Saturday, January 11, 2014

Subsumption Architectures and Common Currencies

One motivation for asserting a common currency is in explanations of orderly behaviour. In this case the common currency is a 'proximal' currency, in the sense that it is part of the cognitive economy of whatever is doing the behaving. (For a note on the distinction between ultimate and proximal currencies see this posting.)

Shizgal and Conover provide a pithy statement of the inference to a proximal common currency here:

In natural settings, the goals competing for behavior are complex, multidimensional objects and outcomes. Yet, for orderly choice to be possible, the utility of all competing resources must be represented on a single, common dimension (Shizgal & Conover 1996).

(For a discussion of Shizgal and Conover's paper see this posting.)

One challenge to this inference arises from those who think that it is an error to attempt to explain orderly or appropriate behaviour by reference to representations of any kind. If there aren't any representations, then there aren't any representations of value (or utility, or reward, etc.). And if there aren't any representations of values, there isn't a proximal common currency. A leading figure in the anti-representation tendency is Rodney Brooks, formerly of MIT, whose 1991 manifesto 'Intelligence without representation' is still regularly cited.

The aim of the working paper 'Subsumption Architectures and Common Currencies' is to assess Brook's challenge in the specific area of motivation, or preference. This involves a slight change of emphasis from most discussion and debate over Brooks, because that debate focused largely on representations that were counterparts to beliefs (in the sense of representing facts about what the world was like) rather than representations which were counterparts to desires (in the sense of representing what actions were worth doing, or what goals had what current value).

The positions of Brooks on the one hand, and Shizgal and Conover on the other, seem clearly opposed. If orderly choice requires a common currency that is internally represented, then whatever subsumption architectures can do, they can't produce orderly choice. But if they can produce orderly choice without central representations of value, then Shizgal and Conover are incorrect about the necessity for such representations.

Assessing the challenge that Brooks' work poses involves getting clearer about what kinds of order behaviour can exhibit, and also what kinds of processing and internal working can produce what kinds of order. I've presented working talks on this at a few departmental colloquia, but not yet taken the topic to a conference or produced a written version.

This paper is now being worked on collaboratively, by myself and Blaize Kaye. Our first conference presentation on the topic (by him) will take place in January 2014. And I hope that we'll be able to post a working draft within a few weeks. 

Related postings on this blog:

Important sources: Rodney Brooks


No comments:

Post a Comment