Designing for Mutually Beneficial Decision Making in Human-Agent Teaming

Abstract

This paper presents a joint decision-making framework between human and artificial intelligent agents in an effort to create a cohesive team uninhibited by each other’s actions. Based on the well-known Recognition Primed Decision-Making Model, our framework expands upon RPD’s single decision maker to be more Human-Agent Teaming (HAT) oriented. Specifically, our framework includes three layers of shared cognition to ensure both a consistent level of transparency between members and the efficient completion of the task. The first layer provides itself as a foundation of expectations that provides familiarity recognition in a situation. The second layer categorizes the environmental features into relevant decisions informing the symbiotic nature of who should and how to enact decisions collaboratively, which is the third layer. Altogether, this mutually beneficial decision-making model emphasizes transparency so that both humans and artificial agents are equal partners in completing tasks in unique situations.

Balancing the Scales of Explainable and Transparent AI Agents within Human-Agent Teams

“…With this pairing of humans and AI teammates harmoniously balancing each other in their talents, researchers have investigated methods to improve the relationship between these types of teammates. One such factor is in building a comprehensible platform of communication that is bi-directional for both humans and AI (Mallick, Sawant, McNeese, & Chalil Madathil, 2022). In further detail, humans need to understand how AI makes a decision and how they are able to communicate that information, as well as the AI needs to understand how and why the humans are making the decisions they do so they can learn from it and gain a better understanding of the environment.…”

Can We Build it? Yes, We Can! Development Procedure of High-Fidelity Simulation Environments for Human-Agent Teams

“…Teammates allow themselves to repetitively simulate the appropriate teamwork behaviors that are required of them to work cooperatively with each other as a means to achieve high cohesion with each other. High team cohesion is a significant proponent of effective HATs (McNeese, Schelble, Canonico, & Demir, 2021) as it encompasses the comfort humans and AI have with each other and expresses it through fluid decision-making where each action mutually benefits the other as they learn the task and learn about each other's capabilities (Mallick, Sawant, McNeese, & Chalil Madathil, 2022). In this way, post-trained humans and AI understand each other's distinct responsibilities as it relates to the overall task and what aid would look like should one entity need it.…”