Grant: Alliance Grants

Autonomous systems face a major sociotechnical integration challenge when they must collaborate with human operators and that challenge is hindering their effectiveness and use. Specifically, human-AI agent teamwork comes with cognitive costs and new skill requirements for both human and artificial agents. Collaboration overhead costs need to be minimized, along with improving shared understanding and mutual adaptation. We propose to help overcome those capability gaps through human-AI co-learning of teamwork and taskwork. Our main hypothesis is that to be effective, human-autonomy teaming (HAT) systems must not only focus on human and AI-based counterparts learning how to perform the required taskwork, but they must also learn how to engage together in the required teamwork processes for effective mission management and adaptation. An adaptive command and control (C2) process with adjustable human-autonomy collaboration is therefore proposed herein. This framework will be instantiated and tested in simulated human-in-the-loop maritime operations scenarios where a human supervisory controller manages several unmanned vehicles to perform tasks in challenging environments requiring agility. Experiments using a synthetic high-fidelity testbed will objectively and comprehensively measure the impacts of augmenting the HAT system with the envisioned co-learning, shared understanding, and adaptive C2 capabilities. The key goals of this partnership are thus to 1) contribute to the acquisition of knowledge for the design of effective HAT systems; 2) offer new applied cognitive science perspectives on human and AI-agent collaboration; and 3) iteratively test successive versions of a proof-of-concept capability to de-risk uncertainties and inform design of each subsequent iteration.