Transcultural Believability in Embodied Agents: A Matter of Consistent Adaptation
Vienna Workshop "Agent Culture", 24-25 August 2001
Fiorella de Rosis
Dipartimento di Informatica
Università di Bari
To be 'believable', an Embodied Agent should give an impression ofconsistency in the way it appears and behaves: the various aspects ofits physical appearance (its face and body and the way it wears)should be consistent among themselves and with its its behavior (theway it talks, gestures and acts in the world). Appearance andbehavior should evolve in time, consistently with changes in theencountered situations; in fact, believability is also a function ofconsistency between appearance, behavior, personality and the contextin which the Agent lives: academics may wear jeans in their dailywork while managers are implicitly obliged to wear more elegantdresses; they both employ an informal communication language intechnical meeetings, a more formal one in policy-making meetings, andso on.
The idea of 'consistency' originates from the respect of acoherent set of norms and standards which are established, eitherformally or informally, in the environment in which the Agent issituated. Culture is an essential component of this environment andplays a crucial role in influencing the Agent's believability:academics in Southern Italy tend to dress less informally than theircolleagues in Northern Italy, they gesture more and speak more aloud,tend to adopt a more 'relaxed' working style and so on. However,adherence to norms and standards is a matter of personality, as well:ambitious people tend to avoid relaxed working styles, even inSouthern Italy.
Obviously, not everything is culture-dependent: somecross-cultural rules, for instance about the personality traitimpressions of the face, have been demonstrated ('attractiveness','facial maturity' and 'gender' effects) and should be considered ininsuring a 'first level' of believability in Embodied Agents.However, adaptation to the context is essential for personality andculture-dependent features, which are -probably- the majority ofthose that have to be considered when building an Embodied Agent.These features are closely interrelated: if their relations are notconsidered appropriately, adaptation therefore risks to result in alack of consistency in the Agent's appearance and behavior.
The questions that I would like to discuss in my presentation arethe following:
- How should adaptation to the cultural context and to personality traits be implemented so as to really insure believability? Are personality traits culture-specific or culture-independent?
- Is independence between the Agent's 'Mind' and its 'Body' a mean to favour adaptability to the mentioned factors?
- Which aspects should be considered, in adapting the Agent's Mind and Body to its personality and to the cultural context?
- How should the Agent's Mind be represented, so as to enable adapting its mental state and its reasoning style to its personality traits and to the cultural context in which it is situated?
- How should the Agent's Body be linked to its Mind and, again, to the context?
- and similars
My proposal will go in the direction of fine-grained cognitivemodels of Agents' Minds in which goals, norms and standard arerepresented explicitly with their relations with the Agent's systemof beliefs. Building on the experience of our research group about 3Dand cartoon-like Embodied Agents, I will discuss how personality andcontext factors may be represented and how they influence the Agent'sinterpretation and reaction to events occurring in the externalcontext. I will discuss, as well, how uncertainty in these relationsmay be taken into account and which models promise to give anacceptable representation of the dynamics of these phenomena.