Hugh Clapin (1998) Information is not Representation. Psycoloquy: 9(64) Representation Mediation (8)

Volume: 9 (next, prev) Issue: 64 (next, prev) Article: 8 (next prev first) Alternate versions: ASCII Summary
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 9(64): Information is not Representation

Commentary on Markman-Dietrich on Representation-Mediation

Hugh Clapin
School of Philosophy,
The University of Sydney
Sydney 2006


Markman & Dietrich's proposed minimal notion of representation -- information-carrying "mediating states" -- won't work because not all representations carry information about their contents.


compositionality, computation, connectionism, discrete states, dynamic Systems, explanation, information, meaning, mediating states, representation, rules, semantic Content symbols
1. Markman & Dietrich (1998) are right to address the concept of representation. After forty years of the cognitive revolution, more and more cognitive scientists are coming to the realisation that this core notion of cognitive science is poorly understood. The idea of an uncontroversial minimal notion of representation -- mediating state -- is attractive and timely. But sadly, I don't think it is up to the job.

2. Markman & Dietrich define a mediating state as information- carrying, and suggest that all cognitive scientists -- even sceptics about representation -- can agree that cognitive systems have and use such mediating states. Then they consider possible extra properties such mediating states might have to make them the kind of entity used in various kinds of cognitive explanation.

3. Markman & Dietrich say "Mediating states not only constitute the general class to which more specific kinds of representation belong; they capture the essence of representation" (para. 67). But this just isn't the case. Not all internal representational states carry information about their contents. Certainly some people -- e.g. Dretske (1981, 1988) and Fodor (1987, 1990) -- think information is the basis for a theory of representational content. But this is precisely one of the issues which make up the debate over representational content -- a debate which Markman & Dietrich explicitly claim to be avoiding (para. 7).

4. Not all explanatory appeals to representation in the cognitive sciences are appeals to mediating states as defined by Markman & Dietrich. They acknowledge that most AI systems don't use mediating states (para. 13), but don't recognise the consequence: that virtually no AI systems are representational, given that being a mediating state is essential to being a representation. That they need to brush aside forty years of AI research highlights the central weakness in Markman & Dietrich's analysis of representation: only some of what are commonly called "representations" in the cognitive sciences are plausibly mediating states.

5. Even those who suggest causal covariation as the core reduction of content don't think that EVERY representational state is information-carrying. Fodor's informational theory of content, for example, doesn't hold this. ("All that's required for "cow" to mean COW, according to the present account, is that some "cow" tokens should be caused by (more precisely, that they should carry information about) cows, and that noncow-caused "cow" tokens should depend asymmetrically on these" (Fodor 1990, p. 91). [Note that some typographical niceties have been lost in this quotation because of limitations of ascii publishing.]) For Fodor SOME tokens carry information about their contents, but not all. There are many reasons why not: here are a couple of important ones. First, if meaning were information, and all tokens carried information about their referents, then there would never be error: anytime "cow" was tokened in the brain, it would have to have been caused by a cow, since the token of "cow" by hypothesis carries information about (i.e., is caused by) cows. A second sort of worry is that the sentence "Whatever I'm looking at right now, it is not a cow" could never be true if every instance of "cow" was directly caused by a cow, or in some other way carried information about cows.

6. So Markman & Dietrich's notion of a mediating state is not one that can form the basis of all accounts of representation because it presupposes an implausibly extreme answer to the question of representational content.

7. The point of mediating states is that they are meant to provide a basic common ground between representationalists and representational sceptics. Markman & Dietrich suggest that even van Gelder (1995) can agree that the Watt governor has (information-bearing) mediating states. But in his paper "What Might Cognition Be, If Not Computation?", van Gelder (1995) says "no simple correlation between arm angle and engine speed can be the basis of the operation of the governor" (p. 353; see also van Gelder 1998). This may not be to deny that the arm angle carries any information AT ALL about engine speed, however I think it is a clear challenge to Markman & Dietrich's contention that the representational sceptics ought to agree that mediating states are implicated in all non-representational models of mind.

8. I'm not here defending van Gelder -- I don't know what to think about this particular case. Prima facie it seems uncontroversial that any system operating in an environment will have states which carry information about that environment, and it seems very likely that such systems will exploit that information in achieving their goals. This is the conciliatory point that Markman & Dietrich bring to our attention. However the fact that one of the anti-representationalists in their sights -- van Gelder -- explicitly disagrees with this minimal claim deserves some response from Markman & Dietrich.

9. I completely agree with Markman & Dietrich that it is a mistake to assume that "there will be only ONE set of [representational] properties that will suffice for all cognitive processes" (para. 67). I think this is a particularly positive feature of their target article. We should begin a careful taxonomy of which representational properties are required for which cognitive processes, without assuming that all representations share all the same properties. Maybe, however, it is also a mistake to assume that there is any representational property which is NECESSARY for all cognitive processes. Certainly Markman & Dietrich's attempt to find a minimal characterisation of representation isn't successful. Not all representational states carry information about what they represent, and so not all are mediating states.


Dretske, F. (1981). Knowledge and the Flow of Information. Cambridge, Massachusetts: MIT Press.

Dretske, F. (1988). Explaining Behavior. Cambridge, Massachusetts: MIT Press.

Fodor, J. A. (1987). Psychosemantics. Cambridge, Massachusetts: MIT Press.

Fodor, J. A. (1990). A Theory of Content and Other Essays. Cambridge, Massachusetts: MIT Press.

Markman, A.B. & Dietrich, E. (1998). In Defense of Representation as Mediation," PSYCOLOQUY 98(9).

van Gelder, T. (1995). What Might Cognition Be If Not Computation? Journal of Philosophy 92(7) 345.

van Gelder, T. (1998). The Dynamical Hypothesis in Cognitive Science. Behavioral and Brain Sciences, 21 (5): 615.

Volume: 9 (next, prev) Issue: 64 (next, prev) Article: 8 (next prev first) Alternate versions: ASCII Summary