Special Section: Intelligent Virtual Agents Guest Editors' Introduction

This issue’s special section on intelligent virtual agents encompasses four articles based on research initially presented at the 10th International Conference on Intelligent Virtual Agents (IVA) held in Philadelphia at the University of Pennsylvania in September, 2010 (Allbeck, Badler, Bickmore, Pelachaud, & Safonova, 2010). Intelligent virtual agents are animated characters capable of autonomous interaction in dynamic social environments. Work in this area seeks to endow virtual characters with capacities for perception, cognition and action, often including the ability to engage in dialogue with human users. A virtual agent must perceive the world in which it exists, both virtual and real, often including human participants’ language and gestures. It must reason about those perceptions as well as decide on how to act on them in pursuit of its own agenda. The design and evaluation of such sophisticated artifacts is a multidisciplinary effort, requiring the integration of research spanning artificial intelligence, social psychology, linguistics and computer graphics and animation. Since its inception in 1998, the IVA conference has played a central role in drawing together researchers from these fields who share a common vision, to bring these interactive characters to life, and understand the nature of our interaction with them. IVA research has become especially concerned with the bold challenge of realizing naturalistic face-to-face interactions between virtual (and often human-like) agents and human participants, typically seeking to create virtual agents that interact with people using the same verbal and nonverbal behavior that people use to interact with each other. These virtual agents have gone by a variety of names, most notably embodied conversational agents (Cassell, Sullivan, Prevost, & Churchill, 2000) and virtual humans (Rickel et al., 2002; Gratch et al., 2002). ECAs or virtual humans have been proposed for a wide range of educational, social, medical and training applications. Given the broad scope of IVA research, the four articles presented in this section only illustrate a few facets of such research. The articles selected touch on the design of ECAs, how to evaluate our designs as well as investigations into the psychology of human and virtual agent interactions. The paper by Smith and colleagues presents a design and implementation of a companion ECA, an agent designed to establish an affective or emotional relation with a human user. The paper illustrates some of the difficult technical and research challenges faced in creating ECAs, including the analysis of the user’s dialogue, managing the dialogue interaction, reasoning about how to respond, generating a multimodal response as well as how the various components realizing these capacities can be integrated into an architecture. The approach taken leverages the agent’s dialogue interaction strategies, arguing that these strategies support a principled approach to integration. The work by ter Maat, Truong, and Heylen investigates one aspect of the ECA’s behavior: its turn-taking strategies, or management of when to speak. In particular, they look at how different turn-taking strategies, such as interrupting the current speaker or waiting for a pause, influence perception of the agent and the interlocutor’s response to it. Two studies were conducted, one where subjects listened to two scripted agents interact and a Wizard of Oz study where subjects interacted with an agent controlled by a human. The results of the studies argue that the agent’s turn-taking strategy influences perceptions of the agent’s agreeableness and assertiveness and additionally impacts the subject’s own speaking behavior. Having designed and implemented a virtual agent, a fundamental question remains of how we measure and evaluate them. Researchers have often argued that a