Adapting a Human – Automation Trust Scale to an Air Traffic Management Environment

Autor: Joey Mercer, Sarah Hunt, Lynne Martin
Rok vydání: 2014
Předmět:
Zdroj: Proceedings of the Human Factors and Ergonomics Society Annual Meeting. 58:26-30
ISSN: 1071-1813
2169-5067
DOI: 10.1177/1541931214581006
Popis: In response to the demand of future air traffic environments potentially exceeding human operator capabilities, the process of integrating automated decision-making tools into the air traffic management system is underway. However, the current system is lacking a validated standard for measuring a critical element of the human-automation partnership trust. Specifically needed is a valid scale appropriate for and tested on air traffic controllers. To remedy this issue, a two-phase modification of the Human Automation Trust Scale by Jian, Bisantz and Drury (2000) was deployed at the Airspace Operations Laboratory at NASA Ames during 2013. Applied to two different human-in-the-loop experiments, the following results include a scale that supports understanding underlying trust attitude in air traffic controllers while maintaining a high inter-item reliability score. The use of this assessment method when testing new air traffic management tools can assist in understanding potential pitfalls for tool use and implementation. INTRODUCTION The next generation of air traffic control in the United States (NextGen) is evolving into an integrated humanautomation environment, with automation generated information becoming a critical contributor to decisionmaking (Joint Planning and Development Office, 2012). As this integration continues, it becomes increasingly necessary to appropriately assess the human-automation relationship as it pertains to decision-making and safety critical tasks. Underlying trust is one component that has been identified as a key (though not sole) contributor to intent to use and actual usage of an automated system (Lee & Moray, 1994). As such, developing a reliable method for measuring underlying trust is necessary in the assessment of human-automation integration. In the current air traffic control domain where most systems are very safe and the goal is near-perfect performance i.e., delivering aircraft safely and on time, the general underlying trust attitude (Lee & See, 2004) of a controller in regards to an automated system is of as much concern as their ability to detect envrionmentally (such as an incorrect weather forecast) induced inaccuracies / unreliabilities requiring a change in automated tool use by a controller (Kirlik,1993). As such, measuring actual general underlying trust of controllers in an automated system necessitates a method which is not highly sensitive to direct experimental manipulation of envrionmental factors impacting the current accuracy of the automation, but instead responds to their underlying attitude, knowledge in and trust of the system. To this end, in 2013, researchers in the Airspace Operations Laboratory (AOL) at NASA Ames employed modified versions of the Human-Automation Trust Scale (HAT) (Jian, Bisantz, & Drury, 2000) during two human-inthe-loop simulations to assess its efficacy in an air traffic management (ATM) environment. HAT was selected specifically for its empirically based assessment method, and its track record of use in different automated domains (Montague, Kleiner, & Winchester 3rd, 2009; Wang, Jamiseon, & Hollands, 2009). Due to the simulations’ highly specific domain –ATM -, HAT was modified to address an air traffic management environment while retaining the same, or synonyms of, key words in the original scale. It was the concern of the experimenters that presenting the scale without modification would be inapplicable to this technology and population. Therefore, the first study conducted in January 2013 retained seven of the original twelve scale items. Study 2 in September 2013 used the findings from Study 1 and presented eight of the original twelve items. The adaptations and subsequent performance of the HAT scale in both experimental environments is the focus of this paper. METHODS A full report on the simulations cited in this paper can be found in Mercer et al. (2013) and Callantine, Hunt, & Prevot (2014) for Study 1 and Study 2, respectively. As Study 1 was conducted first, adaptations made to the HAT scale were further modified before use in Study 2. HAT’s presentation for both experiments involved: (a) a randomized array using an internet-based survey software suite , (b)a seven point Likert style scale, (c) identical scale anchor points – ‘not at all’ [1], ‘moderately’ [4], ‘very’ [4], and (d) the instructions “Please indicate your current thoughts about the automation for the following:”. All participants were retired air traffic controllers, both male and female, and over forty years of age. Experimenters chose two disparate studies of ATM tools in order to evaluate the usefulness of the modified HAT scale. Test Environments Study 1 Test Environment: Study 1’s test environment consisted of two en route sectors (one-high altitude and one low-altitude) run by two parallel teams simultaneously in the lab for four total participants. Based on historical traffic of the Atlanta airspace, traffic scenarios included flows of arrival aircraft feeding into the northwest meter-fix of Atlanta’s Terminal Radar Approach Control (TRACON). While the majority of the aircraft were Atlanta arrivals, the simulation also included several over-flights. The controller teams metered traffic with a delivery goal of +/20 seconds at the meter-fix. Confederate controllers staffed all necessary adjacent airspace. The aircraft simulated in this environment were equipped with Flight Management Systems (FMS) and Automatic Dependent Surveillance-Broadcast -out capabilities (ADS-B-out). Controllers issued all instructions to the pilots via voice communications. Proceedings of the Human Factors and Ergonomics Society 58th Annual Meeting 2014 26
Databáze: OpenAIRE
Externí odkaz: