Shared Story Ownership between Player and Game
Game Design Consultant
Per definition, Interactive Storytelling in videogames requires interaction, that is, communication emanating from both the player and the game. However, in current games, the player's ability to express his intent is severely limited, leading to poor control over storytelling. This article presents a different way of envisioning this interaction, one that is inspired by tabletop role-playing games.
If interactivity is "the capability of acting on each other"  , then the stories found in the current generation of videogames are not interactive. The player can trigger and sometimes select the events that compose these stories, but he can't act on the stories or affect the quality of the experience they offer.
This stems from two mental blocks that plague game designers and writers: That the player's avatar should be his sole mean of communication with the game and that the avatar's successes and failures are equivalent to the player's. Both these blocks are the consequences of a laudable goal: to increase the player's immersion in the game world as well as his empathy toward his avatar and, in effect, to emulate movies. But games are not movies. They are an interactive medium.
Caveat: The author is not a research scientist and this is not a scientific paper; these are just thoughts from a game developer who has been working on the problem of interactive storytelling in videogames for ten-odd years.
While repetition is only used for stylistic purpose in non-interactive stories, it is at the heart of games. Indeed, Raph Koster (2005) argues that the opportunity for the player to learn patterns in a safe environment is what makes games fun. Learning can only occur if similar situations present themselves regularly and if the means to influence them can be understood and mastered. Such a situation and the interactions it affords is what we’ll call a pattern.
The number and variety of patterns available in a videogame depends on the number of commands the player has access to and the meaning they can take depending on the context. For instance, shooting at a creature conveys the intent to harm it, while shooting at a lamp can mean trying to break it and make the area dark. Learning patterns and their uses is important since succeeding at game challenges usually require a certain level of mastery in that regard.
A player can only remember and master a limited number of commands and original patterns, that is patterns specific to the game world (such as the effects of a spell or piece of equipment never encountered before, in any other medium). This is why most games rely heavily on patterns taken from the real world or on well established game patterns (like the health pack) and why they use generic commands, like “shooting at something” in the example above. Such commands, therefore, can’t convey meaning by themselves: only the patterns do.
The narrative artificial intelligences in the current generation of story-based games all implement the same interaction loop: the player acts, then the AI interprets what the changes brought to the game world by these actions mean and, finally, changes the game world too to reflect the inferred meaning or illustrate plot advancement, which can open new actions to the player. Inferring meaning from the use of patterns by the player is a complex task because, as we've seen, the player's vocabulary is fully action-oriented. Did the player mean to shoot at the foreign dignitary across the park or was it the result of a random shot? Does he want to strike fear in the villagers by walking among them fully armed or did he forget to put his weapons back in his inventory? There's no way for the game to know. So games usually assume that any occurring event linked to a recognisable meaning is purposely produced by the player (yes he tried to assassinate the dignitary, set the secret services on him; yes he wanted to scare the villagers, let the fearful healer hide). This can be very frustrating for the player because, except if the game explicitly warns him beforehand, he has no way of anticipating which of his actions convey implied meaning and which don't  . To minimize this frustration, game designers tend to limit implied meaning to actions that have easily understood quasi-mechanical effects linked to game creatures' senses and behaviours: The player walks slowly in order to move silently and remain undetected by the guards (Thief, 1998), or he makes a noise in order to draw a guard to the place where the noise was heard (Metal Gear Solid, 1998). There is meaning here but it is canned functional meaning.
Since the player can only use action-oriented patterns to express meaning  , the narrative AI interaction loop becomes quite simple: it just checks if predetermined patterns are used or conditions met to trigger story events (for instance, when the dungeon door is opened; when the dragon is killed; etc.) The player is thus taken from one story point to another, living out the story but having no control over it. Action Adventure games use this method exclusively, usually allowing parallel story threads (or side-quests) to give back some control to the player (he can, for instance, try to find a lost dragon-slaying sword before confronting the dragon or try his luck without). Triggered events are not about understanding what the player wants to do; they just add a narrative frame to his actions. Such stories are not interactive.
Nowadays, the most successful models of interactive storytelling are to be found in tabletop Role-Playing Games (RPGs)  .
Thirty years ago, RPGs started where story-based videogames are now: players did little more than live out pre-written adventures, their only freedom being the choice of their moment-to-moment actions. But RPGs quickly evolved, each new game bringing innovations such as eschewing the adventure model and replacing it with a fully developed world the players could explore and change (Steve Perrin, 1978); giving the players the means to control fate and luck (G. C. Klug, 1983); having the story affected by characters' passions (Greg Stafford, 1985) or flaws (Steve Jackson, 1986); encouraging all the players in the group to take turns as the storyteller (Jonathan Tweet, 1988); or even driving inexorably the characters toward madness and death (Sandy Petersen, 1981). RPGs continue to evolve today, still experimenting with form, rules, and themes. Yet computer games cling to a thirty year old model.
