Symbolic representations and common - sense reasoning in open multi - agent systems

Abstract

This thesis concerns with intelligent autonomous software agents that populate open computational environments, in which they interact for various purposes, e.g. competitively in the case of auctions or resource allocation problems, collaboratively in the case of distributed problem solving or parallel processing, joint planning, etc. We use the term open to characterize a computational environment in Hewitt's sense, that is to describe an environment that is dynamic, continuous, unobservable (or, at best, partially observable) and non-deterministic.Agents in such environments possess, unavoidably, information that is incomplete, imprecise, maybe even incorrect, due to the very fact that the environment is open and, at the very least, agents join and leave it as they please. The information exchanged between agents may be delayed, or distorted by noise during its communication, and in any case, as the environment evolves, it is bound to change.The interactions among agents in any multi-agent system are typically governed by norms. These may refer to restrictions on communication means among agents, or to particular coordination mechanisms, liveness and safety properties of the system etc. In some application areas, such as ecommerce, additional norms may regulate the agents' behaviours, resulting from agreements into which the agents enter willingly, and possibly from the protocol that governs the e-market. Norms prescribe what each agent is obliged, permitted, prohibited, empowered and so on, to do during its life in the particular environment. Autonomous agents decide for themselves which norms to subject itself to and whether to comply with the norms of their environment. This decision-making is all the more challenging when an agent has to perform it in circumstances where its available knowledge is incomplete/imprecise/incorrect. This thesis addresses the need and requirement for common-sense reasoning agents in open computational environments. We discuss and illustrate our proposals with reference to an e-commerce example. First, based on the example scenario, we identify the requirements for the representation of the norms that govern the environment and the specifications for the environment itself. Nevertheless, our proposals are generally applicable to any case where multiple agents interact and their interaction is governed by some contract as a coordination or collaboration mechanism.Primarily, this thesis examines and motivates the need of agents to fill in information gaps by resorting to assumptions. Agents need to be able to identify and use assumptions dynamically, in any open computational environment, as well as in the particular application context of an e-commerce example. We present a novel approach to dynamic assumption identification and hypothetical nonmonotonic reasoning inspired by the syntax and semantics of Default Logic, without however resorting to proof, which is notably computationally hard. We discuss in detail what distinguishes our approach from other work on dynamic assumption...