Virtual Pedagogical Agents: Naturalism vs. Stylization

In discussions on naturalism vs. stylization in the design of virtual pedagogical agents (VPAs), the smooth communication argument is one of the most central in favour of visual naturalism. Yet, this argument dissolves if we separate design into the levels of: (l) linguistic performance, where the support for naturalism is strong; (2) dynamic visual appearance, in the sense of bodily behaviour, with increasing design freedom still matching positive user responses; and (3) static visual appearance, as the underlying, inanimate, visual model, with naturalism a well defined state, but stylization spanning a complex, multidimensional design space. A comparison with cartoons and animated movies here suggest a potential for stylized designs. A consequence of not recognizing these different levels of design, is that a (reasonable) strive for naturalism in some aspects ends up in a non-reflected strive for overall naturalism – not realising that a certain degree of naturalism in visual dynamic appearance does not require the same degree of naturalism in visual static appearance, and that, in particular, naturalism in linguistic behaviour is practically independent of naturalism in visual static appearance. The identification argument in favour of naturalism regards VPAs’ potential to take advantage of human social affordances. Close emulating of human beings will enable us to recreate phenomena that are pedagogically valuable in the human–human context, such as emotional support, identification and role modelling. When a student experiences similarity with a role model, the efficiency of the role model increases – therefore preconditions for role modelling will be superior if a character looks really like a human being. This line of reasoning may have intuitive appeal, but is not supported by empirical evidence. On the contrary, it seems even easier for people to experience social relationships, and identify themselves, with stylized characters. Behind conflicting arguments on visual naturalism we find different motives for research and development of VPAs. One is to obtain smooth and beneficial interaction between human and computer in pedagogical contexts. Another is to model and understand human behaviour: human dialogues, gestures, facial expressions, etc. With respect to the latter motive, but not the former, naturalism is a self-evident ultimate goal. Therefore, when researching naturalism–stylization in VPAs, the scientific modelling of naturalistic human behaviour and the pragmatic approach focusing on the development of a usable tool, must be distinguished. Only this way will we be able to benefit fully both from the pedagogical potentials of VPAs – for role modelling, identification, the use of stereotypes, etc. – and from their potentials as powerful test-beds for theoretical modelling.