PAICS: Philosophy of AI and Cognitive Science

Arguments against internal representations

In a few weeks, Zoltan Dienes will give his excellent COGS Open Lecture in defence of the notion of sub-personal representation. He and I were chatting about this, and I volunteered to provide him with a list of the arguments against internal representations that I have encountered. Here’s what I came up with. If you have any additions/corrections/questions, please comment!

Internal representations…

…Get in the way/are a bottleneck; the world is its own best model (Brooks)
…Are homuncular (Harvey)
…(the term) means so many things to so many people that it is meaningless/useless (Harvey)
…(the concept) is incoherent; our only clear notion of representations are public ones (Harvey)
…Are static, whereas cognition is dynamic (van Gelder)
…Are observer-relative (syntax) (Searle)
…Are observer-relative (semantics) (Searle)
…Require an indirect theory of perception, which is false
…Are not explanatorily necessary
…Are not computationally/mechanically necessary
…Are not found in the brain
…Have externally-individuated content, but computation/mechanism must be local
…Presuppose a sharp divide between subject and object, mind and world, whereas subjects enact their worlds. (Varela)
…Suffer from the frame problem
…Require an untenable sense-model-plan-act cycle
…Are disembodied, whereas cognizers are embodied
…Assume a sharp subject-world boundary, but there is no such boundary
…Must be grounded to do explanatory work, but cannot be
…Are based on a technological metaphor
…Cannot account for: qualia, emotion, pain, intransitive consciousness, etc.
…Are inert: representing the world to be P cannot ever cause someone to do something
…Are discrete/atomistic, while real mental content/propositional attitudes are holistic
…Can give enabling explanations of mental states, but not constitutive accounts of them (McDowell)
…Require an algorithmic/formal account of mind, whereas Gödel proved the mind is non-algorithmic/non-formal
…Are sub-personal and thus only have “as-if” content, whereas mental states have genuine content (McDowell)

For what it’s worth, I don’t think any of the above show that the notion of internal representation is not a valuable one for explaining the mind.

About these ads

November 14, 2010 - Posted by | General, philosophy of mind

8 Comments »

  1. Here’s another one, from Michael Morris:

    Internal representations…

    …Have stipulative content, whereas a process is only really cognitive if its content is non-stipulative

    Comment by ronchrisley | November 14, 2010 | Reply

  2. I have a couple of quibbles about what is Harvey’s position, having emailed back and forth with him at some length on this topic.

    I don’t think he’d say that internal representations *are* homuncular; rather, that any discussion of mental representations (his preferred term) inevitably invites an implicit homuncular reading, unless one is clear (explicitly) about who is doing the representing, and to whom. I happen to agree with him on this point.

    Also, I know from that discussion that Harvey’s (current, anyway) view is not that “our only clear notion of representations are public ones”. He accepts (whereas I do not) a distinct category of mental representations. His position is, instead, that *public* representations are primary and mental representations are parasitic on the public ones.

    Comment by jackoflantern | November 14, 2010 | Reply

    • Thanks Joel. I should have made it clear that the attributions are only meant as a rough guide, and are not meant to be academically rigorous. For example, it might indicate a position that was once, but is no longer, held by the person mentioned.

      Also, since each argument is presented as a single line, it will usually be inaccurate if taken to be a description of the argument itself. I was only intending each of them to a mnemonic that might remind one of an argument already encountered before.

      That said, I don’t agree with Inman that his four-place relation (“P is used by Q to represent R to S”) applies to all notions of representation. Perhaps it works well for most cases of communicative representation, but in the case of what most people mean by “internal” (I prefer “cognitive”, or “mental”, despite the title of my post) representation, no subject need be aware of the existence of the representation, let alone intend it to be for anyone. So unless S (and Q) can non-metaphorically be something less than a subject (in which case talk of homuncularity is misplaced), the four-place relation doesn’t typically apply to cognitive representations. To insist that it does is to insist on a homuncular (mis-) understanding of internal representation. It is also to insist that our only coherent notion of representation is the notion of representations for public communication (one of the three views I attributed to Harvey in my list).

      Most people who talk of cognitive or internal representations intend for them to be naturalisable, and thus they intend them to have a semantics that is as objective and observer-independent as any other scientifically respectable property. So they reject the claim that the notion of representation they are dealing with is one that is relative to some subject S (or Q) whom the representation is “for” or who it is “being used by”.

      The nearest approximation to Harvey’s relation that I can come up with that does justice to the notion of internal representation as usually used by exponents of the notion in philosophy of cognitive science is:

      “P sub-personally represents R and thereby partially explains S’s personal-level representation of Q”

      (where Q is usually R).

      Comment by ronchrisley | November 14, 2010 | Reply

      • As I was typing my reply, an email arrived in my Inbox, sent by Inman, advertising the next Alergic talk: Paul Chorley form the CCNR on “Cortical Mementos: Representation and Memory in Reinforcement Learning”. So perhaps this objection to internal representations is de facto withdrawn?

        Comment by ronchrisley | November 14, 2010

  3. Actually, you have missed one, which is the most important in my view, though difficult to summarize other than with a label.

    Internal representations..

    …fall victim of Wittgenstein’s rule-following considerations. (Wittgenstein(!), Brandom)

    Comment by Tom Beament | November 16, 2010 | Reply

    • Really, “most important”? I don’t know what Brandom does with this, but as far as Wittgenstein himself is concerned, I think his arguments are not directed against internal, sub-personal representation (especially since the notion wasn’t really in currency then). Internal representations are intended to constitute a subject, rather than be person-level representations that subjects interpret and use (the target of Wittgenstein’s argument). Is there a good reference for what Brandom does with this?

      Comment by ronchrisley | November 17, 2010 | Reply

      • Yes, most important! Of course, you are right about Wittgenstein not specifically targeting the idea of internal representations, but those who have used the rule-following considerations to do so explicitly invoke him. Brandom has done so in several places but one I can find easily is his paper “Modality, Normativity, and Intentionality”, Philosophy and Phenomenological Research Vol. LXIII, No. 3, November 2001, and reprinted in the latest edition of the Lycan (and Prinz) collection . However, the clearest presentation of this argument is in Tim Thornton’s book ‘Wittgenstein on Thought and Language’. The connection between the rule-following considerations and representation is normativity. I presume that internal representations must be normatively evaluable, otherwise they wouldn’t be representations, and then the rule-following considerations apply.

        Comment by Tom Beament | November 17, 2010

      • A better way of putting the argument:

        Internal representations…

        …must be normatively evaluable, but Wittgenstein’s rule-following considerations show there can be no naturalistic account of normativity (Brandom, (Tim) Thornton)

        I think something like this must lie behind the arguments you attribute to McDowell and Michael Morris – it’s why internal representations can only have ‘as if’, or stipulated, content. Otherwise you could accuse them of begging the question (and I’m sure you would!).

        Comment by Tom Beament | November 18, 2010


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Follow

Get every new post delivered to your Inbox.

%d bloggers like this: