Department of Neuroscience and Biomedical Engineering, Aalto University
About the Primacy of Social Interaction in Human Brain Function
We humans are embedded in social interaction with other people who form a central part of our environment: we think about them, attend to them, interact with them, and interpret the nonverbal and verbal social cues they convey. Still much of the current neuroscience, even “social neuroscience”, examines brain functions of isolated persons who are observing other persons’ actions from third-person view. We have been advocating the idea that the research on the brain basis of social cognition and interaction should move from this kind of “spectator science” to studies of engaged participants. Recent advances in neuroimaging now in fact allow to study the brain basis of social interaction by simultaneous recordings of brain activity of two participants engaged in natural communication. One important research question is whether social interaction is emerging from lower-level perceptual, motor, and cognitive functions, as is typically assumed, or whether it could be the default mode governing e.g. perception and action. Such primacy of social interaction would challenge many current ideas about human brain function.
USC Institute of Creative Technologies
The role of a lifetime: Dialogue Models for Virtual Human Role-players
Some dialogue systems are designed to be instruments to allow a user to efficiently solve a task, using natural language. For these systems, human dialogue related to this task acts as a 'proof of concept’, but might not be the desirable end-state, e.g. if the different knowledge and abilities of a computer might lead to differences in interaction style being more efficient or effective. On the other hand, just as humans sometimes act as role-players in learning exercises, games, or other activities, some dialogue systems also act as role-players in these activities. For these cases, I argue, human-like dialogue is more important, especially if the purpose of the dialogue activity is to learn or practice interactive skills that should transfer to other humans. However, for role-play dialogue, the most appropriate metric is not the familiar “Turing test” of indistinguishability with humans, but rather activation of the same socio-cognitive skills that are appropriate for human interaction.
In this talk, I will present examples of role-play dialogue systems from a wide variety of activities, genres, and roles, focusing on virtual humans created at the Institute for Creative Technologies. I will examine different dialogue models that are appropriate for different kinds of roles, depending on the appropriate interactions for the role and activity. I will also go over human-oriented evaluation metrics for these systems, and show that dialogue with artificial systems can lead to very human results.