Secure Electronic Marketplaces Based on the Multi Agent System Avalanche

Abstract

The future of electronic commerce will be shaped by open, heterogeneous and complex structures, consisting of many independent marketplaces. For conducting transactions, businesses have to be present at many of these marketplaces all over the Internet. But software agents, who interact and negotiate on behalf of their human (or organizational) principals, can act as representatives here. The multi-agent system AVALANCHE is a prototype for an agent based secure electronic commerce marketplace environment. The processes are coordinated by the use of autonomous, self-interested software agents, representing consumers and small businesses. The marketplace itself offers a directory service, but never actively influences the agents' activities. The agents communicate by passing messages bilaterally using an iterated contract-net protocol. Businesses who try to sell and buy goods first configure a (mobile) agent in their web browser or PDA and send it to any electronic marketplace to negotiate autonomously. In case that the agent's demand is not met, it can move to other marketplaces in search for a better offer or demand situation. If the agent finds a promising situation, it automatically starts a negotiation process, which will eventually lead to a market transaction. The software agents are able to adapt their changing environment to their negotiation strategies through the use of evolutionary algorithms. Since the agents are conducting their businesses over an open and insecure network, some technical security mechanisms have been implemented. The code-obfuscated agents digitally sign their offers and encrypt their messages. But the self-interested behavior of the agents demands additional mechanisms, which exceed technical security, to ensure co-operational behavior. Currently a reputation tracking mechanism is implemented that evaluates a partner's former transaction behavior and thus influences both partner selection and negotiation strategy for future transactions. In case that an agent plans to conduct a transaction with an unknown agent, it may be able to source the reputation information from either the marketplace or other known agents, depending on the system of information collection, storage, and distribution. This rating of reputation is then updated during the settlement phase of any transaction.