Life-Like Agent Design Based on Social Interaction

Abstract

This study focuses on the design of a life-like agent interface that considers the social aspects in human-agent interaction. We have designed and incorporated life-like guide agents into an exhibition guide system on the hypothesis that a user of the system establishes a social relationship with a life-like interface agent similar to a relationship with another human.

We studied the effect of inducing the users’ affiliation need by incorporating life-like guide agents in the C-MAP (Context-aware Mobile Assistant Project) Exhibition Guidance System. This system features a personal mobile assistant that provides visitors touring exhibitions with information based on contexts (spatial/temporal locations and individual interests). The user of this system carries a hand-held guidance system called PalmGuide while touring an exhibition. PalmGuide maintains user contexts. A life-like personal guide agent runs on the user’s PalmGuide and provides tour navigation information such as introduction of exhibit articles and recommendations of what to visit next. Information terminals called Information Kiosks are installed at each exhibit. Information Kiosks usually provide general information about the exhibit. When a user with a PalmGuide comes to an Information Kiosk, the user can connect her PalmGuide to the Information Kiosk by infrared communication. Then, the use r’s guide agent migrates to the Information Kiosk and personalizes it for her personal interests and needs.

We designed the behaviors of the guide agents as follows: Until the fourth access to an Information Kiosk, the user’s guide agent returns to the original PalmGuide, after it provides relevant information of the exhibit and the recommendation for the next visit. However, at the end of the user’s fourth access, the guide agent does not return to the original PalmGuide. Instead, the agent disappears from the current Information Kiosk by saying that it will go to an Information Kiosk and will be waiting for the user there. The user is expected to chase her agent by herself. When the user follows the agent to the recommended Kiosk, the agent thanks the user. When, on the other hand, the user comes to a different Kiosk, the agent reappears and gives a complaint to her. In either case, we expect that the agent’s reaction induces its user’s affiliation need and has an effect on the user’s subsequent behaviors. We deployed our guide agent system at our annual Open House exhibition, and observed visitors behaviors toward the agents.

We compared acceptance rates of recommendations before and after each user obtained feedback from her agent to her reactions toward the agent’s recommendation at the fifth access. We found that people, who did not follow her agent’s recommendation at the fifth access, tend to behave differently toward the guide agents’ recommendations before and after the fifth access, when the agents simply went ahead and waited for their users at the recommended Information Kiosk. We believe that the guide agents’ action enhanced the users’ affiliation needs toward the agent. The users try to recover the affinitive relationship with the agent when they did not access the recommended Kiosk and obtained a complaint from the agent in the fifth access. The changes in user behaviors are controlled by the enhancement of affiliation needs, which are effected by social interactions between the users and their agents.