Title: Preventing the misuse and disuse of automated systems: Effects of system confidence display on trust and decision performance
Margaux Price
Dissertation Defense
4:00 pm, Friday, October 5th, Room 419 Brackett Hall
Dr. Richard Pak (advisor)
Dr. Leo Gugerty (committee member), Dr. Chris Pagano (committee member), and Dr. Joel Greenstein (committee member)
Abstract:
Complex decision-making may be aided by forms of automation known as decision-support systems (DSS). However, no DSS is completely reliable and so it is imperative that users know when they should and should not trust it (calibration of trust). Previous research has shown that providing users with information about the DSS’s confidence in its own advice (“system confidence”) can help improve the calibration of user’s trust of automation and actual system reliability on a trial by trial basis. The current study examined how the nature of the presentation of system confidence information affected user’s trust calibration. The first study examined the attentional demand of each display, while the second study examined their effect on trust and performance on a decision making task. The results of this study indicate that there was no effect of system confidence display type on subjective or objective trust. The lack of differences in performance or trust between the control condition (no system confidence display) and other displays raises doubts about whether users were utilizing the system confidence information or simply solving the problem on their own. The type of decision task may be crucial in determining whether to provide system confidence and these results suggest that it should be tested prior to implementation against a control group, unlike previous studies. The results of these studies have implications in the design of DSS, especially given the difficulty of providing accurate system confidence information to users. The time and resources that would be required to provide such a display may not be beneficial if it has no effect on user trust or decision performance. However, the use of an anthropomorphic display may be an ideal way to present decision support information because overall calibration between expected system reliability and users’ behavioral trust was more accurate.
