Re: Conditioned Taste Aversion (was metablather)


"Michael Olea" <oleaj@xxxxxxxxxxxxx> wrote in message
news:0teEf.24424$F_3.20724@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Glen M. Sizemore wrote:


"Michael Olea" <oleaj@xxxxxxxxxxxxx> wrote in message
news:PzoDf.37115$PL5.753@xxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Glen M. Sizemore wrote:



"JAK" <jak@xxxxxxxxxxxxxxxx> wrote in message
news:1138474442.180086.155810@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Michael, I'll await your response.

Glen, in deference to the task before Michael, I would still like you
to isolate a specific, fundamental experiment as a target for
discussion. Michael's point is well taken. It's tough to "boil the
ocean" (or even a lake). Please declare something a bit more
manageable.

Thanks - JAK


Well, I did talk about a class of experiments. In general, the
experiments
I am talking about are concerned with the fundamental nature of
conditioning; is it a temporal phenomenon, or does it require the
notion
of some long-term correlations between stimuli or between responses and
consequences. These experiments must be considered together - that is
where it becomes confusing. But, anyway, it seems to me that Michael's
task is not that great. He made some assertions about learning and
probability distributions etc. I would like a basic, simple statement
from
him concerning how these issues relate to classical and operant
conditioning.



It seems to me that the old "temporal contiguity notion" is holding its
own. I lean in that direction for a couple of reasons. A big reason is
that "unsignalled, delayed reinforcement, produces lower rates of
response
than immediate reinforcement. In order for correlation to play any
role,
the relevant time scale would have to be very, very short. But the
noise
at such time scales is so great that one wonders how differences
(between immediate and delayed reinforcement) could be detectable.
Plus,
there's the whole issue of the "molar" view starting out talking about
long-term correlations, then being forced to squeeze the time scale
down
to one that
approaches that at which the "contiguity" people operate. There are
other reasons, but the issue is sort of complex. That complexity, and
the large experimental literature that exists in a filed that Michael
has denigrated, is part of the source of my challenge to him.

Garcia, J. and R. A. Koelling. 1966. Relation of cue to consequence in
avoidance learning. Psychon Sci 4:123-4.



http://www.uwf.edu/psych/bmikulas/Webpage/learning/section%20two.htm#relationofcue
(fixed broken link; thanks, Joe)

I'm quite familiar with this paper by Rescorla.

Actually I was pointing to the paper by Garcia and Koelling, which reports
experiments on what has subsequently sometimes been called "conditioned
taste aversion".


Hmmmm. I'm a bit surprised given the topic that I suggested, which came from
your reference to "learning probability distributions."




I'm sure you are familiar with these, but for the rest of
CAP, these are experiments in which rats were made sick to their stomachs
when they drank water, either because the water was toxic, or because the
rats were subjected to ionizing radiation. In the Garcia and Koelling
study
the water was either "bright and noisy" or "tasty" (sweet or salty), or
both. Two results are interesting: 1) the conditioned rats avoided tasty
water, but not bright and noisy water - you might say, if you were in an
interpretive mood, that they formed associations between nausea and
gustatory cues, but not between nausea and audiovisual cues; 2) taste
aversion occurs even when there is a long delay (as in hours) between cue
(taste) and consequence (nausea). Moreover, when drinking "bright and
noisy" water, or drinking "tasty" water was paired with mild electrical
shock the rats subsequently avoided "bright and noisy" water but not
"tasty" water. Waxing interpretive once again, you might say the rats
associated audiovisual cues with shock, but did not associate gustatory
cues with shock. Get sick to your stomach, must have been the last thing
you ate, even if that was hours ago; get shocked, probably has nothing to
do with gustatory stimuli but rather with some audio or visual or tactile
event in close temporal contiguity with the shock event.

All of this would seem to make sense from an evolutionary biological point
of view (and, as I might argue later, time permiting, from a statstical
learning theory point of view) but it does seem to pose, at least at first
glance, some problems for the notion that classical and operant
conditioning alone are enough to account for all forms of learning.

