symbolism

Symbolism

To provide some intuition, a first definition of a symbolic architecture is an architectures that utilizes symbol manipulations in a fixed manner to represent its processing. Subsymbolic architectures do not use symbols to represent their processing. A common alternative to symbolism is to use analog representations and transformations.

Now the term symbolic architectures will be defined in more detail. A natural question to ask is what is a symbol? Allen Newell considered this question in Unified Theories of Cognition. He differentiated between symbols (the phenomena in the abstract) and tokens (their physical instantiations). Tokens "stood for" some larger concept. They could be manipulated locally until the information in the larger concept was needed, when local processing would have to stop and access the distal site where the information was stored. The distal information may itself be symbolically encoded, potentially leading to a graph of distal accesses for information.

Newell defined symbol systems according to their characteristics. Firstly, they may form a universal computational system. They have memory to contain the distal symbol information, symbols to provide a pattern to match or index distal information, operations to manipulate symbols, interpretation to allow symbols to specify operations, and, capacities for there to be: (a) sufficient memory, (b) composibility (that the operators may make any symbol structure), and (c) interpretability (that symbol structures be able to encode any meaningful arrangement of operations).

Finally, Newell defined symbolic architectures as the fixed structure that realizes a symbol system. That it is fixed implies that the behavior of structures on top of it (i.e. "programs") mainly depends upon the details of the symbols, operations and interpretations at the symbol system level, not upon how the symbol system (and its components) are implemented. How well this ideal hold is a measure of thestrength of that level.

The advantages of symbolic architectures are:

1. much of human knowledge is symbolic, so encoding it in a computer is more straight-forward
2. how the architecture reasons may be analogous to how humans do, making it easier for humans to understand (the flip-side of 1)
3. they maybe made computationally complete (e.g. Turing Machines)

   Subsymbolism

   (I use the term subsymbolism because these approaches could be used to implement a symbolic system at a higher level -- they do not necessarily preclude symbolism.) As Newell pointed out agents that react in their environment must do three basic mappings:
   
   1. from the environment to that of internal representation (sensing)
   2. internally between representations, ("thinking") and
   3. from the internal representation to the environment (actuating)
   There are methods of doing this with out using symbols. Two well known alternatives are Neural Networks and Subsumption Architecture by R. Brooks. Brooks' observation is that some of the difficulties of computing the proper action given a rich analog environment may be avoided if a symbolic (discrete) internal model weren't used as an intermediary (cf. Subsumption Architecture). Brooks argues, for example, that it is very unlikely that flies recover 3-D images, reason about the goals of humans with fly-swatters, use naive physics to land on ceilings, etc.
   
   The advantages of subsymbolic architectures are:
   
   1. they may be faster, which is important for dynamic environments
   2. they may be cheaper