Abstract | ||
---|---|---|
ABSTRACTConversational agents have been touted for their potential to support individuals over time as health coaches or personal assistants, but have yet to live up to this potential. Wizard-of-oz (WOz) methods enable researchers to test early prototypes of conversational applications before they are fully implemented, with a human “wizard” filling in the gaps in functionality. Current WOz methods, however, are more commonly used for studies in a lab setting, rather than deployment studies, which more accurately capture users’ interactions in-the-wild. We argue for the need for WOz methods for deployment studies that address key challenges, namely the need for easy-to-prototype technology that works reliably in the wild, as well as the user's expectations for 24/7 availability. We describe an initial approach that begins to address these challenges, as well as the insights gleaned from a two-week WOz study of t2.coach, a conversational agent health coach for diabetes self-management. We argue that the findings from our WOz study could not have been identified from a lab usability study with the same prototype, and the need for the research community to further develop methods for WOz deployment studies. |
Year | DOI | Venue |
---|---|---|
2021 | 10.1145/3411763.3443446 | Conference on Human Factors in Computing Systems |
Keywords | DocType | Citations |
wizard-of-oz, conversational agents, chatbots, deployment study | Conference | 0 |
PageRank | References | Authors |
0.34 | 0 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Elliot G. Mitchell | 1 | 4 | 5.88 |
Lena Mamykina | 2 | 1124 | 98.05 |