GEMINI: Gestalt Mixed-Initiative Intelligence

ABSTRACT: SIFT proposes to build GEMINI (Gestalt Mixed-Initiative Intelligence), a shared workspace where humans and machines collaboratively assemble intelligence products. In a human-machine team, the collaborators have vastly different-- albeit complementary-- cognitive capabilities. Thus, a central research challenge is to expressively and soundly translate from analyst-centric interfaces (for human understanding and high-level fusion) into structured knowledge and goals (for goal-directed autonomy) and back again for intuitive support. The analyst will use an intuitive user interface (UI) to fuse information and define objectives. GEMINI will decode the UI's elements into structured information and decomposable goals, and then assign goals to specialized agents-- such as graph-matchers, document retrievers, information extractors, and image processors-- to augment the analysts workflow. It will scan agent-generated information for relevance to the analysts objectives and focus, and then present relevant information within intuitive displays using Adaptive Interface Management (AIM) technology. Having pioneered work on AIM, goal- and task-based autonomy, information extraction, and data fusion, SIFT is well-positioned to build and transition GEMINI to customers that need it. By facilitating the bi-directional human-machine information flow and leveraging both parties' innate strengths, GEMINI will enable human-machine teams to identify threats and insights faster and more accurately than ever before.; BENEFIT: By facilitating the bi-directional human-machine information flow and leveraging both parties' innate strengths, GEMINI will enable human-machine teams to identify threats and insights faster and more accurately than ever before.GEMINI's integration of goal-directed autonomy in an analyst-centric fusion setting will yield synergistic benefits for the human and for the autonomous agents: the analyst receives mission-relevant information in intuitive interfaces to support high-level information fusion; the agents receive the analyst's hypotheses, context, objectives, and high-level fused information to support goal-directed automation.GEMINI consequently has the commercially-viable feature of reducing the time and technical expertise that an organization requires to remain competitive.The extensibility of GEMINI's analyst interface-- including displays for monitoring and control-- and of GEMINI's autonomy manager-- including a goal- and knowledge-based interface for agent metacontrol-- will reduce the cost and effort of extending GEMINI to additional customers with novel information needs.This facilitates integration of diverse autonomous agents toward common objectives, as we will demonstrate in our Phase I effort.Ultimately, GEMINI's extensibility will help it generalize to strategic intelligence, tactical intelligence, and other human information fusion settings.