Dennett and Fodor

1
Dennett and Fodor

• Nyelv és elme, 2007 tavasz

2
Three Kinds of Intentional Psychology

• Folk psychology
• Intentional system theory
• Sub-personal cognitive psychology

3
Intentional reduction

• Church Dennett
• Effective folk psych
• procedure
• Turing machine Intentional system
• Every mental phenomenaon of folk psych is
  characterizable in terms of intentional systems.

4

• The final step of reduction is that of folk
  psychology to intentional system theory.
• Microrealization is diverse and no particular
  type of it is the mark of the mental.

5

• Intentional stance the predictions it gives rise
  to are more heuristic, more prone to error
• (The design stance is better at prediction the
  physical stance is the best) ? IF we can handle
  it. But we cannot run a chess program in our
  head it is not feasible
• Intentional interpretation is more economical and
  more heuristic, still it is fairly powerful in
  its predictions.

6
The ultimate nature of intentionality

• (1) We attribute intentional states to other
  people and some systems.
• (2) This is all there is to intentionality if a
  systems behavior is sufficiently powerfully
  predicted by taking the intentional stance by
  attributing to it beliefs, etc. then that
  system is an intentional system ? moreover, to be
  an intentional system is to have beliefs.
• Two examples (1) The light-searching vehicle ?
  at a maximum, an extremely simple intentional
  system (the design stance may be good enough to
  explain its behavior)
• (2) A volcano feared by some native people
  despite possible animistic projections it is not
  an intentional system (eruptions not well
  predicted by belief-attribution).

7
Is this notion of intentionality vacuous?

• Skinners behaviourism every use of the
  intentional stance is in reality a use of the
  design stance. That is powerful enough to predict
  behaviour.
• Dennett Holistic logical behaviourism the
  intentional stance is real and needed. Skinners
  success was due to such a simplification of the
  system and environment that the design stance
  became useful (rats pressing levers, etc.)
• Another example of the ineliminability of
  intentional talk

8
The two black boxes
? ?
B
A
? ? red ? ? green Cut the wire ? no light goes
on.
9

• On pushing either button, a very complex signal
  (10000 bits long) went through the wire.
• Scientists opened box B a supercomputer was
  inside. Input stings were processed by a program
  and resulted in the corresponding light going on.
  
• Every sting sent by A was either a red or a green
  string.

10

• However, changing any such string virtually
  always resulted in the amber light going on.
• The overwhelming majority of the strings were
  amber. (Still there were lots of red and green
  strings, and A output only such strings.)
• There appeared no systematicity in which string
  was in which category.

11

• Open box A ? another supercomputer which
  generated the strings according to a complex
  algorithm.
• The programmers interviewed what unified the red
  strings was that they were binary versions of
  true English sentences typed in ASCII code.
  (Green false sentences. Amber neither of the
  foregoing.)

12
Dennetts suggestion

• Intentional idioms immediately helped to reveal
  the regularity in the systems behavior. Now
  nothing was easier than to generate red and green
  strings, whereas before, nothing was more
  difficult.
• The regularity captured by the semantic notion
  gives rise to a causal regularity it is not
  epiphenomenal, says Dennett.

13
Propositional Attitudes

• Constraints on a theory of PAs
• I. PAs are relations.
• Reasons to believe
• (i) Intuitive plausibility
• (ii) Existential generalization (John believes
  its raining ? there is something John believes)
• (iii) Taking believes-that-p an atomic predicate
  is an implausible view.

14

• II. Account for opacity
• (Substitution of coreferentials plus substitution
  of logical equivalents is not allowed in the
  scope pf PAs. A theory of PAs should explain this
  datum.)
• III. The objects of PAs have logical form. Mental
  states have to be compositional, if they are to
  explain behaviour in an effective way.
• IV. The account of PAs should mesh with empirical
  accounts of mental processes.

15

• Carnaps theory PAs are relations between people
  and sentences.
• I. is satisfied.
• II. Freges condition the opacity of belief is a
  spacial case of the opacity of quotation.
• III. Logical form of the objects of PAs no
  problem ? the sentence is the object of the
  belief the sentence has logical form done.
• IV. This is the interesting part. Some evidence
  cited that even on this count Carnaps theory
  does quite well.

16
However

• Fodor has objections.
• 1. The identity condition of beliefs differs from
  that of sentences (Bill bit Mary Mary was
  bitten by Bill. 148-9 different sentences same
  belief)
• 2. Belief and language (X believes that its
  raining X can believe this even if he does not
  speak English) (Belief in non-linguistic animals?)

