TAC Assist: Team Award Collaborative Assistants
Abstract
Good teams reliably produce good mission outcomes by adapting to their changing situations. Individuals on the team know to adapt by awareness of relevant information and application of that information with an understanding of mission and team context (i.e., team aware decision making). The team must communicate to ensure all team members have sufficient information for a coordinated and coherent effort while managing the cost of collaboration. We propose to develop Collaborative AI- enabled (CAI) assistants to help teams manage interdependence and enable adaptive and resilient team performance. The core of our technical approach is combining joint activity graphs (JAGs) with machine learning (ML). JAGs are a novel approach to representing how teams conduct joint activity. JAGs capture the structure, process and data flows of joint activity, which allows them to be reasoned over to understand interdependencies and interpret relevance of information. This enables them to support the key properties desired in CAI agents (cooperative, coordinated, communicative). JAGs will enable our CAI agents to dynamically determine information relevance to drive effective collaboration and enhance coordination through support for the management of interdependencies. While developing the joint activity model of a mission can be systematic, determining the expected value of decisions within that solution space is complex and heuristic-based. We propose using ML to determine the expected value of different decisions within a graph, as well as the value of information that might be communicated across the team. JAGs provide a unique framework for exploring the action-communication space. This analysis can then be utilized to support a human operator in decision making under different situations. Our CAI agents will collaborate in two ways: supporting individual performance and facilitating team coordination. To support the individual, our CAI agents will use their knowledge to understand and track information relevance. This will be used to ensure individuals attend to essential information and understand important interdependencies. They will assist with planning, identifying underspecified plans, brittle points and conflicts. They will also help track execution to identify errors, conflicting information and bottlenecks. Additionally, our CAI agents will facilitate teamwork and coordination between team members. They will model each individual team member looking for misalignments in common ground and facilitating resolutions. They will track communication and coordination and ensure teams manage their interdependence efficiently. We will use mission planning and execution for disaster response as our test domain. Mission planners can become overwhelmed by available data from real-time data sources making a common operating picture tough to build, maintain, understand and translate into action. Our CAI agents will help assess, curate, process and manage this data. They will also assist the warfighter in interpreting and decision making based on this data. In Phase 1, we will have a two- person mission planning team and use an event-based simulation. For Phase 2, we will replace our simulated data with human generated data by having a larger human team conduct the activity directed by the mission planners and those new human team members will be responsible for generating the data and reporting to facilitate mission management. We will evaluate teams based on team performance and team process measures
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Oct 13, 2023
- Source ID
- N000142312886
Entities
People
- Matt Johnson
Organizations
- Florida Institute for Human and Machine Cognition
- Office of Naval Research
- United States Navy