Engineering visual displays to influence choice in automated decision support systems

Abstract

The task of choosing between decision alternatives presented on a visual display is ubiquitous. Automated decision support systems (DSS) provide a powerful means of improving human decision-making outcomes, but they can also introduce deleterious effects in the form of automation bias (e.g., commission errors and errors of omission). Research has shown that informationally equivalent display designs can lead to significant differences in terms of decision-making outcomes. The current study examined how the influence of visual display design factors on decision making can be leveraged to increase compliance with an automated DSS and reduce potential automation bias. To this end, a series of four experiments were conducted. In each experiment, participants completed a simulated route navigation task in which they were tasked with choosing one of four different routes that were described by four different attributes in order to navigate to their destination. Experiments 1 and 2 were designed to explore how display design factors could be used to establish a decision environment that influenced participants choices in a predictable manner. Results revealed that highlighting an attribute in yellow to increase its perceptual salience increased the likelihood that participants would choose the route that was strongest on the salient attribute. Experiments 3 and 4 applied this salience effect to the design of an automated DSS which recommended one of the four routes to participants. By increasing the salience of an attribute, choice share in favor of the route recommended by the automated DSS increased by as high as 15%. However, this increase in compliance came at the cost of increasing commission errors; participants chose the recommended route even on trials in which it was inferior to other alternatives. For this reason, salience effects should be applied to cases in which the cost of commission errors is low or when automation reliability is high. Informationally equivalent display design factors can be manipulated to increase compliance, but to reduce automation bias, the display design must communicate the logic underlying the automation’s recommendations.