Behaviourist, dispositional, functional

Roger Bishop Jones

JL and I had an exchange about the relationship between "behaviourist" and "dispositional" (I wanted to know the difference).
I was Googling today and found an account by Prior of the difference between behaviourism and functionalism which casts some light on this.
I've lost the Prior site, but no matter, there is a more complete account under "functionalism" by Janet Levin at SEP.

I propose here a little discussion of how Turing, Grice, Ryle and Carnap may be placed relative to the varieties of behaviourism and functionalism described by Levin and each other (dispositional appears in the definition of behaviourism).

At SEP under the heading "Antecedents of Functionalism" we find a section on the Turing test and a section on behaviourism.  I discuss the latter first.

Two varieties of behaviourism, empirical and logical or analytic, are distinguished, and are considered precursors to corresponding variants of functionalism (which seek to remedy defects in the behaviourist theories).
Both of these behaviourisms seek to explain something in terms of behavioural dispositions (so, both the behaviourisms and the subsequent functionalisms are dispositional).  The empirical behaviourist explains behaviour in that way, the logical behaviourist explains the meanings of mental concepts in that way (Malcom, Ryle and perhaps the later Wittgenstein are mentioned).

The principal difference between behaviourism and functionalism is that the former deny the reality of mental states and seek explanations of behaviour or language which are independent of mental states or concepts, whereas the latter insist on mental states and offer explanations of mental states or concepts which may involve other mental states or concepts.  It sounds as if there is a metaphysical aspect here, which relates to Ryle's desire to refute Cartesian dualism, but there are more concrete differences as well.  The functionalist allows himself to talk about mental states when giving an account of mental concepts, he does not have to give an account exclusively in terms of behaviours.

Functionalism comes in three varieties according to its origins.  Different varieties come from early AI research (and therefore connect with Turing), and from the empirical (or "psycho-") and logical behaviourisms (the latter connecting with Ryle, the former being an empirical theory rather than a philosophical one or a philosophical theory about empirical scientific theories of mind).

The description of "machine state functionalism" given by Levin is I hope, not quite correct (it would otherwise reflect badly on Putnam), because she seems not to understand the difference between a finite state automaton and a turing machine (the difference is the tape, which is a big difference).  If we recast her definition appropriately then Putnam's account of this kind of functionalism asserts that any creature with a mind "can be regarded as" (this seems very weak) a Turing machine and that its mental states are then to be identified with the states of the Turing machine (and that would have to be the state of the automaton and the content and position its tape, though Levin talks only of the state of the automaton).

I quote Levin for the nub of her description of the other two kinds of functionalism.

"What is distinctive about psycho-functionalism is its claim that mental states and processes are just those entities, with just those properties, postulated by the best scientific explanation of human behavior."
 
"the goal of analytic functionalism is to provide “topic-neutral” translations, or analyses, of our ordinary mental state terms or concepts"

So how do our four protagonists relate to these various kinds of behaviourism and functionalism?

This is my present impression.

Turing so far as I can see from "Computing Machinery and Intelligence" does not fit well into any of these schemes.
First of all, he is very clear that his test does not test for thinking, and he explicitly disavows any opinion about whether machines can think.  I think the manner of his dismissal might be thought to implicate a similar attitude to other mental concepts.

Here is a salient quote:

 "It will simplify matters for the reader if I explain first my own beliefs in the  matter. Consider first the more accurate form of the question. I believe that in about fifty years' time it will be possible, to programme computers, with a storage capacity of about 10⁹, to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning. The original question, "Can machines think?" I believe to be too meaningless to deserve discussion."

He does not appear to be interested in analysis of the meaning of mental concepts, and therefore is neither an analytic behaviourist nor an analytic functionalist.
Furthermore, the above quote I believe gives the central purpose of his paper, which is to make a claim about what computers will be able to achieve at some point in the future.  He not only lacks interest in the meaning of mental concepts (beyond what is unavoidable is describing his thesis), but also does not appear to be interested in scientific theories about the mind.  This I think excludes him from being considered an empirical behaviourist or functionalist, at least as far as we can see from this paper (the paper does not exclude this,
though we might think his assertion that the "Can Machines Think?" is meaningless does exclude him from being an analytic behaviourist or functionalist).

Turing is closest to a "Machine state functionalist", not surprisingly since they were probably influenced by his work.  But does he fit the description?
Does he think anything with a mind "can be regarded as" a Turing machine?
Possibly.
But this is not, on my reading, the thesis which his paper presents.
I believe he is asserting functional identity between minds and Turing machines (to each mind there is a corresponding Turing machine with which it is functionally identical), though he does not say that explicitly.
But "can be regarded as" I still think, though vague, a little too strong. 

Let us now consider Grice, which I am ill qualified to do, but perhaps these remarks will provoke a better informed response from Speranza.
Grice seems to me certainly to be an analyst and to be interested in the meanings of mental concepts.
I doubt that he is any kind of behaviourist, but it seems to me that he might well be an analytic functionalist.
In relation to Turing, my impression is that there is little contact, neither agreement nor disagreement.
Insofar as Grice is interested in automata, I would imagine this to be concerned with the insights which might be gleaned from them into the nature of language or other philosophical problems.
Would he have any interest in predictions about what computers will be able to do at some point in the future?

