saying is that we
should _redo_ experiments like these in a way that takes into
consideration the possibility that the subjects are input
control systems.

So one might look for the `minimal controlled disturbance' instead
of the `just noticeable difference'. `Common sense' would suggest that
they would be the same, but possibilty not, thinking of phenomena like
blindsight and the fact that the response time for disturbances to controlled
positions of levers, etc. is much faster than the standard `reaction
time' to `respond' to a discrete `stimulus'.

- Avery.Andrews@anu.edu.au

[From Bill Powers (980217.0230 MST)]

Avery Andrews (980217) --

So one might look for the `minimal controlled disturbance' instead
of the `just noticeable difference'. `Common sense' would suggest that
they would be the same, but possibilty not, thinking of phenomena like
blindsight and the fact that the response time for disturbances to
controlled positions of levers, etc. is much faster than the standard
reaction time' to `respond' to a discrete `stimulus'.

The underlying assumption is that one JND represents the same subjective
perceptual difference at all levels of stimulus input. This is, of course,
unprovable, and I see no reason to accept it without proof as a fact.

A more reasonable hypothesis is that a JND represents something like the
signal-to-noise ratio at a given signal level. If a perception is carried
by a bundle of redundant trains of nerve impulses, the noise will be
roughly the square root of the signal level (Poisson distribution). If we
assume that a JND is a constant fraction of the noise, then we would expect
JND to increase with signal level. If that is the true situation, but we
assume JND to be constant, we will arrive at a roughly logarithmic
relationship of perceptual signal to stimulus magnitude -- assuming that
the conscious response to the perceptual signal magnitude is linear.

So the answer you get depends on what you assume. The only reason for
assuming that the JND is subjectively constant is in order to draw a
conclusion about something unmeasurable. Only people who accept that basic
assumption about JND will agree on the relation between the stimulus and
the subjective apprehension of its magnitude. People who make some other
assumption will deduce a different relation. In no case can anyone prove
that the assumption (whatever one is preferred) is true.

The PCT model doesn't depend on linearity. If perceptual signals are some
nonlinear function of physical measures of the stimulus, then so are
reference signals. If I match a perception to a memory of how that
perception was some time ago, the memory includes the same nonlinearities
as the perceptual signal, so the scale of subjective magnitude is irrelevant.

The JND approach requires adding JND's end-to-end in order to plot the
magnitude of perception against the magnitude of stimulus. The minimal-
controlled-disturbance approach that you suggest doesn't require building
up a picture of the total disturance from adding up increments, since we
know what the total disturbance and the total opposing action are without
having to infer either one.

By the way, "controlled disturance" is not what you mean. Disturbances are
not controlled; they are independent variables. You're the linguist; there
must be some way to say what you mean.

Best,

Bill P.

[From Bruce Nevin (980217.2223)]

Bill Powers (980217.0230 MST)--

Avery Andrews (980217) --

So one might look for the `minimal controlled disturbance' instead
of the `just noticeable difference'.

By the way, "controlled disturance" is not what you mean. Disturbances are
not controlled; they are independent variables. You're the linguist; there
must be some way to say what you mean.

Resisted disturbance?

[Avery Andrews (980217 Eastern Oz)]
(Bill Powers (980217.0230 MST)

Avery Andrews (980217) --

>So one might look for the `minimal controlled disturbance' instead
>of the `just noticeable difference'. `Common sense' would suggest that
>they would be the same, but possibilty not, thinking of phenomena like
>blindsight and the fact that the response time for disturbances to
>controlled positions of levers, etc. is much faster than the standard
>reaction time' to `respond' to a discrete `stimulus'.

The underlying assumption is that one JND represents the same subjective
perceptual difference at all levels of stimulus input. This is, of course,
unprovable, and I see no reason to accept it without proof as a fact.

Yes, and there's another bad assumption, which is intuitively that the
activity in which the experimental subject is engaged has no bearing on
the threshhold. a JND experiment could be seen as showing control of
a rather complicated perception where some of the inputs are the
sensory modality being investigated, and others are proprioceptions of
the activity of the subject (along the lines of `if the stimulus has
changed, I have pushed the button). It's not valid to assume that the
other contributers to this perception won't have any effect on the
apparent sensitivity.

A more reasonable hypothesis is that a JND represents something like the
signal-to-noise ratio at a given signal level. If a perception is carried
by a bundle of redundant trains of nerve impulses, the noise will be
roughly the square root of the signal level (Poisson distribution). If we
assume that a JND is a constant fraction of the noise, then we would expect
JND to increase with signal level. If that is the true situation, but we
assume JND to be constant, we will arrive at a roughly logarithmic
relationship of perceptual signal to stimulus magnitude -- assuming that
the conscious response to the perceptual signal magnitude is linear.

So the answer you get depends on what you assume. The only reason for
assuming that the JND is subjectively constant is in order to draw a
conclusion about something unmeasurable. Only people who accept that basic
assumption about JND will agree on the relation between the stimulus and
the subjective apprehension of its magnitude. People who make some other
assumption will deduce a different relation. In no case can anyone prove
that the assumption (whatever one is preferred) is true.

By the way, "controlled disturance" is not what you mean. Disturbances are
not controlled; they are independent variables. You're the linguist; there
must be some way to say what you mean.

Yes, `resisted disturbance', as suggested by Bruce Nevin, is what I meant.
An interesting example of scrambled lexical access.

Avery.Andrews@anu.edu.au