This project’s objective is to develop and validate an authoring-friendly framework for the creation of transparent social intelligent agents capable of being deployed as virtual characters in training simulations or games, or used to drive the behavior of social robots.

It intends to join two disparate AI research areas -- Social Agent Architectures and Machine Learning -- in one framework.

By using machine learning techniques over a solid social theory foundation that defines the structure of the learning task, our framework provides a new methodology where authors can create intelligent characters by providing examples of social interaction.

Since the learned behaviour will be built on top of a social representation model, it will be more understandable and transparent to the authors, thus allowing a finer level of control. We will validate our approach in two ways, one with authors, and another with end-users via a case study in social skill training.