PETEEI: A PET with Evolving Emotional Intelligence Magy Seif El-Nasr Thomas R. Ioerger John Yen Texas A&M University Texas A&M University Texas A&M University Computer Science Department Computer Science Department Computer Science Department College Station, TX 77844-3112 College Station, TX 77844-3112 College Station, TX 77844-3112 [email protected] [email protected] [email protected] emergence of what is now called ‘emotional intelligence’ hasrevealed yet another aspect of human intelligence. Emotions havebeen shown to have a major impact on many of our everydayfunctions, including decision-making, planning, communication,and behavior. AI researchers have recently acknowledged thismajor role that emotions play, and thus have began to incorporatemodels for simulating emotions into agents. However, theemotional process is not a simple process; it is often linked withmany other processes, one of which is learning. It has long beenemphasized in psychology that memory and experience help shapethe dynamic nature of the emotional process. In this paper, weintroduce PETEEI (a PET with Evolving Emotional Intelligence).PETEEI is based on a fuzzy logic model for simulating emotionsin agents, with a particular emphasis on incorporating variouslearning mechanisms so that an agent can adapt its emotionsaccording to its own experience. Additionally, PETEEI isdesigned to recognize and cope with the various moods andemotional responses of its owner. A user evaluation experimentindicated that simulating the dynamic emotional process throughlearning provides a significantly more believable agent.1. INTRODUCTIONArtificial Intelligence (AI) and psychology researchers have longbeen concerned with defining intelligence and finding a way tosimulate it. Many theories of intelligence have been formulated,however, very few included emotions. In Howard Gardner’s book,Frames of the Mind, he describes the concept of MultipleIntelligence [9]. He divided intelligence into six types: linguistic,musical, logical-mathematical, spatial, bodily kinesthetic andpersonal intelligence. Accordingly, a person’s intelligence canvary along these dimensions. Gardner’s theory is importantbecause, by including personal intelligence, it incorporated thesocial and emotional capabilities that people possess, which thenled to the rise of what is called the ‘emotional intelligence’ theory[10]. The importance of emotions in the theory of humanintelligence has recently been strengthened through neurologicalevidence presented by Damasio [4].As a result, many researchers within the agents and AI field havebegan to develop computational models of emotions. Simulatingemotional intelligence in certain types of computer programs isimportant. Computational models of emotions are very useful tomany applications, including personal assistance applications,training simulations, intelligent interfaces, and entertainmentapplications.Recognizing the importance of the emotional process in thesimulation of life-like characters [6], a number of computationalmodels of emotions have been proposed within the agent’scommunity [33, 23]. An important effort in this area is the OZproject [2, 25] at CMU. The OZ project simulates believableemotional and social agents; each agent initially has presetattitudes towards certain objects in the environment. Furthermore,each agent has some initial goals and a set of strategies that it canfollow to achieve each goal. The agent perceives an event in theenvironment. This event is then evaluated according to the agent’sgoals, standards and attitudes. After an event is evaluated, specialrules are used to produce an emotion with a specific intensity.These rules are based on Ortony et al.’s event-appraisal model[21]. The emotions triggered are then mapped, according to theirintensity, to a specific behavior. This behavior is expressed in theform of text or animation [25].Even though OZ’s model of emotions has many interestingfeatures, it has some limitations too. An emotion is triggered onlywhen an event affects the agent’s standards, attitudes or thesuccess of its goals. The notion of partial success and failure ismissing, since typically an event does not cause a goal to fullysucceed or fully fail. Additionally, some emotions, such as hopeand relief, may never be simulated since they rely on expectations.In order for an agent to simulate a believable emotionalexperience, it will have to dynamically change its emotionsthrough its own experience. We have developed a new model foran agent that can simulate this dynamic nature of emotionsthrough learning. In this paper, we will present our model anddiscuss how different learning mechanisms are used to simulatethe dynamic aspect of the emotional process. Additionally, wewill report the results of some user evaluation experiments of asystem we implemented with this model, called PETEEI.2. COMPUTATIONAL EMOTIONIn introducing our model, we will first describe the differentlearning mechanisms that were implemented in the model. In lightof the learning mechanisms involved, we will describe how theyinfluence or facilitate the emotional process. In this way thereader will better understand the specifics of the model.2.1 LearningWe simulated four different learning mechanisms, which include(1) learning about event sequences and possible rewards, (2)learning about the user’s actions and moods, (3) learning aboutwhich actions are pleasing to the user and which actions are not,and (4) the Pavlovian S-R conditioning. These learningmechanisms will be detailed in the next paragraphs, givingexamples of their effects on the emotional process whenappropriate.2.1.1 Learning about EventsThe agent needs to know what events to expect, how likely theyare to occur and how bad or good they are. This information iscrucial for creating emotions. Consider the following: You are astudent in a classroom. Your professor is giving out exam grades.You know that you did really bad, so you are expecting a badgrade. Your professor gives you your exam back, andsurprisingly, you got a very good grade. Now, let’s look at youremotional state. Before you got the exam, you felt some fear,because your goal of passing the class was threatened by theexpectation of a bad grade. After you got the exam back, yourealized that