The evolution of RPGs follows a clear path: each new generation of games gives more narrative powers to the players, powers that were previously the sole province of the Game Master. Sharing story ownership between the players and the storyteller increases the quality of the narrative experience as the story meets the needs and desires of the players more often and more easily. In order to achieve this shared ownership, RPGs define rules about how players can affect the story beyond the actions of their characters. These rules are storytelling patterns.
Sharing story ownership means that the player is no longer a simple actor. He can change the game world beyond what his avatar can do and then slip back in his actor role and deal with or profit from the new situation. This is a huge paradigm shift.
Videogames are often conceived as a series of challenges presented not only to the player's avatar but also, through it, to the player. The player is thought of as playing against the game, his avatar's victories and defeats becoming his. But what if a defeat generates a more interesting story? What if some victories feel pointless and boring to the player (as can victories in random repetitive combats)? What if the player feels that only one outcome to a situation is interesting, however improbable or difficult it is to obtain? Shouldn't the player be allowed to tell the game about it - to play with the game - so that the perceived problem can be fixed?
Harnessing the Narrative Power of the Player
James Bond 007 Role Playing Game (G. C. Klug, 1983) presented its designer with a challenge: what kind of game rules could simulate the world accurately and still allow for the improbable stunts and lucky breaks the Bond-like agents need in order to survive and accomplish their missions? He solved the problem with the Hero Points storytelling pattern. Hero points are a limited resource players can spend to emulate Bond's luck and extraordinary skills. Hero Points let an agent find a sport car with its motor running just before a car chase scene; or shoot (without any chance of missing) an underling who is about to activate the villain's base self-destruct mechanism; or jump unscathed through a hail of bullets; and so on. Hero Points made the game more interesting as taking risks became fun, just like in a James Bond story.
A player in a videogame could have the same ability to bend the world rules when he feels it will make the story more interesting. This could include skewing the odds of random events (perfect skill use, random combats avoidance, etc.), finding an object his avatar needs, having it meet a character "by chance", etc. Conversely, the player could create problems for his avatar (jammed gun, additional guard patrols, sneezing bout, etc.) to increase the dramatic tension or the challenge level.
In such a model, since the game knows the purpose of the events chosen by the player, it can present their outcomes in a dramatic manner (the game sends a patrol near the avatar just when it's about to sneeze) or in the form of new challenges (for instance, the dungeon key wished for by the player appears dangling from a guard's pocket and the imprisoned avatar must grab it without being detected when the guard walks past its cell).
While the above storytelling pattern allow the player to tell the game what he wants, story contexts let it know how the avatar feels and what the player's intent is.
A story context describes the current emotions, feelings or world views of the avatar and, through them, the player's goals. Using this information, the game can alter existing interactions, make new ones available and remove old ones. In current games, action-oriented patterns are routinely modified according to the avatar's physical state - a wounded avatar would move slowly and with a limp - but rarely according to its mental state since it's supposed to be the same as the player's  .
Story-based games where the player can explicitly choose the emotions felt by his avatar are rare. Indeed, emotions are often used to get the avatar moving toward the next plot point ("They killed my partner. I won't rest before I avenge him."), so letting the player control them could ruin the storyline. Again, this mental block requires a paradigm shift. These games aim for passive player empathy (let him feel happy / sad / vengeful / afraid along with his avatar) instead of letting him actively shape the emotional landscape of the story. Choosing feelings and emotions for one's avatar could be a result of empathy or projection, but it could as well be motivated by curiosity or strategic thinking ("I won't let anger fill my heart and render me blind to the true reason why my partner died").
Used that way, emotions / story contexts can themselves trigger story events or be illustrated by the game to reinforce the chosen mood. For instance, let's say the avatar is alone at night in a deserted back alley. Two of the available moods are nervous and confident. If the player chooses to make the avatar nervous, the game reflects this by showing him the world the way his avatar perceives it  : its footsteps sound louder than usual, there are whispering winds, curls of mist, howling dogs in the distance, menacing shadows, and glittering eyes in the darkness. A nervous avatar would attract the attention of some low-life and would get mugged. Conversely, the player of a confident avatar could see a brightly lit alley, rats fleeing before him, and pass through it unchallenged.
Why then would the player want his avatar to be nervous? Is that the "wrong" choice? This illustrates the difference between playing against the game (trying to win every challenge) and playing with it to build an interesting story. Plot complications are interesting. Maybe the mugging would lead the beaten avatar to a hospital and start a romantic sub-plot with a compassionate nurse; maybe the avatar would overpower the mugger and acquire information from him or gain status in the underworld.
