> From: "Parker, Chris" <Chris.Parker@soton.ac.uk>
> Date: Tue, 14 Nov 1995 16:43:21 +0000
>
cp> This escape to ideal experimental organisms producing orderly
cp> predictions for techniques which produce disorderly factual real data,
cp> is best summed up by his reference to words in a song: "a paper doll to
cp> call my own" rather than a "fickle minded real live girl". [When he
cp> says we don't need mathematical models but we do need mathematical
cp> analysis of behaviour, he is talking about observable changes in
cp> performance but not, I guess, statistics!]
Skinner rejects models because he thinks the data will explain
themselves. In no other field has this proved true, and psychology
(behavioural analysis) is no exception.
cp> He suggests that the major loss of experimenters is to those areas
cp> concerned with what is going on inside the organism. "observable
cp> subject matter is abandoned in favour of inferred (traits, perceptions,
cp> experiences, habits, ideas, and so on)". Mentalist concepts should be
cp> abandoned in favour of explanations involving causes from outside the
cp> organism.
Skinner conflates internal mental theory with any internal theory at
all. There can be nonmentalist internal theory, and paradoxically, once
it explains all of behaviour, it may turn out to be the explanation of
the mental too (or as much of an explanation as we can ever have).
All of science is inference: Inference about the causal basis of the
data.
cp> [He thinks that the reinforcer will be the draw of the true
cp> science and analysis of behaviour, but I think the mathematics of
cp> complex contingencies will be a big turnoff and anyway, why waste
cp> introspection, it gets me out of a lot of trouble despite its
cp> location.]
Introspection doesn't get you out of scientific trouble, it gets you
into it. But reinforcement doesn't get you out of it either. Only a
causal theory of behavioural capacity will.
cp> He seems to be concentrating on theories grounded in areas that are
cp> incompatible in many important ways with the area of behaviour.
cp> Statements about the nervous system are not expressed in the same terms
cp> [and therefore units?] as behaviours like secreting saliva. Mental
cp> events like being pleased are in different realms to bingeing
cp> behaviour. [I'm not sure this is still the case with modern
cp> neurophysiological work, and is it not possible to conceive that
cp> certain neural activity is linked to both experience and activity a la
cp> Jeannerod?]
Neural activity is surely the basis of both behavioural capacity and
mental states. To explain the latter, it must first explain the former.
But what is needed is not neural DATA, any more than we need behavioural
data. What is needed is a causal theory of how neural activity (or ANY
activity) generates, in the first instance, behavioural capacity, and
eventually, one hopes, the mental states that ride on it.
cp> Learning is not the same as performance, like getting out of a box
cp> faster, because learning may take place but other factors may result in
cp> no effective change in performance. Latency is also dismissed because
cp> it depends on what the subject is doing at the moment the stimulus is
cp> presented. Rate of response is more promising but is not a measure of
cp> probability. [It would not have been a good predictor of the Hungerford
cp> massacre.] Skinner suggests that we need smaller units of behaviour
cp> that lead to and compose the larger more complex events that we wish to
cp> predict.
Skinner speaks of learning as if it were an internalist inference of the
kind he would eschew. But of course it is just an empty term, hence
harmless for him. It would only become something substantial if, instead
of "learning," it were the mechanism of learning capacity: But for that,
working hypotheses are needed, and those are not forthcoming from
behaviorism.
The wrangle about rate vs probability is a not very sophisticated
version of a philosophical issue, involving frequentists, Bayesians,
and subjective probabilists. Don't bother pursuing it: it's a red
herring. Data won't explain themselves, irrespective of whether they are
expressed in rate or probability terms...
cp> I couldn't see what he was trying to say here. He starts by suggesting
cp> that the Law of Effect is no theory but a procedure as in operant
cp> conditioning of a pigeon, then goes on to say that when you ask why
cp> reinforcement has this effect then theories arise. Then follows a
cp> discussion of extinction curves in which he concludes that curves are
cp> curved possibly because of competition between the normal response and
cp> "frustration" at lack of reinforcement, and novelty (variables
cp> changing) appearing in the extinction situation. His message is look at
cp> more data. [Couldn't learning/conditioning simply occur because either
cp> the brain is hard wired to support the conditioning process or it is
cp> capable of programming itself to do so to sustain certain states such
cp> as a level of blood sugar?]
The two kinds of conditioning: Pavlovian stimulus/stimulus association
and Skinnerian emitted response shaping by reinforcement are trivial. Any
engineer could build a Pavlovian toy or a Skinnerian toy. The trick is
to scale it up, so it can do what WE can do: For this task, neither form
of conditioning is any sort of a guide. What's needed is powerful
hypotheses about what could be the causal mechanism.
cp> While theories are fun, he suggests that we turn to obtaining data that
cp> shows orderly changes in the learning process itself.
cp> [He seems to be saying that there are just too many variables to
cp> have successful theories, we need to concentrate on narrow
cp> areas.]
Remember St. Exupery's businessman in Le Petit Prince, who collected
stars, just for the sake of having them? That's how it is with data
collection. No matter how much data you gather, you have nothing unless
you have a causal theory too.
cp> This might be summed by two of his statements here. First by his
cp> restatement of what he considers to be a theory: "any explanation of an
cp> observed fact which appeals to events taking place somewhere else, at
cp> some other level of observation, described in different terms, and
cp> measured, if at all, in different dimensions. Second by his suggestion
cp> that "To guess who is calling when the phone rings seems somehow more
cp> admirable that to pick up the phone and find out, although it is no
cp> more valuable". [I think he may be wrong when the door bell rings late
cp> at night and you are all alone!] Again he restates the case for finding
cp> relationships between behaviour and variables as a basis for the
cp> science and technology of behaviour. [Dangerous territory?]
A complete misunderstanding of the ways of scientific inquiry: And the
phoning metaphor is way off. Data don't explain themselves; they are
explained by theories. And theories always go ahead of data. It is only
when the data finally catch up that your hypothesis about who was on the
phone is finally confirmed.
cp> [It seems to me that Skinner has a case for a technology of behaviour
cp> because human beings can change their environment and hence the stimuli
cp> and reinforcements that shape their behaviour.
cp> Marketeers and criminals already use it in selling or conning.]
Behaviour can be shaped by rewards and punishments. We all know that.
Anyone who controls the reinforcement contingencies, controls the
organism. In fact, you can get similar effects with straitjackets, or
machines that force you to go through the motions. But control is not
explanation...
This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:56 GMT