Accountable agents and where to find them

In the field of multi-agent systems individual and organizational actions have social consequences, which require the development of tools to trace and evaluate principals' behaviors and to communicate good conduct. This concerns the value of accountability. The aim of this thesis is to present an Accountability-Driven Organization Programming Technique (ADOPT) that attempts to face the challenges of handling accountability in a computational way. The main contribution is to provide a notion of when an organization of agents supports accountability by investigating the process of construction of the organization itself as well as to give the definition of a protocol to be followed in order to design and build an accountability-supporting organization. The core of the analysis is the notion of role and the action of role enactment (or adoption). By defining an organizational role using social commitments it is possible to reason about accountability when an unexpected outcome is detected, namely an organizational goal is not achieved. Actually, with social commitments, the interaction is represented in terms of social relationships among the agents, which provide expectations of the agents' behaviors. The first part of the thesis provides a survey of the state of the art both on accountability in software systems and on multi-agent organizations. The second part begins by showing how a lack of accountability could be source of problems by means of an example developed using the JaCaMo platform. Then, a few principles to be respected in order to build an accountability-supporting organization are outlined and, finally, the ADOPT accountability protocol, where these principles find their application, is presented.