Bottoms Up: Improvisational Micro-Agents Brian Magerko Georgia Institute of Technology School of Literature, Communication, and Culture 686 Cherry St., Atlanta, GA magerko@gatech.edu Casey Fiesler Georgia Institute of Technology GVU Center, College of Computing 85 5th Street NW, Atlanta, GA 30332 casey.fiesler@gatech.edu Daniel Fuller Georgia Institute of Technology School of Literature, Communication, and Culture 686 Cherry St., Atlanta, GA 30332 dfuller3@gatech.edu Allan Baumer Georgia Institute of Technology School of Literature, Communication, and Culture 686 Cherry St., Atlanta, GA 30332 allanbaumer0@gmail.com ABSTRACT This paper describes our current approach in implementing computational improvisational micro-agents, agents that perform one specific aspect of our findings from the Digital Improv Project. This approach is intended to foster bottom-up research to better understand how to build more complex agent behaviors in a theatrical improvisational setting. The Digital Improv Project is a multi-year study at the Georgia Institute of Technology focused on studying real life theatrical improvisers with an aim towards better understanding the cognition employed in improvisation at the individual and group level. Categories and Subject Descriptors I.2.1 [Artificial Intelligence: Applications and Expert Systems]; J.5 [Arts and the Humanities] General Terms Algorithms, Design Keywords Improvisation, synthetic characters, cognitive science 1. INTRODUCTION Improvisational agents have been of interest to the interactive narrative community off and on for decades. For example, the Computer-Animated Improvisational Theater (CAIT) was an interactive theater system that allowed children to control avatars in a virtual world in which intelligent animated agents improvised playtime activities such as playing, fighting, and singing [1][2]. The intelligent animated agents reactively reproduced the activities one would expect in an improvised children’s play space. The agents selected behaviors reactively by decomposing tasks using a broad but shallow approach. Another example is the Improv system [3], in which virtual animated avatars can be scripted to enact a scenario. The Improv system emphasizes variability at the surface level of the presentation—the exact positioning, movements, and gestures of avatars in a virtual graphical environment by introducing noise [4] to produce natural-looking variability. Hayes-Roth and van Gent combined the Improv system and CAIT to produce a non- interactive scenario about a master and servant that can play out three different ways depending on the setting of personality traits for the master and servant roles [2]. The Mobile Robot Improv troupe at the Carnegie Mellon Robotics Institute [5] focused on creating believable agents that displayed emotionally motivated behavior within the context of a narrative experience. The performance relied on Knight’s deconstruction of a dramatic performance [6], which includes defining hero and villain archetypes, inner and outer obstacles that prevent the hero from achieving his or her goal, and the given circumstances that influence how a character attempts to achieve their goal. These are all examples of previous attempts to model improvisational behavior within a theatrical setting. These approaches to computational improvisation have assumed that agents are either autonomous (they improvise according to their own goals and beliefs) or semi-autonomous (they can receive direction from another agent, such as a human or story director agent) while improvising. Mateas and Stern have argued against what they called strongly autonomous agents, noting that they are difficult to coordinate, and opted for weakly autonomous agents in Façade [7]. While the commitment to weakly autonomous agents was a strength of Façade—it allowed for a tight coupling between agent behaviors and story goals—it also forced the agents to use dialogue-based “catch-alls” in an attempt to subtly deal with unexpected user inputs and keep the story moving (e.g., laughing uncomfortably and saying what a “kidder” the user is, then moving on). However, improvisation (“improv”) in an interactive narrative might be required of synthetic characters when unexpected situations arise (such as complex world physics that are hard to model or player actions that are not covered in the authored story space [8]) or when relying on dialogue catch-alls is perceived as undesirable. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. INT3 2010, June 18, Monterey, CA, USA Copyright 2010 ACM 978-1-4503-0022-3/10/06…$10.00.