Effects of transparency on pilot trust and agreement in the autonomous constrained flight planner

Abstract

We performed a human-in-the-loop study to explore the role of transparency in engendering trust and reliance within highly automated systems. Specifically, we examined how transparency impacts trust in and reliance upon the Autonomous Constrained Flight Planner (ACFP), a critical automated system being developed as part of NASA's Reduced Crew Operations (RCO) Concept. The ACFP is designed to provide an enhanced ground operator, termed a super dispatcher, with recommended diversions for aircraft when their primary destinations are unavailable. In the current study, 12 commercial transport rated pilots who played the role of super dispatchers were given six time-pressured "all land" scenarios where they needed to use the ACFP to determine diversions for multiple aircraft. Two factors were manipulated. The primary factor was level of transparency. In low transparency scenarios the pilots were given a recommended airport and runway, plus basic information about the weather conditions, the aircraft types, and the airport and runway characteristics at that and other airports. In moderate transparency scenarios the pilots were also given a risk evaluation for the recommended airport, and for the other airports if they requested it. In the high transparency scenario additional information including the reasoning for the risk evaluations was made available to the pilots. The secondary factor was level of risk, either high or low. For high-risk aircraft, all potential diversions were rated as highly risky, with the ACFP giving the best option for a bad situation. For low-risk aircraft the ACFP found only low-risk options for the pilot. Both subjective and objective measures were collected, including rated trust, whether the pilots checked the validity of the automation recommendation, and whether the pilots eventually flew to the recommended diversion airport. Key results show that: 1) Pilots' trust increased with higher levels of transparency, 2) Pilots were more likely to verify ACFP's recommendations with low levels of transparency and when risk was high, 3) Pilots were more likely to explore other options from the ACFP in low transparency conditions and when risk was high, and 4) Pilots' decision to accept or reject ACFP's recommendations increased as a function of the transparency in the explanation. The finding that higher levels of transparency was coupled with higher levels of trust, a lower need to verify other options, and higher levels of agreement with ACFP recommendations, confirms the importance of transparency in aiding reliance on automated recommendations. Additional analyses of qualitative data gathered from subjects through surveys and during debriefing interviews also provided the basis for new design recommendations for the ACFP.

A Longitudinal Field Study of Auto-GCAS Acceptance and Trust: First-Year Results and Implications

“…Lyons (2013) identified displays/interfaces to be one of the two means to promote system transparency (i.e., shared awareness and shared intent between the user and automation). There is now emerging research on how to build in transparency (via interfaces) for highly autonomous safety systems (such as Auto-GCAS) in a way that supports operator appropriate trust development (Chen & Barnes, 2014;Lyons, Sadler, et al, 2016;Sadler et al, 2016).…”

Trusting Autonomous Security Robots: The Role of Reliability and Stated Social Intent

“…Other research has defined agent transparency in terms of understanding the agent's goals, actions, and reasoning, and projecting future states of the agent (Chen et al, 2018). Research (mostly in the domain of human-automation/agent interaction) has found that transparency in the forms of state awareness and projection (Chen et al, 2018;Ho et al, 2017;Mercado et al, 2016), decision rationale Sadler et al, 2016), and perceived benevolent design (Ho et al, 2017; are associated with higher trust and/or better humanagent interactions.…”

Influence of Culture, Transparency, Trust, and Degree of Automation on Automation Use

“…Transparency at the situation assessment stage provides teleological (why) explanations which may be incomplete, simply conveying relevant features contributing to an automation decision [29] or supplying the logic behind the decision [30] as well. Teleological explanations are both preferred by humans [31] and conform to our premise that endto-end transparency will lead to fuller prediction and more accurate decisions.…”