Ryle I should perhaps have mentioned before Grice, because he preceded Grice and behaviourism preceded functionalism we may think of Ryle (as he is cited by Levin) as being some kind of behaviourist.
Because of his anti-dualistic zeal and his belief that some of our language (if perhaps only at the metalinguistic level) is tainted by dualism, it is hard to see his position as being purely analytic.  He attributes category errors to certain kinds of ordinary discourse about mental concepts.
It seems to me that Ryle is engaged (in his own terms) in both descriptive and revisionary metaphysics, in analysing the metaphysical content and pathologies of language as it is, and in undertaking a profylactic metaphysical synthesis. His descriptive metaphysics reveals pathologies (category errors particularly) which his revisionary metaphysics seeks to repair.
What makes him seem a behaviourist is that his metaphysics is monistic, materialistic.  Perhaps this is enough to call him an analytic behaviourist, but I am inclined to think that to crude a characterisation.

Between Ryle and Turing I see no connection, their ideas seem logically independent, neither supporting nor confuting each other, though possibly mutually sympathetic in a weaker sense.

What of Ryle and Grice?
Of our four protagonists here, these two are probably the closest to each other.
Insofar as Ryle is engaged in metaphysics, this would be of interest to Grice, though perhaps only the descriptive rather than the revisionary metaphysics.
Grice's ontological pragmatism would I imagine leave him poorly motivated to enter into a crusade against Cartesian dualism.
Functionalism is however very accommodating, it does not seem to exclude much, and so Grice might well be an analytic functionalist, the distinction between him and Ryle similar to that between an analytic functionalist and an analytic behaviourist.

I now come to Carnap, the scientifically oriented scourge of metaphysics.
The special features which Carnap brought in his anti-metaphysical fervour are of interest here.
Before Carnap we would expect a positivist to be also a phenomenalist, and therefore to have no truck with Cartesian dualism and to be close to analytical and psychological behaviourism.
Carnap's anti-metaphysics is novel (for a positivist) in being ontologically liberal and pragmatic rather than nominalistic.  This is reflected in his linguistic pluralism, and so, though he worked primarily with ontologies in which there were no mental entities (phenomenal, physical, and theoretical languages), he would not have ruled them out on principle (that would violate his principle of tolerance).  In consequence Carnap could not be a doctrinaire behaviourist, but would admit a behaviouristic scientific theory if it could by empirically confirmed.

He was concerned with scientific method, and I think it plausible that he would have been some kind of empirical or psycho- functionalist.
Carnap's conception of philosophy was as a kind of analysis, but like Turing, he lacked interest in the details of natural languages, for much the same reasons.  He advocated the use of formal notations and methods in philosophy and science, and explained the relationship between formalised concepts and those of natural languages as "explication", a relationship too weak perhaps to allow that the formal theories provided analyses of the natural concepts.

Because of his desire to formalise and his scientific orientation it seems to me that Carnap is closer to Turing than to Ryle or Grice, and I see no conflict between the views of Carnap and Turing.
Carnap's relationship to Ryle and Grice is more difficult.
As a formalist he lacked interest in the minutiae of natural languages, as a positivist one would expect him to be antagonistic to the metaphysical  aspects of Ryle's thesis (nominalism is metaphysics for Carnap).

What is the point of this long ramble?

Well one thing which I think important is a point about varieties of analysis.
Analytic philosophers seem often exclusively concerned with the analysis of ordinary language, and may too readily assume that what other philosophers (or even non-philosophical analysts) are doing is the analysis of language.

Turing and Carnap in different ways offer examples of work which may be misconstrued as concerned with the analysis of ordinary language.
Turing may be misconstrued as giving a test for or an analysis of the meaning of the concepts of "thought" or of "intelligence" when his purpose was simply to argue that humans had no intellectual capacities which could not be realised in a machines.

Carnap can be construed as primarily concerned with the explication of terms in ordinary language, and it is the central thrust of Carus' book on Carnap and 20th Century thought that the notion of explication is the centrepiece of Carnap's philosophy.  However, this flies in the face of what Carnap writes about his motivations and objectives.  His principal aim was to apply the new methods he learned first from Frege to the advancement of science, and to progress Russell's conception of a scientific philosophy using formal methods.
In my view "explication" is merely Carnap's way of connecting his methods with ordinary language for the benefit of philosophers who would not otherwise understand it.

For many philosophers it seems that the only purpose of formal languages is to provide models of aspects of the semantics and logic of natural languages.
But just as one speaks a second language fluently only when thinking in that language rather than translating into it, proficiency in the application of formal languages results in their native use.  A formal language is the best kind of language for a variety of tasks involving the analysis of some non-linguistic subject matter, and the role of natural languages in this process is ancillary.  One begin's with the formal model, and perhaps adds some informal annotation to help the reader see the point.  The analysis yields
no information about language.

This kind of analysis, of problems or of scientific domains, is more important to Carnap than the analysis of language, and should not be confused with it.
Turing is one step further from the analysis of language, for his central purpose is not analytic at all, it is technological, he is talking about what kinds of machine will in the future be constructed.

Roger Jones