Research Agenda

In order to anticipate and influence how large-scale information economies are likely to behave, we take a two-pronged approach. First, we use analysis and simulation to study the collective behavior of models of large populations of software agents employing a variety of economic protocols and adaptive utility-maximization algorithms. Some of our research papers have concentrated on a simple model of an information filtering economy, such as might be embedded in a larger information economy. The model is inspired by information dissemination services that can be found on the Internet today, and sets them in an economic context. In particular, we have focussed on issues of market efficiency, spontaneous niche specialization, and cyclical price wars that may threaten the viability of low-friction economies. The work on price wars has received some media attention. We have also studied models dealing with shopbots and their likely impact on markets, information bundling by software agents, and multi-agent learning in an economic context. Moreover, if the vision of information economy is to be realized, then it must be demonstrated that, within their domain of application, agents can attain a level of economic performance that rivals or exceeds that of humans on average, without introducing undue risk. Otherwise, people would not entrust agents with making economic decisions. One aim in our work on bidding agents is to provide such a demonstration. Through a series of controlled laboratory experiments in which humans and agents participate simultaneously in a realistic auction (a Continuous Double Auction), we show that software agents can consistently obtain greater gains from trade than their human counterparts.

In our second approach to study the dynamics in information economies, we are developing a software platform that supports the interaction of tens of thousands of economically-motivated software agents, and provides common economic componentry such as pricing algorithms and negotiation protocols (developed using insights gained from the more theoretical studies). A first prototype of an information economy called BookMarket is being built on top of this platform. In BookMarket, agents buy and sell a variety of information goods and services that revolve around the sale of books, including various matchmaking and brokering services and the provision of information about prices and reputations. To the greatest extent possible, the agents are based on services already available on the Internet.

The Information Economies project is supported by the IBM Institute for Advanced Commerce. A popular account of some of our work is given in an article in IBM Research Magazine.