But it is far from clear in what sense that the process is not classical or
operant conditioning - or maybe both. The key definitional issues here are:
1.) whether or not there is a contingency (in the sense of dependency)
between 2 stimuli, 2.) whether or not the functional properties of a
"previously neutral stimulus" change, and 3.) whether or not the change in
function is due to the dependency (which raises the issues that Rescorla
discusses). On the other hand, it is possible that the main process is
positive punishment. The three questions relevant would be: 1.) Is there a
contingency between licking a drinking tube and the administration of
x-irradiation? 2.) Is there a reduction in the probability of drinking? 3.)
Is the reduction due to the response-event contingency? Actually, it is
likely that the observations described in the Garcia paper involve both
classical and operant conditioning. The fluid becomes a conditioned punisher
via the contingent relationship between it and the x-rays, and its
ingestion, being contingent on licking the drinking tube, punishes licking
the tube. The long delays possible in CTA training is not part of the
definitional scheme, nor is the specificity of the modalities issue.



In general, the G&K paper has been used, as have a few others, in an attempt
to discredit a certain segment of the scientific community and trivialize
the expansive experimental literature it has produced. In any event, in
order for CTA to be a new process, one would have to show that it does not
meet the definitional specifications, or that it does not function in other
ways like the defined processes. The unusual temporal parameters involved
could, perhaps, be invoked here, but an alternative view is that the
experiment extends the generality and usefulness of the definitions and
their subsidiary processes.



If
nothing else, the results suggest associative hypothesis space is divided
into domains (i.e. product spaces composed of at least two disjoint
cue/consequence domains).

More papers on CTA are available here:

http://www.magnet.neuro.fsu.edu/Papers/classicCTA.html

I have no problem with the
truly random control procedure (TRCP) but it doesn't carry the
theoretical
significance that you think it does. ...

This, and the rest of what you wrote is interesting stuff, Glen, but right
now Hullian drive reduction has me thinking about tacos. Maybe I can pick
up the thread later tonight. For one thing, the RW delta rule can be
derived, given certain assumptions, from Bayesian decision theory.

Maybe so. But the R&W equation was simply meant to show that explanations in
terms of temporal contiguity are almost always possible when contingency (as
dependency) is invoked (with its attendant correlations and probability
distributions etc.) it may be that your Beyesian treatment offers new
things - that is why I asked you to apply it to experiments that surround
the most fundamental issues involving both classical and operant
conditioning.


More
interesting, though, is to find conditions under which BTD and RW lead to
different predictions. Anyway, I'm starving.

Or to describe predictions that don't follow from the major positions
gnerated by the EAB's investigation of fundamental issues. That was my
point.





Later.

-- Michael



.

Relevant Pages

  • Re: Conditioned Taste Aversion (was metablather)
    ... conditioning; is it a temporal phenomenon, ... these are experiments in which rats were made sick to their stomachs ... when they drank water, either because the water was toxic, or because the ... gustatory cues, but not between nausea and audiovisual cues; ...
    (comp.ai.philosophy)
  • Conditioned Taste Aversion (was metablather)
    ... Michael, I'll await your response. ... >>> conditioning; is it a temporal phenomenon, or does it require the notion ... when they drank water, either because the water was toxic, or because the ... gustatory cues, but not between nausea and audiovisual cues; ...
    (comp.ai.philosophy)
  • Re: 50YO neuroscientist learns to do stereopsis for the first time
    ... the process in question appears to be operant ... We cannot account for operant ... conditioning without accounting for physiology. ... Studying physiology, at its own level, would not ever ...
    (comp.ai.philosophy)
  • Re: Conditioned Taste Aversion (was metablather)
    ... possible response rates. ... so admitting CTA as Pavlovian conditioning (a ... positive punishment) does not extend the definition. ... that one can identify physiology relevant to operant conditioning is to ...
    (comp.ai.philosophy)
  • Re: 50YO neuroscientist learns to do stereopsis for the first time
    ... vision, physiologically, without accounting for operant conditioning. ... We cannot account for operant ... Studying physiology, at its own level, would not ever ...
    (comp.ai.philosophy)