Raciocínio derrotável distribuído baseado em argumentação para ambientes abertos e dinâmicos

Abstract

This work presents a distributed and non-monotonic reasoning model, in which distributed entities – agents – may have knowledge that, taken together, lead to conflicting conclusions. Such conflicts are resolved through a rule-based structured argumentation approach with an argument strength calculation function that takes into account the degree of trust agents have in each other and the degree of certainty about the correspondence between heterogeneous knowledge acquired from different agents. It is also assumed that the environment in which these agents are situated is open and dynamic, só that agents can enter and leave the environment at any time, as well as have their beliefs constantly updated. This generates the need for agents to be able to query arbitrary agents about specific knowledge, even without knowing a priori whether the queried agents have such knowledge or not. It is also considered the possibility that an agent has knowledge related to its current focus – that is, knowledge related to its current situation, location, activity, object(s) of interest and/or objectives – that may be relevant só that other agents can reason correctly about a query sent by that agent. Thus, it is proposed that a query sent from one agent to another allows the former to share focus knowledge that is relevant in the scope of the query. Therefore, a rule-based distributed structured argumentation model that formalizes reasoning structures and argumentation semantics is proposed, as well as a distributed query processing algorithm based on this model. Alternative formulas for argument strength calculation based on different intuitions about possible mental attitudes of an agent and that take advantage of argumentation structures in different ways are also presented. This thesis also opens up possibilities for future work, such as in the scope of belief revision and generation of explanations based on the argumentation structures constructed by agents.