Sorensen, R. (1992). Thought experiments. New York/Oxford: Oxford University Press
Never in my (admittedly, not particularly lengthy) life have I bumped into such a remarkable mixture of immodesty and disorder. In Thought experiments, nothing seems to escape prof. Sorensen’s wit. Read “not one single thing” for “nothing”. The book is a disparate plunge into microbiology, logic, psychology, epistemology, pathology, mathematics, biology, grammar and, par Socrate!, metaphysics. To be clear: the nature of the subject matter must have made delving into the above list almost a necessity. As anyone who gave it a bit of attention will concede, thought experiments appear everywhere. But there is a big difference between discussing a type of discourse throughout more than one field and discussing a type of field throughout more than one instances of a type of discourse.
I sense (from the book as well as from some of the reviews it received) that Thought experiments was intended as a monograph. Its explicit taxonomic objective is the first clue. If so, it is an unfortunate by-product of copyright laws that the curious student will have to trudge through 300 pages of the only book with such a telling title. Whoever edited this book is equally blameworthy for not spanking out tens and tens of pages the sole purpose of which is to squash doubts about the author’s literacy. Of course, “freedom of expression” etc. OK. But if not by prohibition, may students more trained in elementary law figure out ways of blocking capital works with capital titles being written after this fashion:
Rearranging the problem this way makes the problem easier, just as it is easier for a child to eat his steak after his mother has sliced it for him.
One of the advantages of watching a movie on a VCR rather than at a theatre is that you are free to fast-forward and rewind.
Predatory birds and snakes maneuver so that their prey can be swallowed headfirst. Consumers of information display similar preferences.
If you have more than one mouse, you have mice. But houses are not hice, nor blouses blice.
When stood up against Vaihinger, Kuhn appears pale and tame. Kuhn dips his big toe into the pool of incoherence. Vaihinger belly flops.
At first glance, covert presumption precluders are as innocent as new-laid eggs.
Thus, thought experiments developed from experiments in the way a spider's spinners evolved from legs.
To close this exhaust-pipe introductory passage, I leave you with the author appearing on page 124 in a photo of superlative irrelevance to the discussion.
OK, from this point on I shall take Sorensen’s book seriously.
Two purposes could be identified as red threads guiding the book: (1) to establish true and interesting generalizations about thought experiments, (2) to offer a resolution to the problem of thought experiment’s epistemic value (rapidly, are thought experiments of any epistemic value). The first one includes (1a) a taxonomy and (1b) a reconstruction of how they work. The second one includes (2a) an explanation of the role of thought experiment in science and philosophy, (2b) an answer to the “new data” problem – “But if you just ponder, then the information you have leaving the armchair is the same as the information you had when you sat down. So how can you be better off?” (p. 76).
You may have noticed that there is no “(x) a definition” point in the above list. Sorensen does provide one but it is at page 205 and it comes after the taxonomy, after a full-blown discussion of most relevant thought experiments and after the logical reconstruction. This being the case, I shall refrain from accepting the sentences below as constituting a definition:
“A thought experiment is an experiment (see p. 186) that purports to achieve its aim without the benefit of execution.” (p. 205)
“An experiment is a procedure for answering or raising a question about the relationship between variables by varying one (or more) of them and tracking any response by the other or others” (p. 186)
Anyway, taclking (1), Sorensen posits the profitability of what he names the cleansing model of thought experiments. He opposes it to other models which, in the past, purported to have found how these scientific and philosophical machineries work. Swiftly put, the cleansing model generalizes that all thought experiments are intended to refute by reduction ad absurdum.
“The cleansing model is inspired by incidents in which you recognize your own irrationality and then change your beliefs to remove the flaw. A friend of mine was a sensitive boy who became worried about the ants that he inadvertently crushed while walking about. His father assured him that he should not worry about killing ants because "there are millions of them." This satisfied my friend for a while. But his new equinamity was terminated by a thought about the growing human population. This illustrates the familiar situation where an inconsistency takes root, is detected, and is then weeded out.
