Mortensen and O'Brien insist that tokening histories are relevant to explanation, yet they admit that such histories are ambiguous or irrelevant to current dispositional probabilities. They may have a reason to suppose tokening histories are relevant to representational content (hence, to computational explanations). Their reason, however, is not present in their response.
2. Mortensen & O'Brien, to my mind, continue to definitionalize in the face of my paper, claiming that it shows only that Fodor's theory is trivially false. They want to add sophistication to Fodor's view via modifications to the definition of representation. Although they do not consistently offer explicit modifications, when they actually offer explicit modifications, the only support Mortensen & O'Brien offer for adopting these modifications seems to be that (1) the modifications rule out certain counterexamples to the definition, and (2) the modifications are "not so implausible" (paragraph 8). In my reply I continue my tactic of looking at Mortensen & O'Brien's suggestions to see how well these apply to theories in cognitive science.
3. In dealing with the rod, cone, and simple cell examples in my paper, Mortensen & O'Brien recommend the following modifications to Fodor's position: First, as an overall strategy "to avoid the objections which turn on the linkage breaking in the external environment one simply rules them out: only breakings of causal chains within the skin are to be considered" (paragraph 7). Their rationale for this consists in the observation that "after all, causal chains outside the skin threaten asymmetry too easily" (paragraph 7).
4. I find this move unsatisfactory for two reasons: (1) It seems to amount to an arbitrary stipulation, unmotivated by actual theory and practice in cognitive science, but consistent with generating a definition that trivially withstands all ALLOWED counterexamples. (2) Mortensen & O'Brien later rely on the external causes of the system's tokening of a state to determine content (paragraph 10). As a result, their move seems to amount to the inconsistent claim that external factors are irrelevant to fixing content, while at the same time claiming that external factors are relevant.
5. As a second modification, Mortensen & O'Brien would alter Fodor's position so as to require that "the true link cannot be broken without breaking the false link, but there are ways of breaking the false link which do not break the true link" (paragraph 7). They seem to view the causal chains in my examples as forming an x, i.e., as having only one common link. As I understand Mortensen & O'Brien, they claim in the case of rods, for instance, that there are ways of breaking the heat to rod response connection without breaking the light to rod response connection. I cannot, however, understand how they can find this claim "not so implausible."
6. I point out in my paper that the causal chains in my examples share all but the initial link, i.e., that they diverge only in their original cause. In the case of rods, for example, body heat causes phantom photons by the same mechanisms operating when a photon of light causes a rod to respond. Body heat being a necessary condition for "normal" functioning can hardly be eliminated so as to break the heat to rod response connection without breaking the light to rod response connection. I take it, therefore, that there is no means of breaking the false link without breaking the true one, and hence that the above claim of Mortensen & O'Brien is implausible after all.
7. Another example may prove useful in making the preceding point. Present subjects with two groups of vertical lines. The groups have equal numbers of lines and the lines in each group have the same length. However, the lines in the group on the left are thicker, making the white spaces between lines smaller than for the group on the right. Subjects normally judge the left side of a picture to have a darker shade of black than the right side. In point of fact, the surface reflectance of the black areas remains uniform throughout the picture. The mistaken judgement is often referred to as "the assimilation effect." If Mortensen & O'Brien's modified definition held, one would expect cognitive scientists to explain the mistaken judgement of relative brightness by noting that there are cases where the mechanisms involved in the assimilation effect could be altered without altering the relative brightness to perceived brightness connection. But just the opposite takes place; the mistaken judgement is explained by showing how the mechanism involved plays an integral part in the normal functioning of the visual system.
8. The most common explanation attributes the mistaken judgement of relative brightness to the fact that the signals from several retinal cells are pooled together. That is, because the inputs from one's visual receptors introduce a certain amount of equivocation no matter what the conditions, one's visual system attempts to reduce this equivocation by summing the resultant activation from several contiguous receptor cells. One infers the surface reflectance of specific patches based upon the intensity of reflected light from the small areas of the patch. One's architecture implicitly controls for equivocation by instantiating the assumption that surface reflectance will be continuous over small areas of the visual field. One's normal inference strategy has the side effect that whenever surface reflectance varies dramatically within some small area, it will falsely raise or lower the value for the surface reflectance for that entire area. That is, when one closely intersperses the dark and light patterns throughout a visual field, the pooling effect serves to inhibit (dark) or to excite (light) cell firing. I emphasize that this pooling has the effect of reducing equivocation due to false signals from individual retinal cells. So, a case of normal functioning -- in fact, of functioning that acts to reduce error -- results in a misrepresentation of relative shading.
9. I turn then to the final suggestion by Mortensen & O'Brien, that the tokening history of the system (actual properties that caused past tokenings) should determine the state's representational content. Mortensen & O'Brien suggest that I claim that historical determinants of representational content are incompatible with Biederman's (1987) theory of object recognition, and that I misunderstand the role of representation in computational explanations in cognitive science.
10. Their claim here is incorrect. I note only that Biederman's theory makes no appeals to the historical causes of a system's states in explaining the system's object recognition abilities -- even in assigning representational contents to states of the system. I go on to point out that Biederman's neglect of tokening histories would be quite surprising if such histories were relevant to determining representational content. "If tokening histories are so important," I ask, "why does Biederman fail to take them into account?" More important, Mortensen & O'Brien do not address themselves to my other explicit criticism of tokening histories (and my answer to the first question), i.e., that historical asymmetries DO NOT support the relevant counterfactuals for using representation to explain cognitive task performance.
11. In short, historical tokenings do not play a direct role in the functioning of the system. So, systems with different histories, hence different representational properties according to Mortensen & O'Brien, can have the same dispositions to token a state, the state can have the same role in the system's operations, etc. As a result, different histories (i.e., different, inconsistent representational contents for Mortensen & O'Brien) do not necessarily impair the system's performance on the object recognition task. This last conclusion would seem to imply that representation, as Mortensen & O'Brien would define it, is irrelevant to explaining cognitive task performance. I take this to be a reductio of their claim.
12. Again, I mention an example to further illustrate my point: Consider the use of conditioning histories in prediction and explanation in psychological behaviorism. Behaviorists appealed to conditioning histories. However, their appeal to conditioning history was legitimate only because behaviorists had a linear function from past conditioning history to current dispositional probability. Past history was relevant to explanation only because one could infer current dispositional probabilities from conditioning history. However, behaviorists could find no function whatsoever from conditioning history to stimulus generalization. So, they could not use conditioning history to infer the system's current dispositional probabilities with regard to novel stimuli. Hence behaviorism could not explain stimulus generalization, as conditioning history did not support the relevant counterfactuals for explaining stimulus generalization.
13. Mortensen and O'Brien insist that tokening histories are relevant to explanation, yet they admit that such histories are ambiguous or irrelevant to current dispositional probabilities. That is, tokening histories do not support the relevant counterfactuals for explaining functioning, yet are necessary to explanations of functioning. Mortensen & O'Brien may have a reason to suppose tokening histories are relevant to representational content (hence, to computational explanations), despite the failure of such histories to support relevant counterfactuals for content-based computational explanations of functioning. They do not, however, present their reason in their response.
Biederman, I. 1987, Recognition by Components: A Theory of Image Understanding, Psychological Review 94, 115-147.
Mortensen, C. and O'Brien, G. (1993) Representation and Causal Asymmetry. PSYCOLOQUY 4(19) fodor-representation.5.
Wallis, C (1992) Asymmetric Dependence and Mental Representation. PSYCOLOQUY 3(70) fodor-representation.1.