Operationalized Intent for Improving Coordination in Human-Agent Teams
Abstract
With the increasing capabilities of artificial intelligent agents (AIAs) integrated into multi-agent systems, future concepts include human-agent teams (HATs) in which the members perform fluidly as a coordinated team. Research on coordination mechanisms in HATs is largely focused on AIAs providing information to humans to coordinate better (i.e. coordination from the AIA to the human). We focus on the compliment where AIAs can understand the operator to better synchronize with the operator (i.e. from the human to the AIA). This research focuses specifically on AIA estimation of operator intent. We established the Operationalized Intent framework which captures intent in a manner relevant to operators and AIAs. The core of operationalized intent is a quality goal hierarchy and an execution constraint list. Designing a quality goal hierarchy entails understanding the domain, the operators, and the AIAs. By extending established cognitive systems engineering analyses we developed a method to define the quality goals and capture the situations that influence their prioritization. Through a synthesis of mental model evaluation techniques, we defined and executed a process for designing human studies of intent. This human-in-the-loop study produced a corpus of data which was demonstrated the feasibility of estimating operationalized intent.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2020
- Accession Number
- AD1113947
Entities
People
- Michael Schneider
Organizations
- Air Force Institute of Technology