Bringing social reality to multiagent and service architectures : practical reductions for monitoring of deontic-logic and constitutive norms
As distributed systems grow in complexity, the interactions among individuals (agents, services) of such systems become increasingly more complex and therefore more difficult to constrain and monitor. We propose to view such systems as socio-technical systems, in which organisational and institution...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Estado: | Versión publicada |
| Fecha de publicación: | 2016 |
| País: | España |
| Institución: | CBUC, CESCA |
| Repositorio: | TDR. Tesis Doctorales en Red |
| OAI Identifier: | oai:www.tdx.cat:10803/398246 |
| Acceso en línea: | http://hdl.handle.net/10803/398246 https://dx.doi.org/10.5821/dissertation-2117-98124 |
| Access Level: | acceso abierto |
| Palabra clave: | Àrees temàtiques de la UPC::Informàtica 004 |
| Sumario: | As distributed systems grow in complexity, the interactions among individuals (agents, services) of such systems become increasingly more complex and therefore more difficult to constrain and monitor. We propose to view such systems as socio-technical systems, in which organisational and institutional concepts, such as norms, can be applied to improve not only control on the components but also their autonomy by the definition of soft rather than hard constraints. Norms can be described as rules that guide the behavior of individual agents pertaining to groups that abide to them, either by explicit or implicit support. The study of norms, and regulatory systems in general, in their many forms -e.g. social norms, conventions, laws, regulations- has been of interest since the beginning of philosophy, but has seen a lot of evolution during the 20th century due to the progress in the philosophy of language, especially concerning speech acts and deontic logic. Although there is a myriad of definitions and related terminologies about the concept of norm, and as such there are many perspectives on how to analyse their impact, a common denominator is that norms constrain the behaviour of groups of agents in a way that each individual agent can build, with a fair degree of confidence, expectations on how each of their counterparts will behave in the situations that the norms are meant to cover. For example, on a road each driver expects everybody else to drive on only one side of the road (right or left, depending on the country). Therefore, normative contexts, usually wrapped in the form of institutions, are effective mechanisms to ensure the stability of a complex system such as an organisation, a society, or even of electronic systems. The latter has been an object of interest in the field of Artificial Intelligence, and it has been seen as a paradigm of coordination among electronic agents either in multi-agent systems or in service-oriented architectures. In order to apply norms to electronic systems, research has come up with abstractions of normative systems. In some cases these abstractions are based on regimented systems with flexible definitions of the notion of norm, in order to include meanings of the concept with a coarse-grained level of logic formality such as conventions. Other approaches, on the other hand, propose the use of deontic logic for describing, from a more theoretical perspective, norm-governed interaction environments. In both cases, the purpose is to enable the monitoring and enforcement of norms on systems that include -although not limited to- electronic agents. In the present dissertation we will focus on the latter type, focusing on preserving the deontic aspect of norms. Monitoring in norm-governed systems requires making agents aware of: 1) what their normative context is, i.e. which obligations, permissions and prohibitions are applicable to each of them and how they are updated and triggered; and 2) what their current normative status is, i.e. which norms are active, and in what instances they are being fullfilled or violated, in order words, what their social -institutional- reality is. The current challenge is on designing systems that allow computational components to infer both the normative context and social reality in real-time, based on a theoretical formalism that makes such inferences sound and correct from a philosophical perspective. In the scope of multi-agent systems, many are the approaches proposed and implemented that full these requirements up to this date. However, the literature is still lacking a proposal that is suited to the current state-of-the-art in service-oriented architectures, more focused nowadays on automatically scalable, polyglot amalgams of lightweight services with extremely simple communication and coordination mechanisms- a trend that is being called “microservices”. This dissertation tackles this issue, by 1) studying what properties we can infer from distributed systems that allow us to treat them as part of a socio-technical system, and 2) analysing which mechanisms we can provide to distributed systems so that they can properly act as socio-technical systems. The main product of the thesis is therefore a collection of computational elements required for formally grounded and real-time e¬fficient understanding and monitoring of normative contexts, more specially: 1. An ontology of events to properly model the inputs from the external world and convert them into brute facts or institutional events; 2. A lightweight language for norms, suitable for its use in distributed systems; 3. An especially tailored formalism for the detection of social reality, based on and reducible to deontic logic with support for constitutive norms; 4. A reduction of such formalism to production rule systems; and 5. One or more implementations of this reduction, proven to e¬fficiently work on several scenarios. This document presents the related work, the rationale and the design/implementation of each one of these elements. By combining them, we are able to present novel, relevant work that enables the application of normative reasoning mechanisms in realworld systems in the form of a practical reasoner. Of special relevance is the fact that the work presented in this dissertation simplifies, while preserving formal soundness, theoretically complex forms of reasoning. Nonetheless, the use of production systems as the implementation-level materialisation of normative monitoring allows our work to be applied in any language and/or platform available, either in the form of rule engines, ECA rules or even if-then-else patterns. The work presented has been tested and successfully used in a wide range of domains and actual applications. The thesis also describes how our mechanisms have been applied to practical use cases based on their integration into distributed eldercare management and to commercial games. |
|---|