Supporting User Fusion of AI Services through Conversational Explanations

As systems become more complex, they become harder to understand both in terms of their composition (by system developers) and in terms of their operation (by system users). In dynamic systems the services may have been composed in real-time without human involvement, to support a specific user request. It is therefore even more important that the ability to understand the composition, the data sources and the component services is possible. There may also be restrictions relating to what each user is allowed to know about different aspects of the system, services or datasets, and all of the elements of the system may be widely varied and volatile, with the whole system subject to rapid and pervasive change. In this paper we define a conversational mechanism to enable interaction with such a system. Able to handle volatility and variety through the simple conversational interface, this proposed capability enables the explanation of the system and the results at a number of levels. We map our basic architecture to the JDL data fusion model (at all levels) and view this particular conversational explanation capability as belonging in level 5 (user refinement).