Angela Li Sin Tan, DSO National Laboratories
Decision automation is designed to support the cognitive functions which were previously performed by human operators. Together, the decision automation and the operator works towards a common goal. The operators can direct his attention to other tasks as they can depend on the automation for the functions which has been automated. However, decision automation can fail and operators need to monitor the automation. The algorithm may also be too complex insofar that the operator cannot understand what the automation is doing.
There is a need to balance the operator’s trust and reliance on the decision automation. Operators would use the decision automation only if they trust the automation. The trust and reliance is built upon (1) operator’s understanding of what the automation is designed for; (2) the automation performance; and (3) how the automation achieves its goal (Lee & See, 2014).
An experiment was conducted to test the effectiveness of the Motivation-Expectation Space in communicating automation strategies. The Motivation-Expectation Space (MES) is an orthogonal representation where means-ends relations are linked vertically, and cause-effects relations are linked horizontally. In a study that examines operator’s thought processes in troubleshooting faults in petrochemical plants (Tan, Helander, Ho, 2007), it was observed that the operators persistently traversed both the means-ends and causal spaces as they explained their solutions. A horizontal causal dimension was thus added to Rasmussen’s means-ends vertical links.
The orthogonal MES was originally designed to represent the operator’s thought processes. Here, it is hypothesized that representing decision strategies on the MES orthogonal structure facilitates the scanning of the information. This, in turn, reduces the time and effort needed to process the information pertaining to how the decision automation generates its solution. The increase in the availability of the decision strategies, like in Oduor’s study (2006), is hypothesized to improve operators’ trust and reliance on the automation. Specifically, the following hypotheses were tested:
1. The decision automation that explains its decision strategies will generate higher subjective rating of trust than the decision automation that provides no explanation.
2. The decision automation that explains its decision strategies will generate higher rates of appropriate use (not following the wrong suggestions given by the decision and not ignoring the correct suggestions) than the decision automation that provides no explanation.
3. It takes less time to interpret when the decision strategies are represented using the MES orthogonal format than when they are plain textual descriptions.
To test these hypotheses, strategies modules in two different formats: graphical MES and textual, were designed for an in-car navigation decision automation.