Archive: August 21st, 2008

Allen Newell: Physical Symbol Systems

[Readings] (08.21.08, 11:16 am)

Symbol systems are most important development in recent work of cognitive science, linguistics, and psychology. An interesting note: “Thus it is a hypothesis that these symbols are in fact the same symbols that we humans have and use everyday of our lives. Stated another way, the hypothesis is that humans are instances of physical symbol systems, and, by virtue of this, mind enters into the physical universe.” (p. 136) This treatment is important because it establishes the symbol as the media by which the mind interacts with the physical world. Secondly, it seems to echo the notion of the Jungian symbol. Though this is probably not what Newell is intending to connect, the Jungian symbol exists at a cultural and semantic level, and could be used to extend symbol manipulation to be more of a trans-cognitive phenomenon.

Symbol systems have their roots in mathematical logic, as well as in previous work in philosophy, linguistics, etc. Newell mentions Whitehead specifically on the importance of symbols. The work on symbols is existing in parallel between cognition and computer science. Newell is arguing simultaneously for the importance of symbol processing in computation, and also in human thought. These two are intrinically linked in his argument, making AI a natural conclusion. This link extends from a broad cultural history of likening humans to machines in thought and function, stemming from Cartesian Dualism. The human created machine, through mathematics, is something that reaches towards the platonic ideal of pure disembodied meaning. The proclamation of AI is a natural conclusion from this thinking, where human thought belongs in this world of perfected formalism. That Newell should conclude that physical symbol systems are “simply evident” is a continuation of this mode of thought.

Formal definition of a symbol system “SS”: memory, operators, control, input, and output. Memory is a list of symbol structures, or expressions. An expression is a list of symbols, with a type and roles associated with the symbols. Newell writes an expression formally: (Type: T, R1:S1, R2:S2, … Rn:Sn). The number and the roles depend on the type, and the symbols may be repeated. An operator takes some symbols as an input and produces some symbols as an output. The symbol system has several operators, which seem to relate to classical computer IO and memory operations: assign, copy, write, read, execute, exit if, continue if, quote, behave externally, input. No example is given of this system in operation, so we cannot easily see how these properties will cause the system to behave.

The symbol system described is seems to be fairly “garden-variety”, but has the property of universality. This seems to be more than computational universality, but relate to interaction and responsiveness with input and environments. Newell compares this to Weiner’s Cybernetics, where systems used feedback to appear purposive. Newell points out that SS is limited because of its ability to behave in the world, and input symbols. The most significant limitation that he describes is on the limitation of computation, and the existence of non-computable functions.

Newell continues to express concern over this and describe computational universality and relating the capacity of symbol systems to the Church-Turing thesis. It seems, though, that this is getting beyond the problem of cognition. Cognition is about how humans think, and, by virtue of being physical ourselves, we are limited by the laws of computation. The concern over completeness seems unfounded to me. The extreme generalization of Turing’s minimal functions seems to imply that most any symbol system is bound to be universal.

Further embracing the idea of universal machines, Newell forms a definition of symbol systems: “Symbol systems are the same as universal machines.” (p. 154) This argument goes in the direction that symbol systems and universal machines are equivalent, or that they can simulate each other.

Applying this principle: The purpose of symbols is in their ability to signify or stand for something, and Newell describes this as the process of designation. Newell mentions several other words: reference, denotation, naming, standing for, aboutness, or even symbolization or meaning. “The variation in these terms, in either their common or philosophic usage, is not critical for us.” I find this casual rejection very fascinating, as the relation between cognition and the various forms of symbols, especially in terms of metaphor, (and in Piercian linguistics: symbols, icons, and indicies), these variations of meaning are extremely important. The designation intended by Newell is a mechanism used in the means by which one universal machine might represent and simulate another.

The other capacity is interpretation, the ability to derive symbols from given input. Later, discussing assignment, Newell mentions some examples of symbols that might be used to designate things. Symbols processed by machines must be totally and fundamentally arbitrary, even though the words and symbols used by humans in various contexts encode a great deal of information into the symbol itself. One particular example is the word “unhappy” which references the symbol “happy”, even though associated meanings may be more than mere opposites. Also mentioned are labeling conventions in geometry. These sorts of conventions are exactly the type of cognitive extensions that are encouraged by others.

The physical symbol system hypothesis: “The necessary and sufficient condition for a physical system to exhibit intelligent action is that it be a physical symbol system.” (p. 170) General intelligent action: “means the same scope of intelligence seen in human action: that in real situations behavior appropriate to the ends of the system and adaptive to the demands of the environment can occur, within some physical limits.” This does hold and is backed up logically, but fails to describe further context of the ends of the system or the demands of the environment. These details are things that must be carefully constructed and supplied. Furthermore, the hypothesis also is addressing specifically the idea of rationality, which is a loaded and biased term. The hypothesis focuses on rationality in prefrence to a more general “phenomena of mind”. This is described as a preference, but it is also a severe limitation, as humans are not necessarily rational.

Representation and knowledge also get a special treatment. Representation is the quality by which symbols might map from aspects of one object to aspects of another. The idea is that the symbol system has an image of the object in its symbolic structure. This is a useful concept, but is subject to a great deal of value judgements and concerns in terms of formulating the structure of representation. The relation to knowledge is framed in an equation: “Representation = Knowledge + Access”.

Reading Info:
Author/EditorNewell, Allen
TitlePhysical Symbol Systems
ContextNewell is one of the establishing voices in AI, and helped to pioneer traditional symbolic AI.
LookupGoogle Scholar