Self-Organizing Communicating Semantic Objects for Augmented Intelligence : A Study of Fine-Grained Reasoning for QA Agents

Intelligent question and answer agents are becoming popular on consumer devices, e.g. smartphones, but their performance can be less than satisfactory when presented with vague questions. To understand the fine nuances of a user's information need, we propose the use of augmented intelligence through reasoning with semantic objects, which can communicate among each other and self-organize to form knowledge structures in a semantic workspace. Self-organization allows for dynamic multi-granularity of reasoning about real-world concepts at multiple levels of abstraction in response to user's information need. Preliminary evaluation using an iterative adaptation of discounted cumulative gain measure with a prepared domain-specific questions bank, as well as dynamically defined domain-related questions tests, has been conducted to demonstrate both effectiveness and efficiency of the proposed approach.