Developing A Dynamic Testbed: Designing for Trust in Complex Human Agent Teams

There has been a great deal of research examining trust in human-agent teaming. However, the majority of this research has focused on a human-agent dyad with low-interdependency and limited time to study the trajectory of trust. As human-agent teams (HATs) increase in agent capabilities, team size, and heterogeneity, there is a need to enable research paradigms to understand the nuances of trust dynamics in HATs. To support this, we propose that research testbeds should contain four principal dimensions. First, agents in testbeds should be enabled to enact varying levels of autonomy, especially at the highest level. Second, the testbed must support performance of a task that requires true interdependence between the human(s) and the agent(s). Third, the testbed should be able to integrate multiple agents, with varying capabilities and roles, which differ from each other and that of the human. Fourth, testbed scenarios must be long enough to allow sufficient time for (a) trust to be established, (b) trust events (e.g., violation and repair) to be injected multiple times, and (c) the impacts of the events to unfold.