Team Design Patterns for participatory development of First Response Human-Agent Teaming

Open Access
Article
Conference Proceedings
Authors: Tatjana BeukerTina MiochMark A. Neerincx

Abstract: First responders (FRs) work in complex and dangerous environments in which information is often uncertain and incomplete. Advancements in sensor and artificial intelligence technology pose great potential for supporting FRs to stay resilient throughout dynamic mission developments. To avoid creating higher workload or requiring more expertise, a recommended strategy is to use a human-centered design approach that regards the human and machine as team members (Neerincx et al., 2016). For the ASSISTANCE project, we develop a module that will offer FRs support during incidents involving hazardous substances, displaying information about current and predicted gas cloud distributions using constantly updated input from meteorological services, chemical sensors, and FRs. In this research we investigated how teamwork between FRs and the module should be designed to facilitate adequate decision-making. For this purpose, Team Design Patterns (TDPs) were recently proposed as they describe roles and responsibilities within a team in an and reusable manner (Van Diggelen & Johnson, 2019). An additional benefit is the possibility to actively involve different stakeholders in the design process, incorporating relevant expert knowledge in the learning and reasoning of the AI-agent (van Stijn et al, 2021). We created three TDPs, each assigning different roles and responsibilities to the AI-agent and FRs, and assessed these with FRs in an online survey. In the first pattern, FRs collaborate with an ‘Informing Agent’ that is responsible for keeping FRs updated about the current and predicted situation. The second pattern describes the cooperation with an ‘Advising Agent’, which additionally gives mission-relevant recommendations. In the third TDP, FRs collaborate with the ‘Deciding Agent’, which also can make decisions regarding actions and carry them out. The collaboration with the first two agents was evaluated by presenting two scenarios to the FRs in which they had to handle an incident in simulated collaboration with either the ‘Informing Agent’ or the ‘Advising Agent’. The ‘Deciding Agent’ was presented at the end and FRs were asked to indicate their acceptance and preference. The results showed that preference and acceptance varied across FRs and different decisions. For example, when deciding where to allocate measurement teams, hazmat officers, who are experts in this task, tended to prefer an AI-agent with less responsibility whereas on scene commanders with less expertise for this task tended to prefer the AI-agent to take more responsibility. Additionally, comments given by the FRs indicated that trust and sufficient explanation of the underlying decision model were important factors that influenced willingness to collaborate with the AI-agent. In conclusion, this paper shows how TDPs can be applied to systematically involve end-users in the Human-AI system design process. The results point towards recommending a design solution in which the AI-agent can adapt its collaboration style to different FRs and decisions it is assisting with. Next steps include defining and investigating how to design such an adaptive cooperative system, taking into account ethical aspects such as trust of the FRs in the information and recommendations of the system and transparency of the decision model. ReferencesNeerincx, M. A., van Diggelen, J., & van Breda, L. (2016). Interaction design patterns for adaptive human-agent-robot teamwork in high-risk domains. In International conference on engineering psychology and cognitive ergonomics (pp. 211-220). Springer, Cham.Van Diggelen, J., & Johnson, M. (2019). Team design patterns. HAI 2019 - Proceedings of the 7th International Conference on Human-Agent Interaction, 118–126.Van Stijn, J.J., Neerincx, M.A., ten Teije, A.T., Vethman, S. Team Design Patterns for Moral Decisions in Hybrid Intelligent Systems: A Case Study of Bias Mitigation, AAAI-MAKE 2021 Spring Symposium.

Keywords: Team Design Patterns, Human-Agent Teaming, Human-Machine Systems

DOI: 10.54941/ahfe1001151

Cite this paper:

Downloads
12
Visits
34
Download