Virtual Pedagogical Agents - Beyond the Constraints of the Computational Approach

Abstract: Virtual pedagogical agents (VPAs), i.e. computer generated characters in pedagogical roles, populate the digital society in increasing numbers. They are found in educational programs from preschool to university. In a broader educational context, they function as virtual medical counsellors, physical exercise coaches and guides on city web pages – and they also appear in edutainment and infotainment settings. The main reason for adding virtual (pedagogical) agents to educational software is their potential to motivate and engage. In order to develop and exploit this potential, there are several issues that need to be resolved. This thesis (based on three papers) begins with an overview of research on motivation and engagement in VPAs. The outcome of this overview is that visual aspects of VPAs in much are neglected. The thesis continues with an argumentation on the importance of visual aspects in VPAs, followed by a presentation of two empirical studies. The two studies examine possible correlations between visual and social/ communicative characteristics in VPAs and social/communicative styles in users. Taken together, there is strong support for the claim that users’ visual/aesthetic experience of VPAs is too important with respect to the goals to motivate and engage, to be treated as a secondary issue. Thereafter follows a survey on visual properties of VPAs in relation to visual stereotypes and novel possibilities as well as risks are discussed. The conclusions of this thesis are: (i) that it is highly probable that visual aspects in VPAs can be related to pedagogical outcomes; (ii) that there may be significant relations between visual and social characteristics of VPAs and social/communicative styles of learners; and (iii) that VPAs may reproduce stereotypes from everyday real life human-human interaction, as well as from traditional visual media – something that, in combination with the increased interactive potential of VPAs, may even strengthen certain detrimental effects of idealized stereotypical instances. The research agenda proposes a further exploration of visually related aspects of VPAs – not least in order to handle the outcomes of user evaluations that otherwise will continue to be blurred by the impact of uncontrolled visually related variables. Furthermore, the research agenda argues that there is a problem with user evaluations based upon the concept of ‘the User’, i.e. the averaged standardized user. The reason for this is that users may diverge into categories with almost diametrically opposite qualities as to, for example, personality, communicative style and learning style. In order to reveal important correlations between agent characteristics and user characteristics, it is therefore necessary to identify such groups of users. The research agenda also points at the need for ethical and legal research regarding VPAs, as there now is a golden opportunity to attempt to address such questions in advance. And finally – virtual agents may have a not yet thought of potential as research tools.