Introduction

The coming synthesis of agent technology and electronic commerce offers the potential for a free-market economy of interacting agents to emerge on the Internet. Electronic commerce would reap all the benefits that agent technology can provide; at the same time, the "almighty dollar" would fill a crucial function as built-in motivator and facilitator for the adoption, success, and survival of agents.

Let us therefore suppose that the Internet is evolving toward a a free-market information economy in which billions of software agents exchange a rich variety of information goods and services with humans and--much more commonly--amongst themselves [2, 4, 12, 14]. Is the emergence and survival of this economy guaranteed? The answer is far from clear. Within the context of market-driven optimization, a growing body of work demonstrates the success of economic mechanisms in coordinating the actions of agent populations [1, 3, 5, 11, 13]. But by definition, market-driven optimization presupposes a common overall goal to be striven for, in the form of a predefined measure of system performance, along with the implicit power of the system's designers to tune its parameters to help it achieve its optimum. In an open system like the Web, no such goal exists. Each agent has its own set of objectives, which may coincide with, conflict with, or (most probably) simply be incommensurable with, the goals of other agents [10]. The success or viability of the economy is determined on an agent-by-agent basis: if an agent can achieve its goals, the economy "works"; if not, it doesn't. Therefore the success of market-driven optimization can provide encouragement, but not proof, of a free-market agent economy's ultimate survival. Of even more concern is previous research on large systems of interacting, self-motivated software agents showing that they may be susceptible to the emergence of wild, unpredictable, disastrous collective behavior [8, 9].

All of the above motivates a general, wide-ranging study of economically motivated autonomous agents. In this and a pair of companion papers [6, 7], we have chosen to work within the context of a simple model of an information filtering economy, in which information brokers, which purchase selected information goods from a producer agent and resell them to selected consumers, play the central role. Ref. [7] examined the dynamics of the brokers' price-setting behavior in a competitive environment. This paper focuses on the complementary aspect of broker specialization, as measured by the brokers' individual selections of which, and how many, types of information good to carry.

In section 2, we describe the details of the model. Section 3 presents analytic results on specialization in an economy containing only one broker, which are then applied in section 4 to help understand the behavior of a multi-broker system. We close with a discussion of the overall results and of directions for future work.