Designing invisible handcuffs : Formal investigations in institutions and organizations for multi-agent systems

Abstract

The work presented in this thesis suggestively moves from the consideration that what makes social notions difficult to grasp lies probably in their intrinsic "invisibility". Like Smith's notorious "invisible hand", the notions of institution and organization are something which just cannot be pointed at by simply looking at the outside world. Still, they are there and they are almost ubiquitous. The prime research question of the work is therefore foundational. If we aim at a science of such things as institutions and organizations, how should we think of them? Or, to carry on the metaphor, how can we make those things somehow "visible"? Interestingly, recent developments in the field of multi-agent systems (MAS) have clearly pointed at the need for an answer to such question. In MAS a number of autonomous pieces of software ---the agents--- interact in order to execute complex tasks. In such systems the crux of the matter is to design agents' interaction in such a way that, on the one hand, agents remain autonomous and, on the other hand, that the system exhibits global desirable properties. In human societies, institutions and organizations have developed as means for pursuing this exact aim. They set invisible boundaries ---"handcuffs" --- for the activities of the individuals in the society. If such "handcuffs" need to be designed in order to coordinate software agents, then a formal theory of institutions and organizations needs to be found which can ground the design process. The work presented conceives of institutions as systems of constitutive rules. Following Searle, constitutive rules are statements of the type "X counts as Y in context C" ---the so-called counts-as statements--- and they underlie the whole construction of institutional reality. It is our thesis that by means of these statements institutional qualifications are imposed on the to-be-regulated domain which provide norms for agents' conduct. Of course, many different institutions coexist which might disagree on the way they look at the same domain. This motivates the formal analysis of the notion of context and the related one of contextual terminology which, from the point of view of the formal machinery deployed in our analysis, underpins the whole work presented here. In a nutshell, institutions can be thought of contextual terminologies, and counts-as statements as their basic building blocks. As to organizations, the thesis focuses on their structural dimension analyzing them as multi-graphs to which properties of transition systems ---expressed in logical formulae---are attached, which specify the impact of the structure on the activities of the agents in the organization. This perspective provides both quantitative methods, based on graph-theory, to compare different organizations from a structural point of view, and qualitative ones, based on logic, to address the types of interaction which different organizations put in place. Finally, on the grounds of these results, a comparison of the two notions of institution and organization is provided, which makes explicit the different aspects that each of these two notions stresses in the conceptualization of social interaction between agents.