CA paper No.3
- Title & Reference:
Q. Vera Liao, Matthew Davis, Werner Geyer, Michael Muller, and N. Sadat Shami. 2016. What Can You Do?: Studying Social-Agent Orientation and Agent Proactive Interactions with an Agent for Employees. In Proceedings of the 2016 ACM Conference on Designing Interactive Systems (DIS ‘16). ACM, New York, NY, USA, 264-275. DOI: https://doi.org/10.1145/2901790.2901842
- Memo:
Problem
human-like features and social dialogues. it is unclear whether this is a good match for professionals who seek productivity instead of leisurely use. (social dialog와 human-like feature가 professional한 목적으로 사용될때도 좋은건가? )
Objects of research
- focused on studying individual differences in social-agent orientation, defined as the preference for humanized social interactions with an agent interface, such as having natural conversations and social dialogues.
- explored how the individual differences led to differences in user requirements for agent system design and user behaviors that can be used to infer such orientation
- examined how users reacted to agent proactive interaction
- new feature self-introductions / crowdsourcing user questions / social messages
What they did
- conducted a 17-day field study of a prototype of a personal AI agent that helps employees find work-related information. Using log data, surveys, and interviews. (Text-based Agent. Female Face, female persona)
Findings
-
some users appear to be more social, with more relational conversations, politeness, attention to the robot, and self-disclosure.
-
such tendency can be predicted by whether the user greets the robot at the first encounter (Agent를 처음 만났을때 인사를 하느냐 하지 않느냐로 예측가능함)
- the strongest predictor of user opinions is the system performance
- users with high social-agent orientation tended to have more positive opinions of the system
- high social agent orientation were also more tolerant of negative system performance
- perceived interruption of agent proactive interactions was associated with less positive opinions
- General experience with conversational agents was associated with more critical opinions. (Agent 사용 경험, 나이등이 영향을 줌)
-
individual differences in social-agent orientation had significant impact on user opinions
-
users with high social-agent orientation did not judge the system solely by its utilitarian value, but also appreciated its sociality designs.
-
participants who perceived the proactive interactions to be interruptive had less positive opinions of the system.
-
perceived interruption was a stronger predictor of user opinions than perceived friendliness of the agent (★★★★★. interruption하지 않는것이 중요. 특히 업무환경에서는 더 그럼.)
- interruption could be a more crucial design consideration than agent personality
-
They made index and measures: socializing index, politeness index, agent-grounding index
-
found that users with high social-agent orientation could be identified from behavioral signals, including asking socializing questions, being polite, and engaging less in asking agent-grounding questions intended to retrieve desirable answers
-
Interruption 관련해서, frequency of proactive message랑, Message를 보내는 대상이 Agent든, 사람이든 상관없이, message오는 것을 방해받는 것으로 인식하는 개인의 일반적인 경향 (★★★ )
-
바쁜 스케줄과 Social Contact 많은 Professional들은 Agent proactive interactions을 더 싫어함
- 단순히 Message 빈도나, Contents 내용이나 길이의 문제가 아님.
- associated with the general user aversion(혐오) to unsolicited messages at work.
- Proactive Message가 이용자에게 explicit한 communication value를 생성할 수 있다면, 덜 interruption 하다고 느낄 것이다.
-
proactive agent interactions may harming their overall user experience (★★★★★)
-
High Social-Agent Orientation User Requirements:
- improving conversations: improving the capability of handling continuity and turn-taking, awareness of user status, having more variations in answers
- present a personality: providing more subjective and opinionated answers (단순히 객관적 정보 전달이 아닌), presenting relational behaviors, memorizing previous conversations, and initiating meaningful proactive interactions
-
Low Social-agent orientation User Preference : people with low social-agent orientation preferred more machinelike solutions
- including features from conventional information-search tools (정보검색기능): providing desired information with less required user input, providing context aware recommendation
- improve the transparency and affordance of the system
-
communicating what user action is possible in design (★★★★★. What agent can do)
-
carefully embed the explanation and instruction on how to use the system into the dialogue
-
developing personalized agent interfaces(★★★★★). For example, we envision a customizable agent that allows users to choose the levels of social attributes
-
user attribute awareness by identifying user groups that are more or less susceptible to interruption, potentially based on demographics or social status. (Interruption에 덜 민감한 그룹을 찾아내는 것 필요)
- We suggest not only context awareness but also user-attribute awareness for personalizing agent proactive interactions.
Leave a Comment