Speculative Abductive Reasoning for Hierarchical Agent Systems

Answer sharing is a key element in multi-agent systems as it allows agents to collaborate towards achieving a global goal. However exogenous knowledge of the world can influence each agent's local computation, and communication channels may introduce delays, creating multiple partial answers at different times. Agent's answers may, therefore, be incomplete and revisable, giving rise to the concept of speculative reasoning, which provides a framework for managing multiple revisable answers within the context of multi-agent systems. This paper extends existing work on speculative reasoning by introducing a new abductive framework to hierarchical speculative reasoning. This allows speculative reasoning in the presence of both negation and constraints, enables agents to receive conditional answers and to continue their local reasoning using default answers, thus increasing the parallelism of agents collaboration. The paper describes the framework and its operational model, illustrates the main features with an example and states soundness and completeness results.