Joseph Mikhael

#20082998

Oct 25, 2001 (Week 7)

Phil 673 - Mental Representations

Robert Cummins: Representations, Targets and Attitudes;

Chapters 1-3

CHAPTER 1 – INTRODUCTION

Problem of Mental Representation

• 4 difficulties surrounding mental representation:

1. Content: Which contents are represented in the mind?

2. Form: What form does mental representation take (images, symbolic structures, activation vectors)?

3. Implementation: How are the mind’s representational schemes implemented in the brain?

4. Definition: What is it for one thing to represent another? (p.1)

• Book deals with 4, which is, according to Cummins a philosophical as opposed to an empirical issue, but will also look at hoe 1 – 3 constrain 4.
• My chapters deal mostly with how 1 constrains 4.
• 2 fundamental constraints on a philosophical theory of mental representation:

• Explanatory Constraint: The theory should underwrite the explanatory appeals that cognitive theory makes to mental representation (philosophical constraint).
• Implementation Constraint: The theory should be compatible with the best scientific stories about what sorts of things actually do the representing in the mind/brain (scientific constraint). (p.2)

Naturalizing Content

• In defining the content of representations, we want to use terms that are used in contemporary science.
• We also want to avoid semantic terms like ‘means’, ‘refers’, ‘true’ or else the definition would be circular.
• Lastly, we want to avoid intentional terms such as ‘believes’, ‘desires’, ‘intends’, since these bring about the same mysteries as those between representations and what they are representing.
• “We do the foundations of this science no service, then, if we define representation in mental or cognitive terms.” (p.4)

CHAPTER 2 – CONTENTS AND TARGETS; ATTITUDES AND APPLICATIONS

Beginning with Error

• Theories of mental representation are often criticized for a failure to do justice to misrepresentation, or what Cummins will henceforth refer to as error.
• Cummins generally looks at the causal theories in explaining this problem.
• The proposed solutions, according to Cummins, have been inadequate.
• So, instead of beginning with a theory of representation and account for error, we will begin with a theory of error which will place useful constraints on further theorizing about representation.
• This should at least remove the “staleness” of the disjunction problem
• (A v B ® A).

Targets and Contents

M = <k-kr1, K-KB4, k-kr2>

		k
	B
KN		K

	B	k
KN
	K

		k
B
	K
	KN

P1: the starting position P2: the position after M P3: the position R_P3

actually represents

		k
	B
KN		K

	B	k
KN
	K

		k
B
	K
	KN

R_P1: the representation R_P2: the representation of P2, R_P3: a representation of

of the starting position the position after M P3, the representation S

constructs to represent P2.

• S = chess machine.
• Subroutine: LOOK_AHEAD
• Series of possible moves created by LOOK_AHEAD to prevent a stalemate: M

• S’s use of M should lead to R_P2, but instead it creates RP3, which leads to a stalemate.
• Since M leads to a stalemate in R_P3, S rejects K-KB5 as a response to k-kr1.
• However, S’s tokening of R_P3 is error.
• There is a mismatch between what R_P3 needs to represent (P2) and what it does (P3).
• 1^st important distinction: P2, in this case, is the target of the system, while P3 is the content of the representation.
• “Tokening a representation is error when the target of tokening it on that occasion fails to satisfy its content” (p.6)
• R (representation), in this case, is a data structure that represents the position after M.

• But R_P3 does not represent the position after M, but P3, instead.
• “Targets are determined by the representational function of tokening a representation on a particular occasion in a particular context, not by the content of the representation tokened.” (p.7)
• Independence of targets from contents makes error possible.
• If one determined the other, then there would be no mismatch, and therefore no error.
• A theory that says what it is for R to be true of T (targets) is only part of the story about representations.
• Need a theory of target fixation, a theory that says what it is for R to be applied to T.

• 2^nd important distinction: Representations mean their content, but intend their targets.
• This is the difference between intentionality and meaning.
• Error occurs when there is a mismatch between what is meant and what is intended.
• How do targets get fixed?
• Representational function of the mechanism + the current state of the world ® Intenders.
• e.g. chess subroutine CURRENT_POSITION
• Named after its target.
• Mechanism that runs it (chess program) is an intender.
• One can evaluate if there is error by checking the content of CURRENT_POSITION and comparing it to the actual current position (although what we are actually doing is comparing the content with our own)
• Therefore, intenders, not the representations, determine the target.

Falsehood and Error

• 3^rd important distinction: Falsehood and Error.
• e.g. of error without falsehood: S represents x as G, but what is needed is a representation of x as F.
• However, if x really is G, there is no falsehood, but there is error in the intender.
• Targets are determined by the function of tokening a representation on a particular occasion, in a particular context Þ Satisfaction conditions vs. Truth conditions.
• Therefore, truth isn’t the opposite of error, but correctness is.
• “p will be error when the target is the proposition that q, regardless of the truth values of p and q.” (p.12)
• If r’s target is a false proposition, but it correctly applies to it, then there is no error, therefore no misrepresentation.

