Modelling Bounded Rationality using Evolutionary Techniques - Bruce Edmonds

3 The Agent Architecture

Although economic agents primarily develop though a process of incremental learning they also use some deductive procedures. In real economic agents these processes may be arbitrarily mixed as well as developed and abstracted over different layers of an organisation. Here we will only look at a model which effectively separates out learning and deduction and comes from an essentially unitary agent structure.

The agent works within a given a priori body of knowledge (e.g. accounting rules). The agent may well make deductions from this in a traditional way and apply these to the current hypotheses. This body of a priori knowledge may also determine the syntax of the models the agent starts with, its principal goals, default actions, fitness functions and the operations to be applied to its models. Typically much of this a priori knowledge can be made implicit in the syntax of the genome (which is the approach we have tended to take).

The agent here has many models of its environment. Once started the agent incrementally develops and propagates these models according to a fitness function which is based on its memory of past data and effects of its actions as well as the complexity and specificity of its models. It then selects the best such model according to that measure. From the best such model and its goals it attempts to determine its action using a search-based, deductive or quasi-deductive mechanism. It then takes that action and notes the effects in the environment for future use. The setup is illustrated below in figure 1.

The development of these models (i.e. the learning) is modelled by an evolutionary process on this population of internal models (similar to that described in [4]). Important restrictions on such agents include the fact that it may have only limited information gained as the result of inter-action with its environment and that any action costs it so that it can not indulge in an extensive exploratory search without this being weighed against the benefit being gained (this is especially true given the course temporal graining of typical economic simulations).

Figure 1: Basic Structure of a Simplified Economic Agent

Generated with CERN WebMaker