The current market model is far from optimal. It is unclear how many aspects of real markets it captures. For now there is no notion of advertising, the actors can't choose where to move, so even the ``favourite'' and ``room reputation'' algorithms are choosing somewhat randomly. The production of goods is strictly additive, that is, there is no way that sellers can bundle different goods together using co-production advantages to lower the price since there is only one product. Discussions with people more familiar with economic market models will hopefully help to remedy these, and other, shortcomings.
If bankruptcy is to be used as a means for regulating the paricipants on the market, then it migth be difficult for real people to interact on the same market as the electronic agents. In a pure electronic systems it's just a way to create adaptive systems, as in the agoric computing of Miller and Drexler or in the computational ecologies of Huberman, but in systems including real people, these mechanisms might be too strong. By limiting the system to only include electronic agents, there are more parameters to modify, hence more possible social control mechanisms that can be examined. We hope to investigate mixed as well as pure systems.
Social control is not something that is obviously just good. If actors are excluded, rightly or wrongly, from the market, chances are that it is fatal to them. We continue to look further in to what kinds of problems can be caused by these mechanisms.
By using reputation agents, an actor looses control over what information he/she transmits to the other actors. This is all left to the reputation agent. Perhaps the actors can better assure that their information is not lost somewhere if they themselves are the ones responsible for sending out the information. Still, this returns us to the problem of collecting and refining the enormous amounts of information that will be available.
Human markets will with extended global connectivity change to be open systems with strongly interacting components through the use of telecommunication, network and agent technologies. If such large systems are ever to be put in practical, everyday use, we must be sure that they cannot run amok.
In open systems it might be impossible to know and to be sure of the detailed behaviour of the components in the system. Furthermore, since components can be added and removed as time goes, the global properties of an open system can change rapidly.
The more the parts in such a system are permitted to interact, the more unpredictable becomes the result of their collective actions. Systems dealing with tasks in our everyday life must be assuredly stable, in the sense that adding components to the system can't open it up to criminal behaviour, permitting new ways to safely fraud people.
In open, Internet based, markets it won't be possible to enforce or assure a certain behaviour of all the actors. By simulating markets with interacting agents with evolutionary stable strategies along with some means to promote rational behaviour in agents (reward for good work and cost of execution), it seems possible to find strategies giving the agent society the emergent property to (without any global control) expel malicious agents.
Socially controlled systems are ecologic systems. Ecologic simulations of agents using different rules for choosing collaboration partners will be used to measure different social methods' efficiency in detecting malicious agents. Since hard security fails in open systems, we must understand how to use these mechanism if we want to create secure open systems for Internet commerce.
___________________________________________________
Lars Rasmusson