Yoonjeong Cha, Jincheul Jang, Younghyun Hong, and Mun Yong Yi. 2019. "Jack-of-All-Trades": A Thematic Analysis of Conversational Agents in Multi-Device Collaboration Contexts. In Extended Abstracts of the 2019 CHI Conference on Human Factors in Computing Systems (CHI EA '19). Association for Computing Machinery, New York, NY, USA, Paper LBW0271, 1–6. https://doi.org/10.1145/3290607.3313045
CHI'19 LBW Abstract
A growing number of conversational agents are being embedded into larger systems such as smart homes. However, little attention has been paid to the user interactions with conversational agents in the multi-device collaboration context (MDCC), where a multiple number of devices are connected to accomplish a common mission. The objective of this study is to identify the roles of conversational agents in the MDCC. Toward this goal, we conducted semi-structured interviews with nine participants who are heavy users of smart speakers connected with home IoT devices. We collected 107 rules (usage instances) and asked benefits and limitations of using those rules. Our thematic analysis has found that, while the smart speakers perform the role of voice controller in the single device context, their role extended to automation hub, reporter, and companion in the MDCC. Based on the findings, we provide design implications for smart speakers in the MDCC.
Recent years have seen growing interest in conversational agents (CAs) in the HCI research community. Conversational agents are now not only integrated into smartphones but also embedded into larger systems such as smart homes, as a ‘smart speaker.’
As an increasing number of devices and services are connected in homes, users are facing new challenges in understanding the complex system and learning how to command tasks. With the increasing adoption of conversational agents in smart homes, they can provide opportunities to mediate interactions between users and multiple devices.
While previous studies provided valuable insight into the user interaction with conversational agents, little attention has been paid to the research on the user interaction with conversational agents in the multi-device collaboration context (MDCC).
As more conversational agents are being embedded into the systems where multiple devices and services connect to each other, it is important to identify the roles of such agents in this particular context. Given more smart speakers and connecting devices become available, the MDCC is very likely to be a dominant future direction of smart speakers.
Therefore, the goal of this study is to identify the roles of conversational agents in smart homes, where multiple devices and services work together. As an exploratory study, our research questions are:
1) Are the roles of conversational agents different in the multiple-device context compared to the single-device context?
2) If so, how do users interact with them differently in terms of their roles?
To find answers to these questions, we conducted semi-structured interviews with the participants (n=9) who are heavy users of smart speakers in the MDCC. A total of 107 rules (usage instances) of users were collected and the situations and motivations behind each rule were asked to deepen the understanding of the usage.
Procedure
To discover the role of smart speakers in the MDCC and to find out the in-depth motivations of using them, we conducted semi-structured interviews with each of the participants. First, we asked them to write down the ‘programming rules’ of smart speakers that they were currently using in ‘If condition, then action’ format. That is, if certain condition of devices or services is triggered, then certain action happens in other devices and services. Then we asked follow-up questions to describe each situation more in detail and to explain the benefits and limitations of using those rules. The average length of each interview was 2 hours, and all the interviews were audio-recorded and later transcribed for further analysis.
Data Analysis
We performed the qualitative content analysis on our participants’ responses for their current usages of smart home devices connected to smart speakers. First, we divided the collected rules into two groups based on whether the smart speaker collaborates with more than one device or service to distinguish single and multiple device contexts. Then the purposes of the rules were also categorized to group the functions of various devices and services. Lastly, by examining the rules and interview data, the themes of the role of the smart speakers were derived with affinity diagramming. Two authors collaboratively categorized all the rules. This was performed with repeated iterations until consensus was reached, and the final themes were derived.
As a result, we derived four themes of smart speaker roles: automation hub, voice controller, reporter, and companion. Based on the findings, we draw practical implications on the future design and research directions of smart speakers. Table 2 shows an example of the affinity diagram derived from the analysis.
In the single-device context, smart speakers were most frequently used as a voice controller (86.4%), to control the connected device(s) or service(s) with voice. Also, smart speakers were used infrequently as a reporter (11.4%) and as a companion (2.3%).
In multi-device contexts, smart speakers were used most frequently as an automation hub (39.7%). It refers to the case where smart speakers connect multiple devices or services to make them work automatically in order to accomplish a common mission without any human involvement. In the MDCC, smart speakers were also a reporter (23.8%), notifying the users with information on the current state of the devices or services. Further, smart speakers were a companion of users (19.0%), supporting the users socially with greetings or personal/family care. Lastly, voice controller was the least role performed by smart speakers in the MDCC (17.5%), in contrast to its dominant role found in the single connection context.
In sum, compared with the single connect context, the thematic analysis highlights distinct roles of smart speakers depending on its connection context (voice controller in the single device connection context; automation hub, reporter, and companion in the MDCC).