Designers of military and civilian systems desire increasingly complex automation to reduce manual labor and costs, extend system capabilities, and improve reliability, persistence, and resilience. Increasingly complex automation increases both the need for, and the difficulty in achieving, accurate trust and reliance decisions by human users. Accurate trust, in turn, demands transparency into the automation behavior and reasoningthe ability to see through and understand what autonomy is doing and why. But there is a fundamental problem in providing that transparency information that the proposed PI has described: the exchange of transparency information almost certainly cannot and should not happen only within the locus of executionthe time and place in which the automation is being used to perform a missionsince this is generally the time of maximal human workload. Instead, this information exchange should be distributed throughout the lifecycle of automation usageto encompass its original design, training, pre-mission planning, post-mission debriefing, etc. The PI's proposed concept, LifeCycle Transparency (LCT), the notion of reducing the need to communicate and process transparency information in high workload periods by shifting transparency information into lower demand periods before and after high demand ones and creating conditions to improve the efficiency of transparency information exchange even in high demand times. This project, called REFLECT (Robust Ecosystem and Framework for LifECycle Transparency) will create a lifecycle-spanning framework within which trust and understanding will develop between humans and automation, embodied in a suite of tools focused at various lifecycle phases, and effectively distributing transparency information and creating conditions for its efficient transfer. Our ecosystem of tools will target the broad domain of complex, automation-supported, planninga domain that spans from very brief time-horizon tactical commanding all the way to strategic planning and policy making for far future contingencies, and in which human teams collaboration before, during and after plan execution is already known to produce transparency benefits even in unplanned-for situations. Our REFLECT framework will be unified through the sharing of hierarchically structured task and goal methodology knowledge in the form of plays and plans that use them, leveraging nearly 20 years of work and tool development (through both SBIRs and many other forms of R&D investment). We have already begun the process of creating HAI structures to expose and explain plan rationales by automated planners within this framework. In REFLECT, we will extend that work, illustrate that it is scalable and feasible, and demonstrate its lifecycle benefits.
