Enactive Representationalism

1
Enactive Representationalism?!
A Springboard for discussion.
2
Premises
3
1. Despite all the technical and
theoretical advances that have been made, it is
still difficult to construct an artefact that
is engaged with its environment in the rich sort
of way required by an enactive approach.
4

• ...Nevermind one that qualifies as a
  concept-possessing, concept-using agent.

• Even with get-your-hands-dirty robotics, we're
  still working out what an enactive approach
  really means.

• Nonetheless, one might well wish to construct
  artefacts with which to better understand
  cognition and conceptualization.

5
2. Understanding cognition requires more than
recognizing the role of context.

• Many GOFAI researchers were aware of the
  importance of context.

• Not just that agent is embedded and embodied in
  certain way.

6
3. More apparently human forms of cognition
build upon simpler, more clearly shared forms of
cognition.

• Abstract thinking builds upon dynamic physical
  engagement with the world.

• Knowing how knowledge underlies knowing that
  knowledge.

• So, too, conceptual knowledge and experience are
  built upon the non-conceptual.

7
4. What I take as an enactive view is necessary
to understanding cognition.

• Cognition is the result of dynamic, physical
  engagement with the environment.

• Boundary of organism/not-organism or
  organism/environment becomes blurred.

8
5. Pace Noë, a representational account (of some
kind) is necessary to understanding concepts.

• Concept sub-propositional component of
  thoughts.

• Concept mental representation.

• Concept something that meets something like
  Evans' Generality Constraint.

9
6. Experience and concepts go, at least to some
extent, hand in hand.

• Difficult to conceive of experience that is
  fully non-conceptualized.

• Experience is more than sensory experience
  (e.g., visual experience), and sensory experience
  is more than current sensory info.

• At least part of the something extra seems to be
  concepts/conceptual expectations.

10
7. To talk about representations, you need to be
clear what a representation is.

• Dictionary definition a likeness or image.

• That which is able to stand in place of
  something else, by virtue of, often at some
  highly abstracted level, some perceived
  resemblance.

• ...As opposed to symbols, where the relationship
  of the symbol to the referent is assumed to be
  arbitrary or at least unimportant.

11
8. Representationalism is not per se
incompatible with enaction.

• Ron probably a naïve representationalism falls
  afoul, sure, but a counterfactual
  representationalism (what sensory experiences
  would I expect to have, were I to engage in
  certain sorts of actions, such as eye movements?)
  need not.

• So what I experience as being over there may
  be what I would expect to see were I to foveate
  over there, because of, perhaps, what I saw the
  last time I foveated over there.

12
Clarifications
13
1. How's this different from Noë's approach?

• Noë sees himself as anti-representational.

• This avoids a Grand Illusion-type argument.

• Noë's approach more strictly forward looking.

• Ron's and my approach, at least as I'm looking
  at it, places more emphasis on the role of past
  experience.

14
So e.g., we see what our experience leads us to
expect to see, hear what we expect to hear, etc.
15
2. The work to date has focused solely on
specifying the non-conceptual content of
experience.

• ...Insofar as one can make any sort of clear-cut
  conceptual/non-conceptual distinction.

• But I think some of the same methods can be
  used for specifying conceptual content in a
  non-conceptual way, which I think is needed for
  avoiding a certain vicious circularity.

16
3. I'm not a (un?)reconstructed definitionist,
honestly!

• But concepts look like definitions for a reason.

• Sure, conceptual knowledge is not a collection
  of dictionary-style definitions.

• ...Nor is it a process of collecting such
  definitions.

• ...Even if you allow the definitions to be
  dynamically updated in interaction with the
  environment.

17
But that's getting warmer!
18
Though concepts may look like definitions when we
try to explain them, they are nonetheless the
result of our dynamic engagement both with our
environment as a whole and with the society of
which we are members so I have my personal
concept DOG, which may vary in greater or lesser
ways from the next person, or from my own concept
DOG at different points in time and I have the
concept DOG that is part of the social space in
which we all share.
19
The Questions
20
I want to build some kind of implementation to
demonstrate the potentials and pitfalls of the
theory of concepts I'm trying to develop as an
extension of Gärdenfors' work.
Doing this and constructing something that
qualifies as a concept-using, concept-possessing
agent at the same time is seriously impractical!
21
What looks much more tractable is an
implementation that might on the surface at least
not be so different from a traditional AI
application i.e., a computer model that one
interacts with by keyboard and screen.
Where does enaction come in?
22
Question One
How much of the cognitive and enactive
requirements of such a model be put off onto a
user (e.g., a test subject) dynamically engaged
with the artefact and with the subject's
environment?
23
Question Two
There's a lot of existing GOFAI applications out
there that are, for what they do, successful
applications. How might an enactive approach
allow us to re-interpret and re-conceptualize the
nature of those applications to see them in a
different light?