psychophysics (was Bridges)

CSGnet Archive CSG1998
1

[Martin Taylor 980104 11:50]

Bill Powers (980103.0418 MST)

I'm afraid the central point is getting lost in a fog of words in which
it is easy (for me) to get lost. Rather than responding point by point
to your long message, I will ask three questions, the answers to which may
assist me in responding effectively.

1. In standard HPCT, is it assumed that one can control the values of
a perception more finely than those values can be discriminated within
the perceptual signal?

2. In standard HPCT, is it assumed that the nature of an input function
changes, depending on what higher level perceptions are currently being
controlled?

3. If a light is flashed and a person says "I saw the light flash" is it
reasonable to assume that there is a perception that incorporates some
sensory data from the light?

To assure you that there are no tricks here, I argue that if the answers
to these questions (even if the answers are hedged) are "No", "No", "Yes,"
then the results of standard psychophysical experiments are useful to PCT.

ยทยทยท

----------------
To answer one point from your message, my earlier comment was a suggestion
that the answer to question 2 might be "Yes" in a future version of HPCT:

+Martin Taylor 971231 17:45
+The question arises as to whether the perceptual input functions operate
+the same way when the resulting perception is being controlled as when it
+isn't. This issue is not ordinarily considered within HPCT, since normally
+the perceptual input function is taken to be whatever it is, and only the
+magnitude of its output is controlled. But it is an issue, one that might
+invalidate the uncritical use of the results of psychophysical studies
+to assess the elements of related control loops.

To this, I add the fact that in the so-called "open loop" situations, we
actually are studying different control systems, not the same control
system with its EFF removed.

You are here arguing that in standard HPCT the answer to question 2
is "Yes", are you not?

Martin

2

[From Bill Powers (980104.1041 MST)]

Martin Taylor 980104 11:50--

1. In standard HPCT, is it assumed that one can control the values of
a perception more finely than those values can be discriminated within
the perceptual signal?

No. But it can control less finely; if the system is proportional rather
that purely integral, the steady state of the perceptual signal will not
exactly match the reference signal. It will be G/(1+G) of the reference
signal, where G is the static loop gain. So it is possible for a
perceptible difference to remain when the steady state is reached. The
actual ability to discriminate would have to be found by matching a model
to the data, deducing the loop gain, and calculating the appropriate
correction factor.

For purely integral systems, of course, the steady state is zero error, so
the correction would not have to be made. Is this information available
from any existing psychophysical experiments?

2. In standard HPCT, is it assumed that the nature of an input function
changes, depending on what higher level perceptions are currently being
controlled?

No. But if a higher system is involved, the overall perceptual function
that you see from outside includes properties of the higher perceptual
function. If the higher perceptual function has a greater threshold than
the lower, the threshold you see could be that of the higher system.

3. If a light is flashed and a person says "I saw the light flash" is it
reasonable to assume that there is a perception that incorporates some
sensory data from the light?

Yes. However, when controlling a lower-level perception with a lower-level
output function, the amount of light-change needed to generate an opposing
action may be less than the amount of light-change needed for the person to
say "I saw the light flash."

To assure you that there are no tricks here, I argue that if the answers
to these questions (even if the answers are hedged) are "No", "No", "Yes,"
then the results of standard psychophysical experiments are useful to PCT.

I never said they weren't, at least some of them. But PCT brings in
considerations that standard experiments are not likely to have taken into
account.

----------------
To answer one point from your message, my earlier comment was a suggestion
that the answer to question 2 might be "Yes" in a future version of HPCT:

Yes, it might. Control of lower systems by varying their parameters is a
future part of HPCT that I have left on the shelf. That's where PCT should
put the subject of adaptive control. Your comment about future
re-evaluations of psychophysics that might result agrees with my opinion.

To this, I add the fact that in the so-called "open loop" situations, we
actually are studying different control systems, not the same control
system with its EFF removed.

You are here arguing that in standard HPCT the answer to question 2
is "Yes", are you not?

No, not at all. If you control a sound intensity perception with a
joystick, the output function includes the muscles you use to operate the
joystick, so the loop gain, lag, and other parameters will include the
properties of that output function. If you have an experimenter vary the
sound intensity while the person says "yes" and "no", different control
systems (using the same input information at the lowest level) are
involved. Also, since now a relationship between a sound intensity and a
"yes-no" answer is involved, a new perceptual function of a higher level
must respond to the signal from the lower system, so the apparent
thresholds and slopes could be different.

The only way to find out whether this difference makes a difference is to
do the appropriate experiments, in which both methods are used and the
results are compared. There may be a few psychophysics experiments in which
this comparison was not done, so at least we would have to reinvestigate
those.

Best,

Bill P.

3

[Martin Taylor 980109 01:00]

Bill Powers (980104.1041 MST)]

Martin Taylor 980104 11:50--

1. In standard HPCT, is it assumed that one can control the values of
a perception more finely than those values can be discriminated within
the perceptual signal?

No. But it can control less finely;
...
For purely integral systems, of course, the steady state is zero error, so
the correction would not have to be made. Is this information available
from any existing psychophysical experiments?

Not in the sense you want, so far as I know. But there are some relevant
psychophysical findings. They are on the increase of detectability with
increasing duration of presentation. There are contextual conditions
that matter for well-understood reasons, but generally speaking detectability
inproves for short durations (less than a few 10s of msec, if I remember
correctly--this is all from distant memory) at 6dB per doubling of the
presentation duration. For moderate durations detectability improves at
3dB per doubling of presentation duration, and for durations longer
than 250(?) msec, detectability does not improve with presentation
duration. So, one might think of an analogy with an integrator having
a leak, but the data seem, again from memory, to have three straight-line
regimes, rather than being the kind of asymptoting exponential one might
expect from a leaky integrator. This is for audition. I don't know about
vision, but from the way faint stars disappear when you look at them in
the twilight, I don't think the integration leak time-constant can be
very long.

For higher-level perceptions, I have no knowledge of experimental results.

Martin