Trust in Automation: Transparency for Enhancing the Reliability of a Health Assistant Chatbot

#human-computerinteraction #artificialintelligence #trustandsafety #interpretability

May, 2019 | Christopher Nguyen

Read Paper

Abstract

The topic under purview was agent transparency and conversational user interfaces. Agent transparency refers to a graphical user interfaces ability to support a human-agents understanding of how an intelligent agent thinks as well as the underlying reasons behind its behavior. In line with past research studies on agent transparency, this paper further explored ways for building trust and developing greater awareness in intelligent systems. The experimental design was a one-way ANOVA with trust and awareness as the dependent variables and transparency levels as the independent variable. Thirty participants were tested in this study and were classified under either non-reliant (n= 25) or reliant group (n=5). Participants placed in the non-reliant group were given a bot that provided them with an invalid health assessment. While, participants placed in the reliant group were given a bot that provided a valid health assessment. Group segmentation was based on participants’ complacency and trust in automation scores prior to the start of their session. The data indicated that there was a statistically significant interaction between transparency levels and the non-reliant participants awareness (F (2,48) =124.625, p <.05). The mean scores for the awareness levels revealed that as transparency increased, awareness decreased. And for this reason, participants were inappropriately relying on the automation either by under or over trusting the automation. Additionally, for the non-reliant participants, there was a statistically significant difference between SAT levels and participants trust levels (F (2,56) =3.562, p <.05). Moreover, the significance of these findings suggests that human workload as well as trust were affected by the amount and type of information that was communicated by the automation.