New Language Constructs and Inferences for the Knowledge Base Paradigm: A Business and Multi-agent Perspective

In Artificial Intelligence, the scientific field of Knowledge Representation and Reasoning (KRR) is concerned with developing formal languages, to represent knowledge, and inference methods to solve tasks using that knowledge. Most of the existing approaches develop a formal language (a logic) together with an inference, to solve some type of computational task. The recently proposed Knowledge Base (KB) paradigm applies a strict separation of concerns to information and problem solving, based on the idea that knowledge is completely independent of the computational task it is used for. A KB system allows information to be stored in a knowledge base, and provides a range of inference methods, all using the same knowledge base. With these inference methods, various types of problems and tasks can be solved using the same knowledge base. In this text we study this paradigm in detail and we prove the hypothesis that this approach has many advantages over standard declarative paradigms. We use an implementation of a KB system: the IDP system with corresponding language FO(·) to model challenging and interesting applications. These applications show the need to provide extensions: we develop new language constructs and formalize new inferences. In the first part of this work, we study two applications from industry. We first look at interactive product configuration. In these interactive configuration problems, a configuration of interrelated objects under constraints is searched, where the system assists the user in reaching an intended configuration. We show that multiple functionalities in this domain can be achieved by applying different forms of inferences on a formal specification of the configuration domain. To this goal, a set of new, derived inferences are defined. A second application is the car-rental system, a prototypical example from the domain of Business Rules: rule-based systems that are used in industry for knowledge intensive applications. We investigate if all necessary knowledge can be represented in our knowledge Representation language FO(·). We propose an extension of the concept of inductive definitions in the form of a ”new“-operator to model knowledge about situations where a new object is created, for which a good formalism was not available. In the second part, we focus on applying the ideas of the KB paradigm and the KB system IDP in a security context, by modelling Access Control applications. We use a KB system in a distributed way to verify if a certain agent has access, based on the policies of other agents. We develop a generalisation of autoepistemic logic to a distributed setting as a language for these distributed policies. Distributed autoepistemic logic (dAEL) is equipped with tools to express references to the knowledge of other agents. We study generalisations of well-known semantics of autoepistemic logic, using Approximation Fixpoint Theory (AFT). dAEL assumes complete mutual introspection between agents, every agent knows exactly what every other agent knows. This is a reasonable assumption when modelling public statements that an agent makes, but it is not reasonable for every multi-agent application. In the last part of this thesis we present another new logic: COLm , an extension of first order logic with constructs for a knowledge modality KA for each agent, a common knowledge modality C and an only knowing modality OA for each agent. We study how these modalities behave together and develop a new semantic structure that can resolve formulas of this language, without assuming any introspection.

Publication year:2016

Accessibility:Open