Illustration of Distinction Between Falsehood and Error

• It is also possible to misrepresent propositions that are true.
• e.g. Suppose you learn the following two propositions:

1. Letters are more easily recognized in the context of words than alone.

2. In chess, one should develop the queen early.

• 1 is true, but 2 is false, and this would make error seem similar to falsehood. But suppose you were to learn the following proposition,

3. Letters are more easily recognized in the context of superwords (set of words and pronounceable nonwords) than alone.

• However, if you forget 3, and were asked on a multiple choice test “Which of the following best describes the word-superiority effect?” and chose 1 over 3, you would be in error, even though 1, by itself, is true.

Representations, Applications and Attitudes

• Applications of representations are not representations themselves.
• e.g. POSITION_AFTER_M tokens R_P3, is actually the application that the position after M is P3. There is actually no representation whose content is ‘that the position after M is P3’.
• Vs. Fodor and Schiller, who have representations with characteristic cognitive functions.
• e.g. Fodor: ‘S believes that p’ = p is in S’s belief box
• e.g. Cummins: ‘S believes that the position after M is P3’ is an application whose content is ‘that the position after M is P3’
• Attitudes are relations to applications.
• The propositional content of an attitude is the content of an application of a representation.
• Attitudes are distinguished in type – beliefs vs desires vs intentions
• Applications, not representations, are correct or incorrect
• Proven by the fact that one can retract an application, not a representation
• e.g. can retract that ‘the sky is blue’, but not ‘sky’ or ‘blue’.
• Applications distinguish between attitude types.
• Attitude is the result of giving a cognitive role to an application of a representation to a target
• e.g. the belief that the current auditory stimulation is a bell Þ |bell| É cas.
• There is a representation for bell and cas, but their binding is an application because cas is such that it temporally contingent.
• e.g. ‘Socrates was a Greek philosopher’
• There is an intender pos (properties of Socrates) that is bound to ‘Greek’ and ‘philosopher’.
• Fodor’s use of representations for attitudes causes the problem of misrepresentation since one can confuse content with targets.
• Cummin’s use of representations removes it – two principles:
• Attitudes are applications with a characteristic cognitive function. The semantic content of an attitude is thus the semantic content of the constituent application.
• Applications are the result of applying a representation to a target. The semantic content of an application is that the representation hits the target. (p.16)
• Representations are true or false, satisfied or unsatisfied. Applications, however, are correct or incorrect.
• “The semantic function of a representation is to represent its content; the semantic function of an application is to hit its target.” (p.16)

Nesting Intenders

• Sometimes, the content for one intender is needed before another intender can be used ® nesting.
• e.g. need to satisfy M before you can use POSITION_AFTER_M
• This shows how there can be an unlimited number of targets, thus making a system systematic and productive.
• The target of tokening r is what S expects to find when r is accessed, not what S needs to represent to succeed.

Can Representations Determine Targets?

• No, for example, the function of tokening RP3 is not the same as the function of RP3.
• “The Eiffel Tower is my target: doesn’t necessarily mean that the target of this representation is the Eiffel Tower.

Target and Referent

• The problem above results from a conflation of targets and referents.
• e.g. representation R whose content is the right hand, whereas intender L’s target is the left hand
• When L generates R = error

Three Theories of Mental Content

• Therefore, we need three distinct theories to deal with mental content:

1. A theory of representational content.

2. A theory of target fixation.

3. A theory of application fixation. (p.20)

• We need the three in order to explain error.

Some Diagnostics

• We can’t just look for a theory of truth to account for error.
• We, instead, need to at least ask which truth the system is after on the occasion in question.
• Also come to the realization that beliefs with the same content can have different targets.
• e.g. contents-of-the-basket intender generates a representation of a cat, a belief whose content is that a cat is in the basket, but the target is the contents of the basket. Location-of-the-cat intender generates a representation of a basket, a belief whose content is that the cat is in the basket, but the target is the location of the cat.

CHAPTER 3 – MORE ABOUT ERROR

Forced Error and Expressive Adequacy

• Forced errors: representational scheme lacks expressive power to provide correct representations.
• e.g. Euclidean geometry inadequate to represent the structure of physical space.
• When a system S uses scheme R to representing a domain T that R is not capable of representing, it is forced to make the “best”, at best, representation.
• Forced error can occur when one groups mice and voles as mice when one does not know the distinction, but when the distinction is known, then the error is unforced.
• Some errors are difficult to determine whether they are forced or unforced.

Accuracy: Degree of Correctness

• Representational error and correctness come in degrees.
• Replace correctness with accuracy.
• However, accuracy is not necessarily linked with truth.
• Instead, accuracy is relation between representation and target.

Seriousness and Inaccuracy

• Seriousness not necessarily related to accuracy.
• Can have seriousness from small errors, or large errors that are not serious.
• Seriousness instead related to the effectiveness of a representation.
• Therefore, a correct representation is one that is effective. This drives Conceptual Role Semantics (CRS)
• “We require a clean distinction between accuracy and effectiveness if the concept of representation is to have the independent explanatory leverage that makes it an important primitive in cognitive science.” (p.27)