Co-Evolution of Bargaining Strategies in a Decentralized Multi-Agent System

Software agents will personalize smart devices to act autonomously on behalf of their human owner. In dynamical electronic commerce environments, these agents could buy and sell goods and services from each other, without using a central market maker. From a system perspective, this creates a decentralized and continuously changing multi-agent system with the need for coordination of supply and demand. In this article we show how such a multi-agent system may be decentrally coordinated while software agents bargain with each other under the constraints of incomplete information, non-equilibrium and time pressure. The agents adapt to a changing environment with an evolutionary learning mechanism. It can be shown that the multi-agent system as a whole shows emergent coordination in absence of a centralized coordination institution. Software Agents on Smart Devices The discussion about using small, mobile devices in the context of ubiquitous computing (Weiser 1991) or pervasive computing seems to be largely hardware-centric. The software components are mostly considered to passively collect information from the environment or centralized web servers and store it in the device’s data shadow somewhere in the Internet. Small, passive devices with cumbersome humancomputer interfaces and a constant need for interaction in ongoing transactions will soon create a demand for autonomously and proactive software components on these devices. Such software agents (Wooldridge 1999)can assist human buyers and sellers in digital business processes and environments, which are characterized by trading and money transactions, to save transaction costs. Digital Business Agents (DBAs) (Diebold Consulting (Ed.) 2001)will monitor other agents and the environment continuously, e.g. by making price comparisons between different suppliers in the onand offline world (Youll et al. 2000), in order to fulfill their design goal if utility maximization for their human owner. They will be able to enter into negotiation with many potential trade partners at once, reaching an acceptable deal and setting up a contract Copyright © 2001, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. in a matter of milliseconds (Preist 1998). In that case, the DBAs may execute a complete business transaction as silent commerce (Schenker 2000) in the background without the need for human intervention. Many, if not most of these transactions will occur locally and decentralized, directly between the devices. A prerequisite is that the human principals of the agents are able to define economic goals, preferences and strategies in computer processable data structures (Kraus 1997). A goal is an abstract representation of a bidder’s demand, e.g. a product identification with a price to match. Preferences can be classified as either hard constraints, which have to be matched unconditionally, or soft constraints, which can be ranked, e.g. a preference for certain colors (Guttman and Maes 1998). Strategies to reach the goals and to negotiate on maximal meeting of the preferences can either be rule-based/argumentative, gametheoretic or heuristic/adaptive (Kraus 1997; Lomuscio, Wooldridge, and Jennings 2000) The use of DBAs is not confined to the consumers, however. Similar software can also be used by producers and retailers. As the consumers use comparison shoppers that allow automated price comparisons between different suppliers (Doorenbos, Etzioni, and Weld 1997; Youll et al. 2000), sellers are expected to begin to dynamically post prices to negotiate with individual buyers (Kephart, Hanson, and Greenwald 2000). The research question pursued in this article is how different negotiation strategies of DBAs compete against each other in an unregulated environment, and the impact on the coordination of the environment as a whole. In the next section of chapter 2, we describe the technical requirements and the setup of a multi-agent system (MAS) that coordinates a model supply chain process using decentralized economic coordination. Some experiments with different bargaining strategies in the DBAs are described in chapter 3. The article concludes in chapter 4 with an outlook into possible application areas and the problems of linking economic concepts and technical realization of market coordination. An Experimental Market Coordination