Overview

Now that we have the theory behind the physical symbol system hypothesis clearly in view we can explore its application to particular information-processing problems. We have already looked at one example of what is often called the symbolic paradigm. This is the SHRDLU program written by Terry Winograd and discussed in section 2.1. SHRDLU inhabits a virtual micro-world. It uses a simple language program to describe that world and to receive instructions about what actions to perform. It would be a very useful exercise at this stage to go back to section 2.1 in the light of the discussion in the previous chapter and work out how and why SHRDLU illustrates the basic principles of the physical symbol system hypothesis.

In this chapter we look in detail at three more applications of the symbolic paradigm. The first comes from research in Artificial Intelligence (AI) into expert systems. This is one of the domains where the symbolic approach is widely viewed as very successful. Expert systems are designed to simulate human experts in highly specialized tasks, such as the diagnosis of disease. They standardly operate through decision trees. These decision trees can either be explicitly programmed into them or, as in the cases we are interested in, they can be constructed from a database by a machine learning algorithm. In section 7.1 we see how machine learning algorithms illustrate Newell and Simon's heuristic search hypothesis.