Finding Schrödinger's Gun

Interactive narratives are branching stories with events that change based on feedback from participants. One method of generating interactive narratives is a planbased process called mediation. A sub-process within mediation called accommodation creates new story content when a participant deviates from the main storyline. We show that a model of character knowledge allows accommodation to find a novel class of branching stories previously inaccessible by the algorithm. Introduction Interactive narratives are available in many mediums such as tabletop role playing games, gamebooks, interactive fiction, and video games. One drawback of authoring interactive narratives is that a branching storyline requires a large amount of story content. If branching narrative is viewed as a story graph whose vertices are story content and edges are user actions, unique user choices are edges that transition to unique vertices. A branching narrative that offers a large number of unique choices will have an equally large number of vertices which creates a high authorial burden or combinatorial explosion of story content (Bruckman 1990). Luckily, AI algorithms can automate the creation of branching story graphs. Planning is one system capable of generating interactive narratives (Porteous, Cavazza, and Charles 2010). Recent research has modeled interesting properties of narrative within the context of planning such as character intention (Riedl and Young 2010) and conflict (Ware and Young 2011). These linear planners can be paired with an algorithm called mediation (Riedl, Saretto, and Young 2003) to manage human interaction as a character within the story. Mediation accomplishes this by building a mediation tree data structure. The mediation tree is a representation equivalent to story graphs (Riedl and Young 2006). However, there is a class of interactive stories that mediation cannot currently generate. This class corresponds to the phenomenon popularly termed Schrödinger’s Gun. In this paper we incorporate a model of character knowledge into mediation to generate this novel class of interactive stories. Copyright c 2014, Association for the Advancement of Artificial Intelligence (www.aaai.org). All rights reserved. Related Work Mediation is a plan-based interactive narrative generation algorithm first created for the Mimesis system (Young et al. 2004). Mediation generates a branching story, represented by a mediation tree, by applying accommodation and intervention to a linear story plan. Accommodation is the process of expanding a mediation tree while intervention prunes away possible branches. There have been several modifications of mediation (Riedl et al. 2008; Harris and Young 2009; Ramirez, Bulitko, and Spetch 2013) but unlike previous work we expand the number of branches mediation can create while building its tree. We accomplish this by identifying a new type of interactive story during accommodation. Our process is a form of alibi generation (Sunshine-Hill and Badler 2010; Li et al. 2014), a method that dynamically creates backstories for non-player characters (NPCs). Sunshine-Hill and Badler use alibi generation to reduce the cost of simulating large crowds by transitioning NPC behavior from a random distribution to an intelligent pattern when observed by a player. Li et al. show that socially believable alibis can be learned from exemplar graphs obtained through crowdsourcing. In this paper we present a mediation system with a model of character knowledge that can replan past events during accommodation, rewriting NPC alibis in response to player actions so they further an author’s goals while building a branching story. Mediation Algorithm In this paper we present descriptions and definitions that span three layered algorithms: planning, mediation, and Schrödinger accommodation. This section briefly describes the planning and mediation machinery on which our work is built as a foundation for later sections.