Underlying our approach is the view of large markets as ecologic systems [5], in which the interaction of the participants determines the success of the individual participant. Interesting results can be drawn from game theoretic approaches to the study of ecologic systems and applied to open electronic market systems (or large markets in general) [5]. Non-cooperative games (games with non-binding commitments) can be used to study the cooperation between self-interested participants[7].
The interesting idea of market-oriented programming and how to create design economies in the WALRAS system is described by Wellman in [11]. In their model the market is a tool for resource allocation, and it is argued that every computational problem can be transformed into one of resource allocation [12]. The agents' only means of communication is the trade of goods, in an protocol of iterated revealing of preference functions. The model does not include unintended malicious collaboration between subgroups of agents since it doesn't permit side conversations and cooperation. (To assure convergence, the resource usage and utility sets cannot be sub-additive.)
Instead of using market-oriented programming to do resource allocation, we consider a market that is an open environment, possibly supporting actors with malicious intentions and goals. Our interest is to study if forms of social control (of which market economics is one) can make it possible to perform computations even in such environments.
The idea of letting agents themselves answer for the security of the system is inspired by the idea of mechanism design. Rosenschein and Zlotkin [10] are using a game theoretic approach for the design of agent communication protocols, assuming that an agent will act rationally.
Soft security mechanisms for intrusion detection have been tried by Crosbie and Spafford [2]. They audit program actions and by using a number of sensors that alert if the behaviour of a program seems unusual they are trying to find intruders who are executing programs not normally executed by the users. (Similar methods are being used by credit card companies who try to discover fraud by searching for card owners who drastically changes their buying behaviour.)
One attempt to detect actors who don't behave as expected is being made by A. Rasmusson[8]. A personal security assistant collects information about the actions made by an actor and is trying to find patterns of abnormal behaviour in actors. This could be used to build a very fine grained reputation system that lets other actors know of abnormal actors.
___________________________________________________
Lars Rasmusson