You are listening to a Podhoc podcast — a platform where anything can be turned into a Podcast to Learn in Motion.
Storytelling, in its oral, visual, or written forms, is fundamental to human culture, acting as a cognitive tool for understanding our experiences. This "narrative intelligence" allows us to organize our lives into coherent stories, a skill that computational systems could potentially emulate. If computers possessed this narrative intelligence, they could interact with us more naturally, understanding collaborative contexts as emergent narratives and expressing themselves through stories.
This capability opens the door to interactive narrative, a form of digital experience where users shape a dramatic storyline through their actions. Whether playing a character, commanding others, or directly manipulating the virtual world, users influence the unfolding plot. While often seen as entertainment, interactive narrative also holds significant potential for education and training.
The core idea is to immerse users in a virtual world, making them feel like integral parts of the story, where their actions have meaningful consequences. This is what distinguishes interactive narrative from other digital entertainment; the player’s actions fundamentally alter the story's direction or outcome. Think of the holodeck in Star Trek, creating immersive fictional realities.
Imagine stepping into the holodeck as Hamlet, but instead of following Shakespeare's script, you decide to confront Claudius in the first act. The critical question then arises: what happens next? How does a computational system determine the subsequent events, and what knowledge does it need to reason about your experience?
These questions delve into creating computational systems with narrative intelligence, moving us closer to automated, engaging experiences. This pursuit also addresses a fundamental quest in artificial intelligence: creating systems that exhibit human-level capabilities, such as storytelling, which remains uniquely human. The third question, about intervening in the fictional world for user benefit, points to practical applications.
While interactive narrative has roots in entertainment, a growing number of researchers are exploring its use in "serious" applications like education and training. In STEM education, it parallels problem-based learning, guiding students through interconnected problems. Similarly, procedural skill acquisition benefits from realistic contexts provided by interactive scenarios.
The central research challenge in interactive narrative is balancing coherent story progression with user agency, two often conflicting needs. A coherent narrative flows logically from event to event, culminating in a conclusion. User agency, however, allows players to act freely, sometimes introducing inconsistencies or making planned future events impossible.
This tension arises because users might act in ways they deem best but are unaware of the narrative's intricacies, or they might deliberately try to steer the story in new directions. The key is to ensure players feel their actions matter, while the overall experience remains coherent and engaging.
A common solution to this challenge is the "drama manager," an intelligent, disembodied agent that monitors the virtual world and guides the narrative. A more generalized concept, the "experience manager," extends this to non-dramatic narratives like educational ones. This manager intervenes, typically by directing non-player characters, or NPCs, to respond to user actions.
The experience manager needs to anticipate possible future states of the user's experience to intervene effectively and maintain coherence. This isn't about predicting the most likely sequence of events, but rather the *best* narrative sequence according to specific criteria, achievable given user agency and intervention capabilities. Projecting narrative trajectories allows for evaluating the global structure of potential experiences.
From a state-and-actions perspective, the virtual world transitions through states based on actions by users and NPCs, and by the experience manager itself. If a trajectory passing through certain states is deemed "poor" and one passing through others is "good," the experience manager's goal is to guide the user's path towards the desirable states. This requires continuous projection and selection of future narrative trajectories.
How does an experience manager determine if an intervention will improve the user's experience? It can reason about criteria like narrative structure constraints, dramatic tension, pedagogical goals, or aesthetic considerations, all defined by a human author. The experience manager acts as an autonomous surrogate for this author.
The landscape of interactive narrative research can be mapped along three dimensions: authorial intent, virtual character autonomy, and player modeling. Authorial intent focuses on how much the human author's original vision constrains the system, ranging from highly prescriptive to allowing the system creative freedom.
Virtual character autonomy addresses how independently NPCs can act from the experience manager. Strong story systems mean NPCs are fully controlled by the manager, while strong autonomy systems have NPCs acting without narrative awareness. Most systems fall somewhere in between, balancing narrative consistency with character believability.
Player modeling examines how the system learns about individual user differences, such as preferences and play styles. While all interactive narratives adapt to user actions, player modeling explicitly builds a model of the user to predict behavior and manage the experience more effectively.
