Foundations of Interactive Storytelling

Stories have most likely been part of the human experience from the earliest days of language, but until recently the storytelling medium has been largely static. Barring different versions of the same story, any given tale unfolds the same way every time one reads it. Computer games promise the potential to move beyond this strictly linear form by offering stories that interact with the player, allowing them to participate in the decisions or actions that shape the narrative. However, at the current time this field is still in its infancy.

The origin of interactive storytelling probably dates back to Tactical Studies Rules Dungeons & Dragons (1974), itself based upon a set of tabletop battle rules - Guidon Games' Chainmail (1971). Similar in terms of mechanics, Dungeons & Dragons had a revolution in its focus, changing the game play to being about characters, not armies. This was the start of tabletop role playing games (RPGs), which were to influence and inspire many computer games to follow.

The tabletop RPG is a 'perfect' interactive storytelling medium, as an unrestricted story develops from the dynamics of the group of players, led by a Games Master (GM). The former decide upon the actions of key characters in the story, the latter takes the role of the rest of the universe - becoming very much the surrogate author of the collaborative tale that develops.

In many ways, the quest for interactive storytelling in games is a quest towards the perfect state that tabletop RPGs already had. The computer game takes the role of the Games Master, and the players remain in control of the lead character's (or characters') actions. The problem is that a human GM can respond to any conceived action on the part of the player, but until Artificial Intelligence advances an astonishing degree, computer games are inherently limited in their flexibility.

What computer games lose in terms of the freedom offered in tabletop RPGs, they make up for in immersion. Any tabletop RPG game is still essentially a group of people developing a dynamic story through words alone - a computer game provides the visual and aural experience that brings it closer to cinema than radio theatre (which is the medium that tabletop RPGs arguably resemble most strongly). This difference in the size of the audience for film compared to radio plays is reflected in the difference in audience for tabletop RPGs versus computer games.

This essay looks at the foundations of interactive storytelling, with a focus on their relevance to computer games. Tabletop RPGs (and the related medium of adventure gamebooks) are an important part of the background and history of this new medium, but will forever by a minority pursuit. Computer games, on the other hand, look set to take the reigns of interactive storytelling and bring them to a broad, diverse and expectant audience.

Defining Interactivity

What is it we refer to when we talk about that which is 'interactive'? Although this may seem trivial, language is seldom a solid foundation for discussion without some definition of terms, and choosing what we mean by 'interactive' and 'interactive storytelling' can only serve to clarify the scope of this essay.

Reactive: We can define 'reactive' as "responding to some influence or input." That which is reactive is associated with a flow of information in one direction. Television is a reactive media: you can only watch (react) to the information it provides. You cannot (at least not prior to the recent introduction of interactive television shows e.g. Endemol's Big Brother series) influence the events that you are watching.
Interactive: We can define 'interactive' as "reciprocally active; acting upon or influencing each other; allowing a two way flow of information." Every computer game is inherently interactive, since the player performs certain activities, sees them occur in the game world, and then adapts their actions to reflect the results in the game world. There is an exchange of information between the player and the game which is bi-directional.
There are numerous degrees of interactivity, however, and as computer games have advanced, we have seen advances in the degree of interaction. From an initial state where control is the only interactive aspect, e.g. Nutting & Associates' Computer Space (1971) or Atari's Pong (1972), we have seen interactive levels of ability (i.e. power ups) with Midway's Pac-Man (1980), interactive environments with Universal's Mr. Do (1982) and so forth.

Understanding the distinction between reactive and interactive media is the first step towards understanding what can meaningfully considered an interactive story.

Interactive Stories

Any game featuring both characters and a story in which one or more narrative aspects changes interactively can be considered an interactive story. The possible narrative aspects that could be made interactive include:

Plot: The most obvious route to interactive storytelling is by creating a plot that varies in response to the player's actions. Perfect Entertainment's Discworld Noir and (to a much lesser extent) Konami's Metal Gear Solid are examples of this.
Character attitudes & personality: A more satisfying way to approach interactive stories is to look at ways that the player's actions might affect the attitudes of characters in the game world. We will look at several ways of approaching this later in this essay.
Theme: Although strictly hypothetical at this time, it is possible to conceive of a game in which the theme varies interactively. For example, we can imagine a game in which the story elements are mediated by the game's story engine in relation to that which the player has paid the most attention. If the player spent considerable time talking to a romantic interest, the theme of the story becomes biased towards a romantic element; if they focussed on violent activity, the thematic details might evolve around an exploration of violence.
Mastering dynamic plots and character attitudes will almost certainly be a prerequisite to exploring interactive themes.

