Human-agent cooperation to support resilience.

摘要

Advancements in automation, including increasing machine autonomy, are changing people's relationships to automation, and moving machines into our more unpredictable human world. Such developments have important implications for the resilience of joint human-automation systems. When humans and machine agents actively coordinate joint goals in complex and unpredictable environments, cooperation is required. While previous research in human-automation interaction focuses on how perceptions of performance influence reliance on or compliance with automation, this research explores social exchange factors influencing human-automation cooperation and system resilience. Specifically, it considers how different social exchange structures and different levels of an automated agent's cooperation influence joint coordination in a dynamic task environment. A microworld was developed and the study was conducted in two parts. Part 1 tests whether different levels of agent cooperation affect human cooperation, given unexpected changes in the task environment, in a negotiated exchange structure. Part 2 also tests levels of agent cooperation, but in a reciprocal exchange structure, to evaluate if reciprocal exchange led to timelier and greater exchange of resources. Results show that the reciprocal exchange structure increased flow of staff resources compared to negotiated exchange, leading to higher joint scores. Participants' cooperation also differed depending on the level of agent cooperation. In negotiated exchange, participants provided more resources to a high-cooperation agent compared to a low-cooperation agent, and in reciprocal exchange, participants provided timelier resources to the high-cooperation agent compared to a low-cooperation agent. This work departs from the typical focus on supervisory control automation and automation performance in terms of reliability, and suggests cooperative control automation and automation performance in terms of collegiality is an important area of research for human-automation interaction that enhances system resilience.