Modelling Electronic Organizations

Institutions are established to regulate the interactions between parties that are performing some (business) transaction. One of the main roles of institutions is to inspire trust into the parties that perform the transaction. (see [2] for more details on the roles of institutions). The main focus of this paper is how an electronic organization should be specified on the basis of the abstract patterns given by the institution on which the organization is formed. I.e. how can we define a (formal) relation between the (abstract) norms specified in the regulations of the institution and the concrete rules and procedures of the organization such that the agents operating within the organization will operate according to these norms or can be punished when they are violating the norms. Current approaches of the problem in the agent community either work at a low level (policies and procedures) or at a very high level, formally specifying norms in e.g. deontic logic. The low level approaches allow an easy implementation, but the problem arises when the correctness of the procedures and policies should be checked against the original regulations. High level approaches are closer to the way regulations are made, so verification is easier to be done through e.g. model checkers. However, high level approaches usually use one or several computationally hard logics like deontic logic ([8, 12]). Although it is possible to capture the norms in this way and even give them a certain kind of semantics to reason about the consequences of the norms, this kind of formalization does not yet indicate how the norm should be interpreted within a certain institution. For instance, we can formalize a norm like ”it is forbidden to discriminate on the basis of age” (when determining the best possible recipient for an organ) in deontic logic as F (discriminate(x,y,age)) (stating that it is forbidden to discriminate between x and y on the basis of age). However, the semantics of this formula will get down to something like that the action discriminate(x,y,age) should not