O V E R V I E W 141
6.1 The physical symbol system hypothesis 142
Symbols and symbol systems 144 Solving problems by transforming
symbol structures 144
Intelligent action and the physical symbol system 150
6.2 From physical symbol systems to the language of thought 151
Intentional realism and causation by content 153
The computer model of the mind and the relation between syntax and semantics 155
Putting the pieces together: Syntax and the language of thought 157
6.3 The Chinese room argument 160 The Chinese room and the Turing
test 162
Responding to the Chinese room argument 163
The symbol-grounding problem 165
Overview
This chapter focuses on one of the most powerful ideas in cognitive science. This is the analogy between minds and digital computers. In the early days of cognitive science this analogy was one of cognitive science’s defining ideas. As emerged in the historical overview inPart I, cognitive science has evolved in a number of important ways and what is often called the computational theory of mind is no longer “the only game in town.” Yet the computational theory, and the model of information processing on which it is built, still commands widespread support among cognitive scientists. In this chapter we see why.
For a very general expression of the analogy between minds and computers we can turn to the physical symbol system hypothesis, proposed in 1975 by the computer scientists Herbert Simon and Allen Newell. According to this hypothesis, all intelligent behavior essentially involves transforming physical symbols according to rules.Section 6.1spells out how this very general idea is to be understood. Newell and Simon proposed the physical symbol system hypothesis in a
very programmatic way. It is more of a general blueprint than a concrete proposal about how the mind processes information. And so insection 6.2we turn to the version of the physical symbol system hypothesis developed by the philosopher Jerry Fodor. Fodor develops a subtle and sophisticated argument for why symbolic information processing has to be linguistic. He argues that the architecture of the mind is built around a language of thought.
At the heart both of the very general physical symbol system hypothesis and the very detailed language of thought hypothesis is a sharp distinction between the syntax of information processing (the physical manipulation of symbol structures) and the semantics of information processing. The philosopher John Searle has developed a famous argument (the Chinese room argument) aiming to show that this distinction is fatally flawed. We look at his argument and at some of the ways of replying to it insection 6.3. In the same section we explore a more general problem for symbolic models of information processing – the so-called symbol-grounding problem.
6.1
The physical symbol system hypothesis
In 1975 the Association of Computing Machinery gave their annual Turing Award to two very influential computer scientists and pioneers of artificial intelligence – Herbert Simon and Allen Newell. Simon and Newell were recognized for their fundamental contributions to computer science. They created the Logic Theory Machine (1957) and the General Problem Solver (1956), two early and very important programs that developed general strategies for solving formalized symbolic problems. In the lecture that they delivered as one of the conditions of receiving the award Newell and Simon delivered a manifesto for a general approach to thinking about intelligent information processing– a manifesto that was intended to apply both to the study of the human mind and to the emerging field of artificial intelligence. Their manifesto hinged on what they called the physical symbol system hypothesis.
Newell and Simon start their lecture by observing that many sciences are governed by certain very basic principles (what they called laws of qualitative structure). So, for example, biology has the basic principle that the cell is the basic building block of all living organisms. Geology is governed by the basic principle (enshrined in the theory of plate tectonics) that geological activity on the surface of the earth is generated by the relative movement of a small number of huge plates.
In their lecture they propose the physical symbol system hypothesis as a comparable law of qualitative structure for the study of intelligence:
The physical symbol system hypothesis: A physical symbol system has the necessary and sufficient means for general intelligent action.
There are two claims here. The first (the necessity claim) is that nothing can be capable of intelligent action unless it is a physical symbol system. Since humans are capable of intelligent action, this means, of course, that the human mind must be a physical symbol system. In this sense, then, the physical symbol system hypothesis comes out as a
constraint upon any possible mental architecture. The second (the sufficiency claim) is that there is no obstacle in principle to constructing an artificial mind, provided that one tackles the problem by constructing a physical symbol system.
The plausibility and significance of the claim depends on what a physical symbol system is. Here are Newell and Simon again:
A physical symbol system consists of a set of entities, called symbols, which are physical patterns that can occur as components of another type of entity called an expression (or symbol structure). Thus a symbol structure is composed of a number of instances (or tokens) of symbols related in some physical way (such as one token being next to another). At any instant of time the system will contain a collection of these symbol structures. Besides these structures, the system also contains a collection of processes that operate on expressions to produce other expressions: processes of creation, modifi- cation, reproduction, and destruction. A physical symbol system is a machine that produces through time an evolving collection of symbol structures.
With this passage in mind we can break down Newell and Simon’s characterization of physical symbol systems into four basic ideas.
1 Symbols are physical patterns.
2 These symbols can be combined to form complex symbol structures.
3 The physical symbol system contains processes for manipulating complex symbol structures.
4 The processes for generating and transforming complex symbol structures can themselves be represented by symbols and symbol structures within the system.
Before going on to explore these in more detail we should pause to note (without much surprise, given that the physical symbol system hypothesis is the brainchild of two computer scientists) that the description of a physical symbol system looks very much like an abstract characterization of a digital computer. We might think of the physical symbols mentioned in (1) as corresponding to the alphabet of a computer language. One very common computer alphabet is the binary alphabet {0, 1}. The symbols in the binary alphabet can be combined into strings of 0s and 1s that are the“words” of the computer language. Computers work in virtue of procedures for manipulating strings– as suggested in (3). Some of these procedures are very basic. These are the programs hard-wired into the computer and written in what is usually called machine language. But, as implied by (4), computers can run programs that“instruct” the basic procedures to operate in certain ways and in a certain order. These programs are written in higher-level programming languages. We need to look in more detail at each of the basic ideas (1) through (4) before seeing how they might be combined in a particular model of information processing. Thinking about Turing machines will help bring out some of the issues here. Turing machines were introduced in section 1.2as abstract models of computation. Newell and Simon make clear in their paper how Turing’s work on Turing machines in the 1930s was the first step towards the physical symbol system hypothesis. Now would be a good moment to look back atsection 1.2.