This, I believe, is a good starting point: instead of debating what is the added value of creating a story to refute a claim, Sorensen begins with the simpler and more salient feature of thought experiment. Whatever may prove to be the added value of using them, one’s inquiry should start there: they are first and foremost tools for refuting a claim.
What type of claim? And how does the tool work? To the first question, Sorensen answers: either a universal claim or a particular claim. According to this division, the thought experiments that refute a universal claim are called necessity refuters. The type of thought experiments that refute particular claims are called possibility refuters. Of course, the claims need not be neatly designed as categorical statements, but their consequence always receives the abovementioned modality.
I should probably mention here that possibility statements are not refuted in the same way necessity statements are. Since possibility statements do not prohibit the truth of any one proposition, their logical falsifiability is zero. This has nothing to do with the field in which such statements occur. Also, regardless of the logical apparatus of a thought experiment “It is possible that x” cannot be used in an argument to arrive at “It is not possible that x”.
To the second question, Sorensen lays down these steps (p. 136):
i. S Modal source statement. Fertile sources of modal propositions include semantic theses (definitions, synonymy claims, entailment theses), testability theses (unverifiability, unfalsifiability, indetectability), feasibility claims, law statements, disposition and intention attributions, validity verdicts, and clusters of these—theories.
ii. S → □ I This proposition draws the relevant modal implication from the source statement
iii. (I & C) □→ W This proposition is read as a subjunctive conditional: if I and C were the case, then W would be the case. This proposition claims that the antecedent, which is the conjunction of the implication and the imagined situation, has a weird consequence.
iv. ¬ ◊W Absurdity. This proposition explains the weirdness as an impossibility
v. ◊C Content possibility. This asserts that the content of a that the content of the thought experiment is a possibility
Again, I grant Sorensen that this is a good place to start. After all, this is nothing less than (a more detailed version) of modus tollens. However, there is a minor point: the order of a sentences. Indeed, they are inconsistent in whatever order one arranges them, and thus Sorensen’s idea that thought experiments boil down to paradoxes is not affected by my rearranging. But if modus tollens goes as follows:
p → q
¬q
Therefore, ¬p
Then I see no reason for going:
p
p → r
r & s → q
¬q
s
Therefore, ¬p
To resume: the theory p predicts that r but since we know that s could be the case, and that the conjunction of r & s yields an absurdity, we conclude to non-p. I will argue later that the 3 extra steps from Sorensen’s model are not necessary – although I see how sometimes they can be an accurate reconstruction of this or that thought experiment. For now, let us observe that in order for such an apparatus to be academically potent, the other party must (be known to accept) s. If the possibility of a certain state of affairs is not granted, no thought experiment can be constructed. For instance, to take one of Galileo’s thought experiments, if the other party does not agree that two things can be attached to form one thing (balls, stones, whatever), the thought experiment has no power to persuade. In fact, the argumentative process goes the other way around: in the opening stage, both parties agree on s (& p, & r, & whatever else is needed), and on this agreement the antagonist builds his case.
The second point to observe is that the model does not fit the definition. Remember experiments being defined (quite adequately) as fiddling with one variable to observe changes in the other? Now what exactly is one varying in the model above? Where is the observed dependency? Let’s imagine the following theory: All ravens are born in the southern hemisphere. Let us now imagine that we know how no white things can be born in the southern hemisphere. If our universal statement implies “All white ravens are born in the southern hemisphere”, we have hereby unearthed an inconsistency. Have we made any changes?
The impression I get here is that the label of “experiment” still fits these uses of discourse because the possibility of s is provisionally accepted by both parties. They take it as their starting point – regardless of whether it is a real possibility entailed by any other theory they agree upon. At the end, in the concluding stage, the protagonist can always go back to the opening stage and revise his acceptance: indeed, many Aristotelians did just that to counter Galileo’s refutation.
The third point to observe – a point which is also stressed by Martin Bunzl in his review of Sorensen’s book – is that the thought experiment and the use of thought experiment are two different objects of study. For instance, our Galilean thought experiment was used to offer (some) support to the idea that velocity is universal and (total) corroboration with the thesis that weight does not matter. But the passage from the latter to the former is incidental and certainly not deductive. Going “Aha! You are wrong” and going “Aha! I am right” are two different steps. Sorensen discusses both, alternatively without making any explicit distinction.
