Institutions (also called normative frameworks) provide an effective mechanism to govern agents in open intelligent systems. An institution specifies a set of norms, with respect to specific normative objectives, that regulate agents’ behaviours in terms of permissions, empowerments and obligations. However, in most real circumstances, several institutionsprobably have to cooperate to govern the same entities simultaneously, which is referred as cooperating institutions in this dissertation. Depending on how individual institutions are connected with each other, three different ways of forming a cooperating institution are addressed: coordinated institutions, interacting institutions and merged institutions. The dissertation firstly presents a formal and computational model for all three types ofcombination. Furthermore, when agent behaviour is regulated by a cooperating institution, consisting of a set of independently-designed institutions, normative conflicts are likely to arise, as each individual institution has its own objective. For instance, a certain action may be permitted (or obliged) by a norm from one institution while being prohibited by a norm fromanother institution. A blunt solution is to ignore or delete the conflicting norm(s) from one or the other institution. A further contribution of this dissertation is however the development of a formally justified fine-grained approach operating on parts of norms that is able to: (i) detect normative conflicts automatically for all the three variants of cooperatinginstitutions, and (ii) resolve these conflicts by automatically constructing a minimal revision of the conflicting norms through inductive learning. In this work, we start with formalising three types of cooperating institution by means of an institutional action language (InstAL ), which can be automatically translated into logic programs under Answer Set semantics.Based on that, we then put forward an automatic procedure that can identify the normative conflicts that may arise, and transform them into negative examples to feed our conflict resolution system implemented by Inductive Logic Programming, through which the conflict-free cooperating institution can be derived by revision of the norms belonging to specific identified institutions. We further demonstrate the proposed conflicts detection andresolution approach in several case studies from the domain of multi-agent systems and legal systems.
|Date of Award||19 Nov 2014|
|Supervisor||Julian Padget (Supervisor) & Marina De Vos (Supervisor)|