NetNews Usenet Archive 1993 #1/NN_1993_1.iso/spool/bit/listserv/csgl/2127

Text File | 1993-01-12 | 8.8 KB | 199 lines

Comments: Gated by [e-mail]
Path: sparky!uunet!zaphod.mps.ohio-state.edu!darwin.sura.net!ukma!psuvax1!psuvm!auvm!COURIER4.AERO.ORG!MARKEN
Return-Path: <@VMD.CSO.UIUC.EDU,@vmd.cso.uiuc.edu:[e-mail]>
Posted-Date: Tue, 12 Jan 93 13:58:55 -0800
X400-Trace: US**AEROSPACE; arrival Tue, 12 Jan 93 13:58:55 -0800 action Relayed
P1-Message-Id: US**AEROSPACE; 930112215855
Ua-Content-Id: CSI NC V2.1b
Message-ID: <00030E68.MAI*[e-mail]>
Newsgroups: bit.listserv.csg-l
Date:         Tue, 12 Jan 1993 13:58:55 -0800
Sender:       "Control Systems Group Network (CSGnet)" <[e-mail]>
From:         [e-mail]
Subject: Re: Psychophysics
Lines: 183

[From Rick Marken (920112.1200)]

Martin Taylor (930112 14:00) --

> My question is twofold: (1) How would one do an experiment in
>psychophysics differently from a PCT viewpoint, and (2) can one use the
>results of classical psychophysics within PCT and if not why not?

In order to answer these questions we would first have to agree on what
psychophysics is and why we would want to do it. I think of psychophysics
as an attempt to find the relationship between perceptual variables, p, (the
psycho
part) and basic physical variables, e, (the physics part). So psychophysics is
 trying
to find the form of the function, f, that transforms physical variables, e,
into
perceptual variables, p. Two proposals for this function are Fechner's Law,
which
 says that p = k log e, and Steven's law that says p = e^k. There are other
aspects
of psychophysics (like multidimensional scaling, which is an attempt to find
the
dimensions of p independent of the physics of e) but I think the search for f
in
the relationship p = f(e) is the main one. Is this a fair description of the
nature
and goals of psychophysics from your perspective, Martin? I hope so; I will
proceed as though it is.

From this point of view, psychophysics is concerned with discovering the
nature of what we call the "perceptual function" in PCT. The PCT concept
of e and p is a bit broader than the conventional psychophysical view (maybe;
although a nice influence of J.J. Gibson was to get people to see that
p could be a representation of complex relationships between simple physical
variables. In PCT, p is what we experience -- from colors to religions
systems--
so we accept that f could be an enormously complex function of quite abstract
aspects of e -- that is, aspects of e that are not "really there" from the
point of
view of physics).

So the basic goal of psychophysics is certainly consistent with the PCT
modelling
goal of understanding the nature of the perceptual function. But there are a
couple
of problems right off the bat; first, psychophysics assumes that there is a
perception
that corresponds to a physical variable (measured by some instrument); that
is, psycho-
physics works "outside in" (taking "physics" as reality and trying to see how
this reality
relates to a person's representation thereof). So the psychophysicist has a
meter that
measures "amps" and decides to find the relationship between this real,
physical
variable and the perception it produces. PCT goes in the opposite direction --
 "inside
out". In PCT, the perception is the only reality; the goal would be to find
out the
relationship between this experience and models of the causes of this
experience
(like physics). So it would never occur to a PCT person to do a psychophysical
experiment -- because such an experiment presumes that you already know what
is
being perceived -- and what is causing that perception.

The second problem with the psychophysical approach to determining perceptual
functions is that it tries to use the subject as an output generating meter --
 assuming
that the subject's responses (like the responses of the meter) are
proportional
to the perceptual input. In other words, psychophysics is based on the idea
that
o = g(f(e)) where g is the function that transforms perception (f(e)) into
responses,o.
This model ignores feedback effects of o on f(e). (Martin says that such
feedback effects are small (or non-existent) in the typical psychophysical
experiment:

>But with respect to influencing what perceptions will occur of the kind
>normally (in the real non-laboratory world) controlled through the signals
>now provided by the experimenter, the loop is broken.  None of the subject's
>responses have any influence on whether the tone will come in the first
>interval or the second on the next trial.

This analysis assumes that  the subject's responses are simply a function of
aspects of the tone perception that cannot be influence by the subject, such
as
tone amplitude. But this implies that for some reason, in this particular
circumstance, the subject is willing (and able) to act as though tone
amplitude causes his or her responses. Far more likely, it seems to me, is
that
the subject's responses (as usual) are part of a loop controlling some
perception(s);
the experimenter simply is not aware of what the controlled perceptionss might
be. There is quite a bit of evidence in the literature to suggest that
responses are
not simply caused by the stimulus in such tasks. Signal detection theory-based
research shows how important are pay-offs and "feedback" information. So the
tone is clearly not just causing responses -- the subject is controlling
something
ABOUT the tone; and signal detection theory can be looked as as an attempt to
analyze the data in a way that shows what the causal relationship between tone

and response WOULD BE if the subject were not also controlling things ?about?
the tones (like the relationship between the occurance of the tone and the
response
given). Still, one wonders how successful this decomposition might be since
the data is generally so noisy -- probably because there has been no formal
attempt to determine the variable(s) that are actually being controlled by the

subject (and also because the subject has such poor control over whatever it
is
they are controlling).

>When a control system is actively controlling, it is very hard
>to know what is inside it, because a lot of variation in its attributes
>have very little effect on anything detectable from the outside.  That's
>what is meant by control.

But there is more than enough around to let you test models of what's inside;
that's what is meant by modelling.

There is much more in your post and I'm sure I haven't addressed all your
questions but let me get back to the original questions, as you do:

>My answer to my own pair of questions is, at present,
>(1) I would not do a psychophysical experiment differently, now that I know
>about PCT

My answer would be -- I just wouldn't do it; it's not necessary. The goal of
psychophysics (determining f) in solved by building a model that actually
controls
the variable that ther subject does. And, more importantly, PCT research
solves
the problem that psychophysics takes for granted , viz. WHAT IS P? That
question
is answered by determining the kinds of variables that people control.

There are several kinds of psychophysical research that could become PCT
research;
"Cross modality matching" comes to mind -- see if a subject can control the
relationship
between tone intensity variations and light intensity variations -- this could
 be done
as a pursuit or compensatory task. If control was not good (which it might not
be -- I think the Stevens correlations are only about .96) then figure out way
 to
improve the control. Then try to build a model that does this; it may very
well be that
the model will require some non-linearity in the perceptual function in order
to
work (and mimic subject behavior). Also, "bisection" experiments could be
PCTized;
all that has to be done is figure out ways to get the correlations (between
disturbance
and output) up around .99 -- then you can start building sensible models).

>(2) I would use the results of classical psychophysical experiments as
>guides to the possibilities of control, by determining the limits of
>resolution for different perceptions.

I think you would be enormously disappointed in the results (at least compared
 to
what you could do by just building models of actual control systems). But feel
free to try -- Bill and Tom and I have been building pretty accurate models of
real control behavior -- a hell of a lot more accurate than the results of any
 classical
(or modern) psychophysical study of the variable being controlled -- so I just
 don't
know that using classical psychophysics would improve things much (except to
make
the classical psychophysicists feel better because we're using their stuff).

Psychophysics, I'm afraid, is just another one of those conventional areas of
psychology that bit the dust in 1973 (the year that "Behavior: The Control of
Perception" was published). Hey, I loved that crap too; I also loved WWII,
JFK,
bell  bottoms and lava lamps -- you get over it.

Best

Rick