Sunday, 16 April 2017

Flynn’s Cat – Part 4: An Exploration of Embodiment

© 2017 Simon’s Cat Limited
< Flynn's Cat - Part 3

There are many methods of evaluating a theory. Simplicity, testability, fruitfulness, power to unify, and so on. So how does embodiment stack up?

Traditional cognitive science in some aspects, has had a lot of success and has a proven track record. It has deepened our understanding of the mind in an unprecedented manner.  It has a power to unify perception, attention, memory, language under the same explanatory framework. Embodiment’s ability to be applied equally well across the range of cognitive phenomena, has yet to be proven, but it is very early days.  Appeals to concepts such as affordances, meshing, world-making, etc have as yet, an uncertain status. That a cat and I might conceive of the world differently is a difficult theory to test. Virtual reality technologies may present possibilities of testing in this area, but any findings would still be speculative at best. Traditional cognitive science is testable, and in certain cases experimental results from studies in embodiment can equally be explained by traditional cognitive science, and yet its explanations are less certain.

It must however be conceded that traditional cognitive science does not do a good job in explaining all varieties of cognition.
Perception, pattern matching and Rodney Brook’s (1997) creatures appear to be better explained by dynamical methods than they do by traditional cognitive science and symbolic representation. There are however, representation-hungry problems, requiring abstraction that are not adequately served without the use of representations.  

If you’ve read my previous posts along with the title, you’ll know I’ve been accompanied in this exploration of embodiment by a cat.  In respect of the title of these post, I am aware that my surname doesn’t sound as impressive as the Austrian physicist’s, but I see no reason why Schrodinger’s moggy should be the only feline philosophical device in town.  So here’s the thing. The cat died in September 2016, and I’m writing this in April 2017.  There is no cat beside me as I write.  She never lived in the house I live in now and to my recollection, never saw a laptop or a cursor. So apart from calling me out as a liar, how do we explain what I’ve been doing?  I spent my childhood in the company of cats. I know their movements well and even though this particular cat never saw a cursor on a laptop, I have no difficulty in visualising her in point mode, following a cursor’s movements around a screen, as I have seen other cats do.  This is now what Clark & Toribio (1994) would describe as a “representation-hungry” exercise requiring strong internal representations and as such is not well served by a dynamical systems explanation or sense-act interactions. While representations are not required for some kinds of activity, it does not follow that cognition never requires representational states and as such, embodiment can only claim a portion of the explanations that traditional cognitive science seeks to address. Could dynamical systems be adopted as an additional tool for traditional cognitive science, broadening its approach and thereby its explanatory powers, but maybe not completely replacing it.

In this pursuit, could representations in embodiment take a different form?  A problem I have had from the beginning of this exploration into embodiment is the idea that representations are seen as discrete entities that can be extracted from a system. It is one perhaps borne out of the beginnings of traditional cognitive science but I’m not so certain still exists. While connectionism in its current form is criticised for its lack of body and world, it does show that the processing of symbols is not necessarily the only means of processing inputs, albeit in a distributed fashion.  It transforms inputs into outputs, and in a manner, represents something about the world through the state that the model enters into. The weights of the connectionist model can be interpreted as a form of knowledge detailing how inputs should be treated in pursuit of outputs. A representation is a stand-in for something else. If I get freckles from staying in the sun too long without sun-blocker, my skin is carrying information about my time in the sun. In a similar manner, a state of me and my nervous system comes to represent something in the real world, when it consistently correlates with that thing in the real world. But perhaps the represenational mechanism must also involve a functional role to cooordinate activity with its environment beyond just mere correlation. As Clark (1997) put it, "the system must be capable of using these inner states or processes to solve problems off-line, to engage in vicarious explorations of a domain, and so on [...] Strong internal representation is thus of a piece with the capacity to use inner models instead of real-world action and search". It would seem then, a redefinition or at least a clarification of what is meant by representation is required, and without it, embodied accounts of cognition, for all the possibilities that they open up, become limited in their ability to explain mental images, the ability to learn from previous experiences, examine options when making a decision, and myriad other cognitive activities that require something that at least acts as a representation would.



References
Brooks, R. (1997). Humanoid Robots. The Cog Project. Journal of the Robotics Society of Japan, 15(7), pp.968-970.
 
Clark, A. and Toribio, J. (1994). Doing without representing?. Synthese, 101(3), pp.401-431.

Clark, A. (1997). The Dynamical Challenge. Cognitive Science, 21(4), pp.461-481.

No comments:

Post a Comment