A Context-Centered Architecture for Intelligent Assistant Systems

We propose a conceptual framework for implementing intelligent assistant systems (IASs) that (1) work on experience base instead of knowledge base, and (2) deal with the decision-making process and not the result only. Considering experts' experience instead of the domain knowledge supposes to have a uniform representation of elements of knowledge, reasoning and contexts. We propose Contextual Graphs (CxG) as such a formalism of representation. A contextual graph represents a task realization, and paths in the contextual graph represent the different practices developed by experts for realizing the task in specific contexts. A contextual graph is an (micro-) experience base on which intelligent assistant systems have to work. Thus, IASs present a more elaborated domain independent architecture for (1) managing the experience base, (2) simulating practice development, and (3) explaining the rational behind each practice in terms of contextual elements used in the human expertise developed. This opens challenges of a new type of simulation, namely a CxG-based simulation with real-time management of context and actions. We are developing such an IAS for supporting anatomo-cyto-pathologists (ACPs) that analyze digital image of slides as part of breast cancer diagnosis. Such domain experts make critical decisions based on imperfectly known or complex domains, without possible automation of procedures, context-dependent decision made under temporal pressure, and highly context-dependent situations requiring each a specific expertise. An example will illustrate this application.