The effects of automated decision aid reliability and algorithm modality on reported trust and task performance.

摘要

As IT systems grow more complex and become more prevalent, understanding the collaborative nature of the relationship between humans and automation becomes more important. Several factors influence the human-automation relationship. Research has shown that trust and perceived reliability are key factors in whether a relationship will develop between humans and automation. Presenting automation reliability and automation algorithms are ways to potentially improve this relationship. To explore this question, an experiment was conducted in which an automated decision aid presented suggestions to participants while they managed a simulated city (i.e., Policity). The goal was to maximize the health of the city's population by adding hospitals, housing, businesses, and other facilities and services. Participants were assigned to conditions where the automated decision aid performed with high or low reliability levels. Based on condition, the decision aid's algorithm was presented to participants in a textual or graphical (diagrammatic) format. Results showed that users' perception of the decision aid's reliability directly influenced their trust in the decision aid. Results also showed that presenting the decision aid's algorithm, regardless of modality (i.e., textual or graphical) had a direct impact on reported trust. Both had a direct effect on human performance. Additional results and implications are discussed.