There are two main types of social science experiments that can, in principle, be conducted in virtual worlds. First, with access to necessary logs of an existing virtual world, one could conduct natural experiments, observing e.g. the effects of the introduction of a new feature. Second, with the necessary influence in some virtual world, one could set up and conduct controlled experiments, that is, tweak some property or rule of the world and observe how it affects the behavior of the participants.
The Arden project, headed by Edward Castronova, has the ambitious goal of building a virtual world for the purpose of conducting the latter kind of experiments. A version of Arden is now available for testing and downloading.
A virtual world specifically designed for experimentation is relying heavily on voluntary agents. The same goes, naturally, for any virtual world. Setting technical difficulties aside, giving the users incentives to participate is probably one of the more difficult tasks of any virtual world developer. The first version of Arden is, according to its developers, not very successful in this aspect.
The tone of the above-linked post on Arden is, however, perhaps more positive than the previous one just a couple of months ago. A more game-like version of the virtual world and some experimentation results are promised for the next year – though probably not yet the ones that “revolutionize social science”.