decision making with multiple imperfect decision makers - Institute of ...

An Adversarial Risk Analysis Model for anEmotional Based Decision AgentJavier G. RázuriDepartment of Statistics and Operations Research & AISoy RoboticsUniversidad Rey Juan CarlosMadridjavierfrancisco.guerrero.razuri@urjc.esPablo G. EstebanDepartment of Statistics and Operations Research & AISoy RoboticsUniversidad Rey Juan CarlosMadridpablo.gomez.esteban@urjc.comDavid Ríos InsuaRoyal Academy of Sciences, Spaindavid.rios@urjc.esAbstractWe introduce a model that describes the decision making process of an autonomoussynthetic agent which interacts with another agent and is influencedby affective mechanisms. This model would reproduce patterns similar to humansand regulate the behavior of agents providing them with some kind of emotionalintelligence and improving interaction experience. We sketch the implementationof our model with an edutainment robot.1 IntroductionWe have recently introduced in [19] the framework of Adversarial Risk Analysis (ARA) to cope withrisk analysis of situations in which risks stem from deliberate actions of intelligent adversaries. ARAhas a Bayesian game theoretic flavor, as in [11] and [17]. In supporting one of the participants, theproblem is viewed as a decision analytic one, but principled procedures which employ the adversarialstructure, and other information available, are used to assess the probabilities on the opponents’actions. There is a potentially infinite analysis of nested decision models arrived at when usingARA. This is in the realm of incomplete Bayesian games, see [8], which avoids the infinite regressby using the common (prior) knowledge assumption. We feel that this is a very strong hypothesiswhich is not tenable in our application domain. We prefer to be realistic and accommodate as muchinformation as we can from intelligence into our analysis, through a structure of nested decisionmodels. Depending on the level we climb up in such hierarchy of nested models, we talk about0-level analysis, 1-level analysis and so on, see [1] and the discussion [10]. [1], [18] and [19] haveintroduced different principles to end up the hierarchy. In this paper, we shall explore how the ARAframework may support the decision making of an autonomous emotional agent in its interactionwith a user.Over the last several years, researchers in the field of cognitive processes have shown that emotionshave a direct impact on judgment and decision-making tasks. This has vertebrated fields such asaffective computing [16], affective decision making ([13] and [4]) and neuroeconomics [6]. Based1