28 Agosto 2017, 15:20 - Fátima Sampaio
Candidate: Joana Carvalho Filipe de Campos Nº 55848/D
Title: Modelling Interpersonal Conflict in Multi-Agent Societies
Location: Anfiteatro PA3 (Piso -1, Pavilhão de Matemática) IST, Alameda
Advisor: Professora Ana Maria Severino de Almeida e Paiva
Abstract: Conflict and conflict dynamics are phenomena intertwined with social change. The absence of conflict only brings stagnation opposed to progress and improvement. Therefore, being able to cope with conflicts and detect them beforehand are social skills very important in life. Throughout our lives, better or worse, we learn how to cope with conflict by experience. Yet, questions such as “how do we know we are in a conflict?” or “what makes some conflicts emerge while others subside?” are difficult to answer unequivocally. In Artificial Intelligence, conflicts also abound in multi-agent systems. Depending on the needs of the systems, different mechanisms (e.g. joint intentions or enforcement of norms) have been applied to cope with what researchers often call “failures”. Yet, as systems become more complex, we can no longer assume that agents will strive for the overall well-being of the system. Thus, a less simplistic view of conflict is necessary, so that agents are able to detect and anticipate conflicts.
In this thesis, we aim to address the problem of representing the interpersonal conflict construct in a cognitive agent architecture. To that end, we propose a model that is based on sound theoretical concepts and it allows to make conflict explicit. However, to create a model that is more flexible and expressive, we believe that theory and real deep data must be combined. More often than not, the models described in the literature are focused on the effects and use high level features as predictors of conflict handling modes. Little emphasis has been given to actual process and to how it unfolds, information that is critical to create agents that detect and react to conflicts in an appropriate way.
In this thesis to capture the subtleties of conflict and its dynamics, we collected data from real usersby the means of cultural probes and the analysis of interactions of children playing a negotiation game. In the latter, we created a dataset that encompasses eleven dyadic interactions of 10- to 12-year-old children playing a mixed-motive – the Game-Of-Nines. Our aim was to trace the conflict process in conflict-prone situations and therefore we sought to observe and trace cognitive processes by analyzing social signals over time. Those signals include gaze, electrodermal activity and the exchange of offers in this negotiation game. Such analysis allowed us to detail our initial model of conflict, which was based only on theoretical concepts, by introducing mechanisms that create a dynamic system that depends on the changing dynamics of the interaction and whether the agent frames the situation as a conflict or not.