Explicit Player Goals
In the dead partner example above, wouldn't the experience be more intense to the player if he chose to seek revenge, instead of being told by the game that he had to? Choosing one's goal can also be done with the help of story contexts.
Story contexts have different scales and can be nested. They govern the short term (emotional reaction to a situation that can lead to opportunities, as above), the mean term (lasting moods that can alter character relationships), and the long term (goals, values, beliefs). Rules link them logically or dramatically. For instance, a string of small victories can make the optimist mood available that, if chosen, augments the avatar's luck but can lead to sudden dramatic reversals of fortune. Or, if the player lets his avatar get angry, it could augment its strength and resilience but close the access to other possible interactions (like negotiation) and require special events to exit this mood (like a retreat in a monastery).
This opens a new emotional space to explore, beyond the physical space and the story timeline.
Using these different types of context, the player can inform the game of what he wants now, in the near future and what the global experience should be about. Is it a story about a character seeking revenge or seeking justice? Is it about attaining a material goal or preserving a relationship? Is it one of noble sacrifice or one of survival against all odds?
If the player were able to make such a choice early in the game, it could colour the whole experience.
Making such games is hard because they must be designed and implemented holistically. Action-oriented patterns (the rules of the game world) and storytelling patterns (how to cheat) are interdependent and determine the kind of stories that can be told and, thus, also the kind of suitable game world.
In order for the player to use storytelling patterns, the relevant rule parameters (such as the probability of random attack by monsters) must somehow be known and accessible to him. In the case of object "creation" or character behaviour alteration, each potential event must be pre-scripted separately or follow precise rules. To ensure that these events can be combined, their game descriptions should use shared concepts and functionalities (like guards reacting to "localized noise" and not only to a sneeze, a cough, a snapped twig, etc.). This requires a well developed library with width (a number of interesting concepts without significant overlaps), depth (detailed behaviours that use these concepts), that can be used to describe all the possible interactions between objects in the world and that contains no inconsistencies. Furthermore, story contexts must be able to alter the characteristics of these events and concepts dynamically, as well as other contexts.
This means that one cannot develop such a narrative AI outside of a specific game. The designer has to make choices regarding the kind of experience he wants to provide the player with. These choices will limit the possibilities of interaction with the game world and with the story and decrease the design's complexity.
For instance, the game could focus on exploring the storytelling possibilities of a confined world (à la Groundhog Day); on James Bond-like rule bending (which would require a simple enough world to be fully described, such as the ones in the early Zelda games - like The Legend of Zelda, 1987); on letting the player set his own goals in a complex evolving world (with the dynamic creation of appropriate challenges); etc.
Current videogames give the player an interaction toolset that is too limited for him to express his intent. If the goal of interactive storytelling is to allow player and game to create together a pleasing narrative experience, such a game must share ownership of the story with the player. This means that the perception of what a player does in the game has to change. He should no longer be confined to playing the part of his avatar. Instead, he should be considered as a performer as well as a part-time director, writer, critic and prop master. The game should adapt the experience to fulfil the player's clearly expressed wishes - while still offering him challenges and surprising him. Videogame storytelling will then truly be interactive.
Steve Jackson, GURPS, Steve Jackson Games, 1986.
G. C. Klug. James Bond 007 Role Playing Game, Victory Games, 1983.
Raph Koster. A Theory of Fun, Paraglyph Press, 34-47, 2005.
The Legend of Zelda, Nintendo, 1987.
Metal Gear Solid, Konami, KCEJ, 1998.
Steve Perrin & al., Runequest, Chaosium, 1978.
Sandy Petersen & al., Call of Cthulhu, Chaosium, 1981.
Greg Stafford, Pendragon, Chaosium, 1985.
Thief: The Dark Project, Looking Glass, Eidos Interactive, 1998.
Jonathan Tweet & al., Ars Magica, Lion Rampant, 1988.
 As www.dictionary.com defines it. It's not reasonable to believe that they all do. Actually, the more implied meanings - and thus the more realistic the game world is - the more jarring inconsistencies appear to the player. Pre-scripted modal dialog with game characters can convey meaning. However, since it doesn't use sophisticated AI (most game dialogs are explorations of a dialog tree or graph) but instead relies on writing quality and choice variety, it is beyond the scope of this article. In a RPG, each player except one improvises the part of a character in a story. The remaining player is the Game Master (or storyteller). His role is to describe the world where the story takes place and the result of the characters' actions. He also plays the parts of all the other characters in the world (the "supporting cast"). The goal of a RPG session is not to win the game but to have a pleasurable storytelling experience. For proof, the current trend of output alteration depending on the avatar's state, like blurring the screen when it is dazed.
 Subjective viewpoint is an effective technique to enhance empathy or suspense. It is used in countless movies.