Note that our definition requires more than both interactivity and narrative elements to be both present: many computer games (which are inherently interactive) have story elements, but that does not make them interactive stories. Only if the form of the story itself is interactive can we seriously make a claim to it being an interactive story.

The Value of Interactive Storytelling

Now we have defined what can be considered an interactive story, we are in a position to discuss why interactive storytelling might be of value - both artistically and commercially. Just because a thing can be done does not make it worth pursuing, and with the medium in its current, nascent form it is necessary to make a case for its importance.

From an artistic perspective, interactive storytelling represents a new, highly expressive art form. Before the 20th century, all art was essentially reactive. A work of art was created, and it created an emotional (or intellectual) experience in its audience. Interactive art creates a dialogue between the art and the audience, creating a more immersive artistic experience.

This issue of immersion is linked to the commercial appeal of interactive storytelling. Many games satisfy the player's need for a sense of involvement in the story by creating the illusion of an interactive story. Either the game forces the player to jump through certain story-based hoops in order to progress, e.g. Sega's Shenmue series, or the game features a gameplay element wholly separate from the narrative, as is seen in Square's Final Fantasy series in which the player is engaged in largely pointless combat between story elements. In both cases, however, the players actions have essentially no effect on the progression of the story. When executed subtly, this illusion can be quite effective; often, however, players are left asking "why can't I do that?" or worse "why do I have to do that?"

Providing a true interactive storytelling experience has the potential to satisfy the audience by connecting their actions to the unfolding story directly. This may be as simple as reflecting the player's actions in dialogue, or as complex as dynamically varying the sequence of events to work around the player's choices. Either way, this non-linearity appeals to the hardcore gamers (including the majority of magazine reviewers) offering a direct commercial advantage. It also implies a greater replay value, as hardcore players replay the game to see the other ways the story could have transpired.

Of even greater commercial value is the potential to unlock a vast reservoir of casual gamers already circling at the fringes of the games medium. The casual gamer lacks the patience (or stubbornness) to persevere with a game that annoys them, but once hooked in become intensely loyal to those games that have grabbed them. As with television and film, there are few better ways to captivate than with interesting characters and plots. However, the viewer expects no control of the outcome in the former media - in games, control of the flow of events is expected, but is usually illusory. Games with relatively easy-to-play game space, but an interesting dynamic story space have the potential to dramatically expand the size of the computer game audience.

Interactive Plots & Non-linearity

The term 'interactive plot' can be taken to be broadly synonymous with the terms 'non-linear story' or 'non-linear plot'. The term 'non-linear' is self-explanatory, but in this context should be taken to mean that the story elements (dialogue with other characters, cut scenes et al) do not occur in a fixed sequence. There is a wide range of extents to which this non-linearity can manifest in a game story. Note also that a game can be non-linear and yet still have a fixed outcome - the issue is the journey through the plot, not its conclusion.

The structure of a game's story determines wholly the extent of its non-linearity, and there are various degrees of interactivity of plot that can be expressed.

Linear

The most basic structure imaginable is linear: the game's story elements unfold in a strictly linear sequence. Sometimes the game play will be non-linear, but the game elements can unfold in a non-linear way. This is true of many games at the moment: the story proceeds as a series of linear checkpoints that the player reaches in sequence, although they may have freedom to explore the game world in a broadly non-linear fashion.

We can visualise the linear structure as a straight line:

The advantages of the linear structure are that it is a reliable and resource-cheap way of approaching the story.

In games with a low budget, the linear structure is often the only affordable approach.

The disadvantages are obvious: it is unimaginative, and the player has no control over the development of the plot. If the game successfully manipulates the player into caring about the same issues as the characters, the disadvantages are minimised - but if the player ever feels trapped in someone else's story, the linear structure is usually to blame.

Branching

The branching structure is rarely used, and for good reason. The branching structure leads to a number of fatal problems, and indeed is in part responsible for the miscomprehension that non-linear stories are intractable or inappropriate for use in games. It is unfortunate that many people in the games industry believe that the branching structure is the only non-linear option available, and this may be in part the reason for the persistence of the linear structure.

We can visualise the branching structure as a tree:

The plot begins at the base node, and certain points in the story are defined as branch points, where the plot splits into separate paths. With no recombination of threads, the branching structure leads to a number of end points equal to the number of branch points plus one (for two-way branching) or even more (for multi-path branching).

