Embodiment and the philosophy of mind Clark 2001

notes (presentation outline)

First I will cover the general idea -- establish common ground: Clark is taking a conciliatory position between GOFAI representationalism and 'embodied' approaches. Judging from the reactionary stance that members of the class took towards simon, newell, and abelson during last week's discussion, I'd like to dwell a bit here and see if we can fish out any folks who are still apprehensive about taking Clark's very reasonable stance on represtations:

traditional approaches make a lot of sense for certain kinds of

thought, specifically the kinds of thought that traditional approaches have been most successful with like, 'internal' things like math, logic, reason and long term planning, abstraction, math and logic.

Second: An account of embodiment as compared to representationalism (trad. AI), in 6 areas (goals, definitions, angle, assumptions, tools, rel to psych and future).

1) goals: Of four possible goals of AI (taken from Russell/Norvig):

|Systems that think like humans. | Systems that think rationally.|

|Systems that act like humans | Systems that act rationally |

Embodied is mostly (most successfully) lower left and GOFAI is mostly (most successfully) upper right

2) definitions (spec. definition of intelligence) besides respective inclinations towards speaking of Intelligent Thought and Intelligent Behavior (see above difference in goals) they share generous ( functional) (empirical) definition of intelligence (c.f. minimally cognitive agents)

3) top-down vs bottom up A top-down account of cognition: GOFAI went straight for the thinking that we experience, postulating atoms of thought that are manipulated.

Accounts of lower level processes (like action) are pursued with the

same assumptions.

Embodied approaches are generally bottom up -- walk before you (run or plan next year's vacation)

4) assumptions GOFAI assumes atoms of thought, rationality. In application and theory, it tends to assume unlimited time, knowledge and computational power

Embodied approaches assume tightly coupled interaction of brain, body and environment

5) Tools / Consequences of above: GOFAI is more an extension of engineering than of cog sci having been much more successful with engineering applications than explaining human behavior. The mathematics behind trad. AI is analytic, and humans can do it with pen and paper.

Computers have made embodied approaches scientific ones. The math behind dynamical approaches is numerica,. Giving up on pretensions towards provability and complete information make researchers in embodied tradition comfortable with using limited information and power available to get behavior, and benefit from evidence and models in which humans do the same. Embodied approaches are not analytically tractable, and don't attempt to be. Allows/favors the following inclination: "real-world real-time responsiveness is clearly in some sense primary"

6) relationship to psychology AI looking at reason. Embodiment looking at psychophysics and motor studies.

7) Future In clarks account (and mine) future is bridges. Difficulty here is that all current tools and results in dynamical systems have a ceiling as to the complexity of given task. dynamic account of chained/sequential actions are weak, and dynamical accounts of the atoms of thought (to the limited extent that they will turn out to exist) are no more elaborate than the following: "Internal representations, then, may be realized not as simple inner states but as dynamical patterns of just about any conceivable kind".