Let us use this model to analyse Searle’s Chinese Room. In want of any definition, I will conclude that Searle’s Chinese Room is a thought experiment simply on the grounds that it is mentioned by many theorists interested in thought experiments. I leave in suspension the question of whether this is a satisfactorily accurate statement or not.
What is the S?
Well, at first glance, it must be something around the claim that:
(S1) Computers can think.
Notice that the chapter in which Searle presents his Chinese room (in Reith lectures) is entitled Can computers think? So it might be justified to start from S1. However, S1 is very ambiguous and it would be unfair to Searle to leave it at that. For instance, he mentions explicitly that what we are concerned with are digital computers, that is to say, machines which manipulate symbols and nothing more. By “nothing more” Searle means something like: however much they interact with the physical world, and however much they simulate human interaction with the physical world, the machinery does not produce intentional states of mind. It cannot “have a semantics”, as he puts it. So we might want to change S1 to
(S2) All sufficiently advanced symbol-operators can have any mental state
Whether this is indeed what Searle’s adversaries had put forward in the past is not relevant. I am, for now, interested in cramming the Chinese Room into Sorensen’s scheme.
The next thing is to draw “the relevant modal implication”. This implication could be:
(I1) A sufficiently advanced computer can understand Chinese
Yet again, we would like to make the implication more exact. Based on Searle’s text, we should like to say:
(I2) A sufficiently advanced symbol-operator, given proper conditions, can have the mental state of understanding Chinese
We need not nitpick: understanding here means “knowing the meaning of the sentences”. So let’s simplify, now that we know what we’re talking about:
(I3) All supercomputers in superconditions can understand Chinese.
Thus,
(I4) If it is a supercomputer in superconditions, it can understand Chinese.
Now, the man locked in the Chinese room is not a very advanced computer. That is one of the points. But as the reductio goes, Searle asks us to imagine that that is the case, or that no significant difference occurs. He wants us to suppose that the cards are very sophisticated, and that the “program” they represent is the same as in the very advanced computer the theory is referring to. Indeed, computers are just that: “people” that manipulate “cards” really well and fast. Historically, anyway, this is in fact very precise.
But what would C be?
(C) Let us Searle is such supercomputer in superconditions.
Now, modus ponens:
(I & C) □→ W: All supercomputers in superconditions can understand Chinese & Searle is a supercomputer in superconditions, therefore Searle understands Chinese.
Now:
(¬ ◊W): Searle does not understand Chinese
Therefore?
This doesn’t make much sense. I mean, it does, but it is too stupid. Searle’s conclusion is that computers cannot think, this is what he explicitly claims to tackle, so the contradictory of (S), something like:
(¬S) No sufficiently advanced computer can have any mental state whatsoever.
Is Searle’s argument so blatantly invalid? Note that the reconstruction above is also the reconstruction made by Damper’s (2006) The logic of Searle’s Chinese argument. The step from (¬ ◊W) to ¬S is always invalid because I & C are less general then S. So, unless some sort of inductive step slips in the scene, all thought experiments (reconstructed by Sorensen’s scheme) which conclude to the opposite of S are invalid. Let me re-state it so that it is clear. By “supercomputer” I mean “very advanced digital symbol operator”. By “supercircumstances” I mean the Chinese Room scenario. (S) is what “strong Artificial Inteligence claimed. The rest has been discussed:
(S) All supercomputers in supercircumstances can have any mental state
(S → □I) If [All supercomputers in superconditions can have any mental state], then [All supercomputers in superconditions can understand Chinese].
(◊C) Let us suppose that Searle is a supercomputer in superconditions.
((I & C) → W) If all supercomputers in superconditions understand Chinese and Searle is a supercomputer in superconditions, then Searle does understand chinese.
¬◊W Searle cannot understand Chinese
Therefore
(¬S) No supercomputers in supercircumstance can have any mental state
This is invalid. And simplified. And, I believe, wrong as a reconstruction.
0 comentarii:
Post a Comment