SPACE-RELEVANT HUMAN-AUTONOMY TEAMING TASK TO STUDY THE MULTI-DIMENSIONAL AND DYNAMIC NATURE OF TRUST.

Leary, Sarah; Jaekeun Sung; Hurd, Victoria; Lee, Christian; Yimin Qin; Zhaodan Kong; Clark, Torin; Anderso, Allison

INTRODUCTION: Human-autonomy teaming is an integral component of civilian and DoD missions. A human must appropriately trust the autonomous system (AS) to collaborate effectively. Historically, trust has been obtrusively obtained via surveys. Because of this limitation, trust is often modeled as static, rather than dynamic, and one-dimensional, rather than having nuance. For example, Cognitive Trust (CT) forms due to rational, logical thinking, whereas Affective Trust (AT) is based on feelings and emotions. CT and AT are two-dimensions of trust. Our goal is to affect CT and AT in participants and obtain survey-based "ground-truth" trust measurements over time, while simultaneously collecting unobtrusive psychophysiological (e.g., skin conductance responses), neurophysiological (e.g., oxygenated hemoglobin), and embedded measures (e.g., button clicks). We can then use these unobtrusive measures as predictors and develop metrics and models that can infer and predict "ground-truth" trust in real-time. METHODS: We developed a space-relevant, humanautonomy teaming task to study multi-dimensional trust. The task was developed using PyQT and displayed on a 2D screen. Participants act as the "supervisor" of their simulated AS teammate in a ground troop monitoring task. The AS receives simulated data captured from ground imaging satellites (e.g., visual and thermal images) and classifies the data as containing troops or not containing troops. The AS then relays the classification and data to the human. The human has the option of agreeing, disagreeing, or ignoring the recommendation of the AS. Throughout the experiment, subjects are compelled to report their trust using a pop-up slider which serves as the "ground-truth". RESULTS: To affect CT, we varied AS "reliability". Reliability is the number of correct classifications the AS makes and is 65\% in the "low reliability" cases or 85\% in the "high reliability" cases. To affect AT, we varied AS "explainability". Explainability is how the AS justifies its classification to the human. "Low explainability" is robotic-like rhetoric, whereas "high explainability" is human-like rhetoric. Our results indicate subject trust in the AS varied with reliability and over time. DISCUSSION: This task is specifically designed to affect multi- dimensional trust when working with an AS. Future work will build models from psychophysiological, neurophysiological, and embedded measures for real-time trust inference and prediction. Learning Objectives 1. Biosignals and embedded measures provide an unobtrusive method of inferring trust. 2. Reliability is known to affect Cognitive Trust. This experiment aims to use robotic or human-like rhetoric to affect Affective Trust as well, which is less studied.

TRUST; GALVANIC skin response; CRITICAL thinking; UNOBTRUSIVE measures; THERMOGRAPHY

Aerospace Medicine & Human Performance, 2024, Vol 95, Issue 8, p448

2375-6314

Academic Journal