Concerning authorial intent, systems can range from highly constrained to those where the system takes creative responsibility. At one extreme are "Choose Your Own Adventure" style systems with manually authored branching paths, ensuring the author's vision but becoming unmanageable with scale.
As complexity increases, the experience manager must assume more creative control, inferring authorial intent from higher-level specifications like plot graphs. These graphs define events and precedence constraints, pruning illogical or undesirable trajectories. Search algorithms then generate and evaluate possible paths.
Moving further along this spectrum, the experience manager possesses more general knowledge and is responsible for constructing narratives from scratch, often embedding a full story generator. The human author provides high-level knowledge, and the system generates the narrative, for instance, using a domain theory of possible actions and their conditions.
A key challenge for generative systems is the "boundary problem" – recognizing and responding to user actions that deviate from the intended narrative. For example, if a user kills a crucial NPC, the narrative can't proceed as planned. The system must then either intervene or accommodate this deviation.
Intervention involves the experience manager acting to prevent boundary crossings, perhaps by having NPCs interact differently or altering the fictional world, like causing a weapon to jam. This aims to keep the user within the pre-defined narrative space.
Accommodation, on the other hand, allows the deviation to occur and then uses the story generator to create a new narrative that resolves any inconsistencies. This might involve, for example, selecting a new NPC to replace the one who was killed.
Story generation itself remains an open research question, with generative experience management aiming to have the system assume creative authorial responsibilities. This is particularly relevant for creating dynamic and responsive narratives that can adapt to unexpected user actions.
The dimension of virtual character autonomy relates to the independence of NPCs, impacting their believability. Believable characters foster suspension of disbelief, appearing as intelligent beings with personality and emotions. Stories are driven by character interactions, making them crucial to interactive systems.
However, a tension exists between the experience manager's goal of global narrative coherence and NPCs' local actions aimed at believability. In "strong story" systems, NPCs act only under the experience manager's direct guidance. In "strong autonomy" systems, NPCs are fully independent agents unaware of the narrative's needs.
Most interactive narrative systems try to find a balance. For instance, the Façade interactive drama uses personality-driven decompositions to realize plot units, with the experience manager providing coherence while characters autonomously determine how these units are enacted.
Under the "strong story" approach, the experience manager has maximum control to shape the user's experience. However, tracking every NPC interaction or dialogue can be computationally intractable. The manager typically operates at a higher level of plot abstraction, realizing actions only when necessary.
In "strong autonomy" systems, the narrative emerges entirely from the uncoordinated decisions of NPCs and the player. This approach, often called "emergent narrative," is common in simulation games or exploratory learning environments where guiding the user to a specific conclusion isn't always the primary goal.
Player modeling focuses on learning about the user's individual differences, like preferences and play styles. While all interactive narratives adapt to user actions, player modeling explicitly builds a model to predict behavior and enhance the perceived experience. This is especially relevant for serious applications like education and training.
For educational purposes, interactive narratives can function like intelligent tutoring systems, scaffolding learner problem-solving within a narrative context. If the system uses a generative experience manager, it can also be responsible for generating problems, adapting them to the learner's needs.
The use of artificial intelligence in interactive narrative systems enhances their expressiveness by taking on creative responsibility, leading to greater responsiveness and a wider range of narratives without sacrificing player agency. This is a rich area of AI and machine learning research.
Significant progress has been made in enabling intelligent systems to deliver engaging narrative experiences for entertainment, education, and training. However, open questions remain about how and when to intervene in virtual worlds, how to generate narrative structures, and how to effectively incorporate believable characters.
Human authorial intent is crucial in interactive narrative development. While branching story graphs are accessible, AI-driven systems require significant knowledge engineering. The goal is to make these intelligent systems accessible to non-expert authors, potentially transforming interactive narrative into a mainstream creative expression.
Storytelling is an intrinsic part of the human experience, used for communication, entertainment, and education. Immersive, interactive systems that engage us in narratives while allowing user agency have the potential to revolutionize how computational systems are used, offering a compelling domain for exploring computational narrative intelligence.
Thank you for listening to this Podhoc podcast.
