Surprising findings showing that many basic rules of social interaction among humans also apply to the interaction of humans with computers²⁶ have encouraged interface researchers to proclaim a new paradigm in human computer interface design: the social interface. Reeves and Nass' findings have especially given a new impetus to research on anthropomorphic agents, suggesting that furnishing an agent with a human-like interface could be an effective way to enhance agent usability. Little research exists however, to verify this claim. This study seeks to illuminate user responses to an anthropomorphic agent if it is used on a daily basis. It examines the influence of traditional usability variables such as control, understanding, trust, and distraction on the acceptance of an anthropomorphic interface. Kaasinen⁸ as well as Wexelblat and Maes³⁶ emphasize that the acceptance of agents (whether or not they have an anthropomorphic interface) strongly depends on these traditional usability variables.

As subject for this study, a commonly used anthropomorphic agent, the Microsoft Office Assistant, was selected. Using a questionnaire that was distributed through email and the World Wide Web as well as on paper, 105 users of the Microsoft Office Suite were asked about their use and acceptance of the Office Assistant. It turned out that the Office Assistant feature was rejected by the majority of the users. As a feature to troubleshoot problems, it was perceived as inefficient. Also, from a "social" perspective, it was rated as not particularly engaging or motivating. Multiple regression analysis revealed that variables influencing the use and the acceptance of the Office Assistant were distraction and trust. Only those users who felt distracted by the assistant and who did not trust in its capability to solve their problems rejected the Office Assistant. It is concluded that to ensure agent usability, its design should emphasize assistance with the user's task over social mimicry. If design efforts to furnish the agent with social traits interfere with the user's principal goal, it is likely that the agent will be rejected.