[Martin Taylor 2016.04.29.10.25]
VH
29.04.16 0820 GMT
You would have to ask the speaker. It's obviously
embedded in a larger idea of what happens when a living
body acts, one of whose criteria is efficiency.
VH: Thanks Martin but don't you think it is
still important for any speaker to at least mention
how their ideas and data fits into a larger framework? I
think it is fairly rare in psychological science. I’m just making a simple point
that this is what appeals to me about PCT as this seems
to happen more often.
Speaking as myself, I have to agree with you. Continuing as Devil's
Advocate (which is an honourable and useful, if not essential, role)
I suggest that it depends on the context in which the talk is given.
The talk in question was apparently given at a conference in which I
presume the audience all had the same context in mind.
In my (PCT) world, this should allow us to constrain
the parameters of the Artificial Cerebellum for the
perceptual control of the relation between foot position
and ground surface location.
VH: I don't know enough about that model but thanks for
the link the other day.
What the AC does according to Bill, who demos it in the paper, is
compensate for the dynamics of the control loop. It is an adaptive
way of producing a Finite Impulse Response (FIR) filter tuned to
compensate for the spectrum of the signal at its input. The result
of this is to produce output that is stronger at some frequencies
and weaker at others. In the time domain, it is a form of
prediction.
As I see the stepping experiment, the inverted pendulum dynamics are
(claimed to be) sufficient to determine where the foot now in the
air will land, and only after it lands can the output affect where
the following step will land. If that is so, then something like the
AC should produce output based on perceptual data for the region
near where the next step should be, in order to provide appropriate
reference values for the following step.
When I talk about the AC, I usually point out that its operation is
based only on the statistics of its input, which means that it also
tunes itself for any regularities (predictabilities) in the
disturbance – so I follow that tradition here, for no reason other
than that it is traditional.
MT: I would guess that in his mind EP is such a "grand
structure". Whether or not
we agree that it should be, if he thinks it is, I guess
there would be where to look for the answer to your
first question.
VH: For me there is already too much research in
psychological science that seem to be an ideas in search
of phenomena.
Continuing as Devil's Advocate, I think we should consider the
feedback loops around observations (inputs to perceptual functions),
theories (perceptual functions) and outputs (experiments that should
influence the data). “Too much” in any of these areas isn’t very
helpful. A sledgehammer will crack nuts quite well, but using one is
a waste of energy. You are suggesting that there is a sledgehammer
of too many ideas. But isn’t that a sign of reorganization going on,
like producing lots of evanescent perceptual functions, most of
which are not usefully controlled and vanish, but some of which do
prove useful and become solidified? Ideas may be low-level
perceptions that can be developed into theories and world views if
they serve to control (explain, describe correctly) the observations
to which they refer, and not if they don’t. Likewise, if there is a
theory or world view, it requires lower-level ideas working on data
to allow it to control the larger structures of data we call
“phenomena”.
I guess it will not have escaped your notice that I am using a PCT
world view in the above. I don’t think this is inconsistent with a
Devil’s Advocate position, because it allows me to see how even this
world view could be reorganized into something different if it
failed to account for something that it “ought to” be able to
account for. And one thing it ought to be able to account for is the
results of the experiments in the talk. Speaking not as Devil’s
Advocate, I think it does. Apparently, the low-level physical
dynamics do, independent of world view (other than that at this
scale of size and speed, Newtonian mechanics works well). PCT can
use such facts just as well (I should hope) as can any world view
that apparently competes with PCT. The AC isn’t the only way PCT can
account for it, but it is what immediately sprung to mind when I
watched the video.
Martin
PS. I'm trying to avoid getting too involved with CSGnet for the
next while, as I have been for the last while, so this message
should be considered an outlier. I hope to make any further
interventions much shorter.
···
mmt-csg@mmtaylor.net
On 2016/04/28 2:01 PM,
Richard Marken wrote:
[From Rick Marken
(2016.04.28.1100)]
Martin Taylor
(2016.04.28.11.04)–
MT...But looking at the blog discussion
between Wilson for EP and Warren and
Rick for PCT, it now seems to me that we
have a situation described to me in
graduate school with the aphorism: “When
two schools of thought contend
vigorously, they are probably both
right”.
RM: Or not.
MT: To which I add
“except probably in those things about
which they say the other is wrong”
because where the other is wrong is
likely to be in an unexamined assumption
of some kind.
RM: Here’s
what Wilson thinks is wrong with
PCT: “…it has no
theory of information or how that
information comes to be made or relate
to the dynamics of the world. It’s an
unconstrained model fitting exercise
,and it’s central ideas don’t serve as
the kind of guide to discovery as a
good theory should central
ideas simply don’t serve as the kind of
guide to discovery as a good theory
should.” So I presume that you agree
that PCT is “probably wrong” about this.
If so, I’d sure appreciate knowing why
you think so. If nothing else, I
appreciate knowing what the heck Wilson
is talking about. Sounds (and feels)
like hand-waving to me.
When
someone says he has a mechanistic model that fits
the data better than another mechanistic model
fits the same data, it doesn’t sound like
handwaving to me. He has offered his code.
Handwavers don’t usually do that. What little I
read in the two blog posts does not suggest any
more hand waving than is the case in PCT
discussions once we get out of the realm of
tracking and smoothly changing variables.
I
wonder where you got the quote attributed to me
that “PCT is ‘probably wrong’ about” something
defined only as “this”. I don’t remember writing
anything that could easily be construed that way,
no matter what “this” is supposed to refer to. I
wouldn’t, because I don’t think PCT is
fundamentally wrong about anything that it claims.
However, PCT practitioners can be wrong when they
get more specific. We can be sure of that, because
if it were not the case, we would not have long
discussions on CSGnet.
I disagree with Wilson that PCT doesn't serve as a
kind of guide to discovery, but I think the
approach to PCT you have generally espoused does
cut off lots of “guides to discovery”, and if
Wilson sees that as “PCT”, he has a point, as I
have argued many times on CSGnet. (“The ONLY real
question for PCT is the discovery of THE
controlled variable” is the way I would describe
your approach). That approach, as we have often
seen on CSGnet, simply discards as irrelevant the
way PCT guides us into social and linguistic
frames, such as when a few months ago you brushed
off my description of the basic structure of
protocols as being a natural consequence of
perceptual control.
Furthermore, you have long argued, as he does,
that PCT “has no theory of information … [or of]
the dynamics of the world”. So I don’t think you
can legitimately turn around and disagree with him
there (though I would on both counts).
As I said, I don't know anything more about modern
ecological psychology than what I read in the two
blog posts and saw in the video about walking
through obstacles. So I can’t judge how handwavy
it is, or how well its models compare with PCT
models. I wouldn’t be too surprised, if models
from both theories describe the same data, to find
that when you deconstruct them they are the same
model. But neither would I be too surprised to
find that they aren’t, in which case your would
have a test for which theory fits human observable
behaviour better.
Martin