Content area
Abstract
A number of useful search and optimization techniques are based on analogies to the natural world. This thesis considers whether there could be a market algorithm, analogous to a genetic algorithm but based on market principles, that could be used to solve computational problems. Can we do with the invisible hand what was done with the blind watchmaker? Importing market-based techniques from economics to bear on computational problems seems more subtle than importing genetic algorithms from biology. A number of pitfalls arise when one attempts to use the market system and theories from economics to organize multi-agent systems: implementation overhead, Pareto-efficiency restrictions, computational costs to agents, game theoretic dynamics, and the approximate and adaptive nature of the market system. We examine each of these and discuss the limitations of the market system in terms of optimization and adaptation. We argue that economic principles apply differently to artificial agents.