Modelling Bounded Rationality In Agent-Based Simulations using the Evolution of Mental Models

6 Conclusion

An evolutionary model of cognition has been presented which has some of the qualitative characteristics relevant to economic agents, namely:

satisficing rather than optimising behaviour;
flexible learning - it can cope with structural change*1, this is facilitated by the parallelism so that it can `flip' between models;

path-dependency - the agent's population of models forms the context for subsequent learning;

serendipidous - the learning process has the power to come up with models not envisioned by the programmer;

boundedly rational - both the number of models and the inference from them can be controlled;

realisable - the implicit parallelism of the evolutionary model makes it a credibly fast model;

open-ended - the structure of the genome allows for theoretically unlimited expressiveness of the agent's models.

It does this using a class of models (evolutionary models) that is being increasingly studied, formalised and understood. Also it allows the programmer to introduce the following aspects of behaviour in a natural way:

the impact of a priori knowledge and the bias of the internal language of representation - since this is explicitly determined by the programmer;
the interaction of learning and inference;
the different types of model evaluation - there are many possible ways of deciding a model's fitness, including the accuracy of the models, the utility the models would have gained, and various aspects of cost and complexity;
different mixes of genetic operators [9].

At the moment such models only indicate their possible use as relevant models of cognition in economic agents, but I hope that the examples presented here persuade you of their potential expressive power.

Modelling Bounded Rationality In Agent-Based Simulations using the Evolution of Mental Models - 17 MAR 98

[Next] [Previous] [Top] [Contents]

Generated with CERN WebMaker