The Role of Symbol Systems

This is a brief summary of Allen Newell's discussion of the role of symbol systems in cognition. This material is taken from Unified Theories of Cognition , section 2.5.

It is a physical law of nature that any processing must necessarily be done locally. Furthermore itt is a basic proposition of information theory that any given technology has a particular limit to the amount of encoding that can occupy a given region of physical space. Thus if a computational system is to have any sufficient complexity there must be some method for utilizing information that is not confined withing a limited region of space. A symbol, represented concretely by a symbol token, provides a means for representing distal knowledge. Since the symbol token is generally a more compact abstraction of the knowledge itself it can be manipulated in a more restricted region of processing space. The assumption is that the symbol token obeys the representational law , that is encoding knowledge X into symbol X', encoding transformation T into transformation T', applying transformation T' to X' to produce Y' and subsequently decoding Y' into Y (in the format of the original knowledge) is exactly equivalent to applying T to X to produce Y. If an elaboration of the symol in terms of the original knowledge is actually needed the symbol also provides the means for accessing the distal knowledge represented by the symbol, i.e. an address.

Symbols are not useful in and of themselves but rather are components of symbol systems, which have the following characteristics:

Memory

Symbols

Operations

Interpretation

Capacities

Newell argues that these characteristics are sufficient for a symbol system to produce all computable functions, that is it is universal. If a large variety of knowledge and goals are to be represented, distal access and univerality are necessary features of the ensuing knowledge system. Thus, a symbol system can realize a knowledge-level system, albeit imperfectly. It therefore follows that a cognitive architecture designed to approximate a knowledge system should have a symbol system as its basis.

Newell argues that the human cognitive architecture is itself realized by a symbol system. His argument rests on his assertion that - given enough time and external represenational ability - humans can approximate a universal machine. Though this hypothesis is obviously impossible to determine experimentally, he takes our efflourescence of adaptation as empirical evidence of his hypothesis. Humans are able to produce such a wide variety of response functions in such a wide variety of situations that it appears that humans are universal machines. It does not seem reasonable that humans would have every single response function included in their cognitive architecture so it would follow that we are actually composing response functions from a much smaller set. Furthermore the human distributed memory system requires some sort of distal access. Since a symbol system can approximate a universal machine and provides the capability for distal access and function composition it appears to be a natural choice as a foundation for the human cognitive architecture. Any cognitive architecture modelled after the human cognitive architecture, such as Soar , should then be built to support a symbol system.


Click here to see a list of architectures that maintain a symbolic world model.
Back to List of Other Meta-level Discussions