Knowledge sharing among agents via uniform interpolation

In our age of big data and the knowledge society, effective processing and sharing of knowledge is crucial. Agents provide a key abstraction within modern development of artificial intelligence systems and software. In current multi-agent platforms, a simplifying assumption is that all agents use the same symbols to represent their knowledge or expertise. The assumption that agents use the same symbols to represent their knowledge imposes a constraint on system developers and agent systems: agents are forced to use the same ontological representation of the world at the cost of having diverse and unique viewpoints. Furthermore, agents typically exist in dynamic environments that require update and revision of their knowledge and beliefs. We focus on agents that make use of description logic ontologies to represent their knowledge and expertise. Ontologies are knowledge bases consisting of logical statements called axioms. This comes with the advantage that a set of logical axioms can entail knowledge that is not explicitly stated by the axioms: lots of information can be implicit. As a consequence, when communicating, agents must take into account the implicit knowledge contained in their expertise. Assuming a common signature is established between two communicating agents, the agents still require methods to extract specific knowledge from their ontologies that go beyond sending a list of axioms. Agents require methods to extract both explicit and implicit knowledge from their ontologies in a way that can be communicated to another agent if the need arises. This thesis re-uses several existing techniques in the logic literature, mainly uniform interpolation, that can be used to extract knowledge from ontologies and can be adapted for agent communication. In particular, our aim is to develop and evaluate novel algorithms which provide support for scenarios where multiple agents are responsible for different knowledge bases (e.g., that capture the agent's different expertise) and have the ability to restrict (specifically on the basis of uniform interpolation) and adapt their knowledge with respect to the signature of their knowledge base that is shared with other agents. This ensures that knowledge shared is understood by the communication partners. We have realised and developed a range of knowledge extraction methods and structures that enable agents to extract knowledge that can sometimes be implicit with respect to arbitrary subset signatures of their knowledge base. We have analysed and evaluated these knowledge extraction methods on ALC description logic ontologies. Our results suggest that the knowledge extraction methods realised are feasible in some practical settings; unlike the knowledge extraction methods realised, the knowledge extraction methods developed are rarely feasible and useful in practical settings, and will require further exploration and research to be useful in practical settings.