17

• 3. When the correlation between sentence
  complexity and processing costs breaks down, that
  is evidence against the objects of PAs being
  sentences.
• Color codability experiment ? English and Dani.
• In fact this is a poor example. Color recall and
  recognition is likely mediated by perceptual
  categorization, and not by descriptions
  formulated in mentalese. (Both are independent of
  ones natural language.)
• Text comprehension and subsequent sentence
  recognition is better evidence for an underlying
  propositional representation (Sachs, 1967) Fodor,
  (1975) has more cases in point.

18

• 4. First language acquisition has to be based on
  an already existing representational system.
• Fodors alternative sentences in an internal
  representatin system are the objects of PAs.
• This answers the above objections, and also meets
  the constraints.
• Mentalese is innate and compositional.

19
Contrasts between Fodor and Dennett

• (1) Whats in the head how are PAs realized?
• Fodor via mentalese formulas in the head.
• Dennett diverse realization in subpersonal
  mechanisms. Not a ground for generalization (?
  Martians might have beliefs as well)

20

• (2) The nature of mental content
• Both are externalist
• Fodor indicator theory (atomistic)
  nomological locking
• Dennett semantic properties are not just
  relational, but superrelational, the relation
  a particular vehicle must bear in order ot have
  content is a relation between the token and the
  whole life.
• ? A holistic view.

21
Fodor against holism

• Earlier Fodor used to hold an inferential-role
  view of content.
• His reason for abandoning it was that it could
  not avoid holism.
• If the content of a representational vehicle R is
  determined by Rs inferential/computational
  relations to other representations, then either
• A) we determine a subclass of other MRs which
  play a role in fixing Rs content
• B) We say that it is the whole computational
  system with all of its representations and
  processes which determines Rs content.

22

• Strategy A would need an analytic/synthetic
  distinction which, according to Fodor, cannot be
  attained.
• - Concepts cannot be defined in terms of other
  concepts
• - Correspondingly, it does not seem possible to
  outline a subset of MRs which determines Rs
  content (via being inferentially/computationally
  related to R).
• Strategy B very plausibly, in no two people will
  R bear all the same relations to other MRs. Thus,
  in no two people will R mean the same thing.
  Which means, intentional generalizations cannot
  be made.
• This is the price of holism and the reason for
  rejecting it.

23
Another line of argument

• Atomism about concepts and the semantics of MR is
  favorable, according to Fodor, because the
  combination of concepts is very hard to
  understand if we assume that concepts are complex
  entities with constituent structure.
• Fodor stereotypes dont combine. Therefore
  concepts cannot be stereotypes.
• The pet fish probem The pet fish stereotype
  should be somewhere in between the fish
  stereotype and the pet stereotype (carp vs.
  cat/dog). But the pet fish stereotype is goldfish
  not anything like a weighted combination of the
  constituent stereotypes.

24

• NOT A CAT if this is a concept, it has no
  stereotype.
• PURPLE APPLE presumably it is derived from
  PURPLE and APPLE (hey, we want to explain
  compositionality)
• So in the APPLE stereotype, the prototypical
  color red is replaced by purple.
• BUT this also means that, in the PURPLE APPLE
  stereotype, the color has a much higher weight
  than in the APPLE stereotype.
• (There cannot be non-purple purple apples,
  although there can be non-purple apples.)
• So, when we combine the elements we also change
  them (the weight change in the property list is
  something over and above mere combination).

25

• Fodor the structure of the stereotypes
  determines their inferential (computational)
  roles they play in the system.
• Thus the fact that stereotypes do not compose
  shows that inferential/computational roles dont
  compose either.
• This is Fodors other reason to abandon
  inferential role semantics (and, with it, the
  notion of narrow content).

26

• (3) Reduction
• Fodor inner mechanisms, computational entities
• Dennett intentional systems abstract theories

27
Fodors critique of Dennetts instrumentalism

• Dennett it is possible to be realist about
  intentional states without being committed to the
  reality of MRs.
• Fodor at the abstract level of intentional
  systems there is no way to explain why the
  intentional strategy works.
• Moreover, Fodor does have a positive account the
  compositionality (hence productivity and
  systematicity) of MRs.
• There is a relation of isomorphism between mental
  syntax and mental semantics syntax maps
  semantic relations.
• The source of this picture is classical AI
  research.

28
Dennetts argument against LOT

• Attributing propositional attitudes to a system
  can be helpful in predicting its behaviour even
  when the system obviously does not have any such
  states (example a chess computer).
• Our automatic behaviors are similar e.g., my
  being hungry and reaching for food this act
  need not be mediated by PAs, inference, etc. Yet
  saying He wants to eat of me does help the
  interpreter to predict my behaviour.