Experimental work on anthropomorphic feedback in user interfaces has shown inconsistent results and researchers offer differing opinions as to the potential usefulness of this style of user interaction. A review of the literature shows that experimental work can be improved and enhanced by taking into account issues that characterise human-human communications. Results from three experiments are reported that exhibit the previously observed inconsistencies but this is arguably a function of task context. An alternative explanation is that the results are a reflection of the cognitive nature of tasks. Overall, the results point the way to further and future results in terms of refining procedures but also in terms of theoretical focus.
Human-computer interaction (HCI) focuses on the dialogue between users and computers through the user interface (UI). The past 20 years have seen UI developments that range from command line interfaces through to modern graphical user interfaces (GUIs) based on the now familiar desktop metaphor. Regardless of the type of UI, the feedback provided to users is an important area of concern and remains an active area of research. Feedback to users as part of human computer dialogue can take a number of forms such as text, graphics, animation, speech, and so forth. Anthropomorphic interfaces are a way of delivering feedback that in some way takes on the characteristics or role of a human. Typically, anthropomorphism involves a nonhuman entity, usually some element of the user interface, taking on some human quality (De Angeli, Johnson & Coventry, 2001), for example a talking dog or a cube with a face that can talk. A well known example is the Microsoft Office Paper Clip. It could also be the actual manifestation of a real human, such as a video of a human (Bengtsson, Burgoon, Cederberg, Bonito & Lundeberg, 1999).
It can be argued that from a common sense point of view anthropomorphic interfaces ought to be of some benefit to users as these in some way or form mimic ‘natural human communication or characteristics.’ However, the usefulness (however defined) and desirability of anthropomorphic interfaces remains a moot issue. While some researchers provide evidence and argument in favour of using anthropomorphic interfaces (e.g., Agarwal, 1999; Cole, Massaro, Rundle, Shobaki, Wouters, Cohen, et al., 1999; Dertouzos, 1999; Guttag, 1999; Koda & Maes, 1996a, 1996b; Maes, 1994; Zue, 1999), others have cast doubt on this style of interaction (e.g., Shneiderman 1997; Shneiderman & Plaisant, 2005).
The work reported and discussed in this chapter aims to review previous work showing that results have been inconsistent and there is not agreement as to the potentially positive or negative influences of anthropomorphic UIs. It will be argued that experiments need to take into account context and Grice’s maxims as outlined by Reeves and Nass (1996). Results from experiments will be presented that cast some light on effectiveness of anthropomorphic interfaces in relation to context and the potential role of Grice’s maxims.
As part of developing user support and increased usability, Microsoft has experimented with anthropomorphic feedback using work by Nass, Steuer, and Tauber (1994) as a foundation. This work formed part of the Persona Project where an anthropomorphic character, a parrot called Peedy, is used to help a user find and play music tracks. A user can interact with the system using automatic speech recognition (ASR) (e.g., Bradshaw, 1997).
Support for the effectiveness of anthropomorphic feedback can, for example, be found in the work of Maes (1994) and coworkers who use the term ‘personification’ to characterise this type of feedback.
Maes (1994) has developed an agent for e-mail handling that is described in the context of agents as ‘learning’ in the sense of becoming an effective personal assistant. This agent takes the form of a drawn face where various facial expressions are used to convey a particular state of the agent, for example ‘working’ and so forth. This ‘facial agent’ can be applied in various other contexts (e.g., scheduling meetings). Experimentation with this type of agent showed that users produced positive comments and seemed to indicate that users judged the input of the agent to be helpful. This can be taken as an indication of positive approval of anthropomorphic feedback.
Further support for anthropomorphic feedback is provided by Koda and Maes (1996a, 1996b) who conducted experiments in which users played poker using the Web. The aims were to understand: