Conclusion

Our investigation of the dynamic behavior of an information filtering economy revealed at least two important effects: spontaneous specialization, which is generally desirable, and cyclical price wars, which are by and large undesirable even to consumers that may on the surface seem to benefit from lowered prices.

We found that specialization is driven by two distinct mechanisms working together. First, if the extrinsic transport and processing costs and are intermediate between the low-cost ``spam'' regime and the high-cost ``dead'' regime, a monopolist broker prefers to offer a small number of categories. Second, competition among multiple brokers encourages them to become monopolists in largely non-overlapping sets of one or a few categories. Niche specialization is typically desirable from the perspective of both the brokers and the consumers.

Standard models of price wars [18] typically lead to a stable point at which no one makes a profit. The news filtering economy is extremely prone to unstable limit-cycle price wars, behavior that can be traced to the multi-humped, discontinuous topography of the profit landscape. Price wars undermine the tendency of the system to efficiently self-organize itself. We found that cyclic price wars could be eliminated in a system of brokers and consumers that had little knowledge of the system state and very simplistic algorithms for updating prices and interests, permitting useful specialization to occur. However, one cannot conclude that individual ignorance leads to societal bliss. The conservative price-setting strategy makes the system less nimble, and more susceptible to failure. Furthermore, even if ignorance led to good collective behavior, it would hardly be a stable strategy: there would be a strong incentive to use a better informed or more intelligent agent that could outperform its weaker opponents. Other effects that may hinder price wars in human economies, such as explicit and tacit collusion, frictional effects, and spatial or informational differentiation are likely to be weaker in agent economies. Price wars may indeed prove to be a serious problem to contend with in large agent economies of any sort, and merit our continued attempts to understand and control them.