The fact that the number of end points is so huge is what is known as the combinatorial explosion. Branch in two ways four times and you have to deal with sixteen different paths (2⁴ = 16) - and each binary branch point that you add to all paths doubles the total number of outcomes. No game could survive such a vast investment in potentially unused resources, and as a result the branching structure (when it does appear) is usually pitifully simple (often just a single branch point).

Parallel Paths

The parallel path approach (which could also be considered a fully recombined branching structure) is a tenable and useful approach to story design, used to some extent in games such as Konami's Metal Gear Solid (1998), and Ion Storm's Deus Ex (2000). Branches recombine at the key story points, creating a structure which is a balance between linear and branching structure:

In the following diagram, those story points which have a green line through them are branch points splitting into three routes - but those paths then recombine before the next main story point. The underlying structure is linear:

every player must through every point on the spine of the story, but there is still some freedom as to how the player reaches the next point.

Metal Gear Solid, whilst essentially a linear game, features a single segment of parallel path structure which does much to improve the player's involvement in the game's story. This sequence, in which the player is tortured repeatedly but has an opportunity to escape between the torture sessions, has three paths through it: the player escape in a couple of different ways from the cell between torture scenes, they may admit defeat to the torturer, or if all else fails, they are eventually rescued from the cell by a deus ex machina. This parallel path segment also governs a split in the games ending: if the player does not give up in the torture scenes, they will end up rescuing the romantic interest - but if they do, she is killed and the player ends up escaping with a buddy character instead.

Although simplistic, this remains a strong example of the power of even a basic non-linear structure, giving the player a genuine moral decision to face with consequences within the framework of the story.

This example also typifies the needs of the parallel path structure's branch points: a primary path (rescue the romantic interest), a secondary path (escape with the buddy character) and a catch-all tertiary path (the player is rescued) to ensure that the plot will move forward whatever happens. In point of fact, this sequence from Metal Gear Solid is two separate branches (resist torture, or give up is the first branch; escape on their own, or be rescued is the second branch) leading in effect to four different paths through this sequence, ignoring the issue of the means by which the player may escape by their own means. Only the first branch is expressed in the development of the rest of the plot, however, and in the branch at the end of the story.

The disadvantage of the parallel path structure is that more is developed for the game than is actually seen. As soon as you have two endings, only players who play the game twice will have any hope of seeing all of the story material, although this disadvantage is offset if it is expected that most players will replay the game (i.e. the game is both short and fun).

The advantage lies in the choices the player has within the plot, and the sense of connectivity with the story that is engendered. Every time a game gives the player a meaningful choice, the become more emotionally involved with the story - although of course this may only be effective if the player knows they had a genuine choice. Because many players do not expect to be given a meaningful choice, sometimes the player will play along with what they believe is expected of them, because their prior experience of games has given them no reason to believe that they are facing a genuine decision. It is important, therefore, to ensure that the player is subtly taught or informed that their actions will affect the outcome, if this is in fact the case.

Threaded

The threaded structure is a powerful, complex way of allowing the player a non-linear story experience. Games such as Perfect Entertainment's Discworld Noir (1999) use a threaded structure to give the player a genuine degree of freedom to explore the story space once the initial formalities of establishing setting, character and interface have been concluded.

The following diagram is an approximation of the structure of Discworld Noir's first two Acts:

Unlike previous diagrams, the course of the plot does not follow a single path in this form of structure. Rather, the story is comprised of a number of different threads that develop largely independently (at least until the closing stages of the story). This parallels the methods that quality novels and films employ to develop the plot, with multiple narrative threads being woven together to produce a satisfying story.

The key to the symbols employed is as follows:

Yellow circles represent the events at the start of an Act. Every Act starts in the same way in this particular example of the threaded structure, although this is not a given
Blue circles represent non-mandatory plot events
Red circles represent mandatory plot events: the player must experience all red circles (and must experience at least one circle with a black centre) to proceed to the next Act.
Red lines show Act boundaries.
Green lines show the places at which threads connect to one another (the places at which new threads are 'unlocked').

For example, starting at the beginning of Act I (the bottom yellow circle), the player is introduced to some back story and placed in the role of a private investigator in a fantasy world. They begin their investigations and soon reaches a point at which they are investigating multiple cases (the green line opens up two other threads, one to the left and one to the right). The player can eventually conclude the first stages of their initial investigation in two different ways (the red circles with black centres), but they may also begin investigating two other (apparently) separate cases (the path to the left and the path to the right). The right-hand thread represents the true conspiracy the player is investigating, and by the end of the first act, the player has already changed their conception of what is going on twice (signified by successive jumps along green lines).

The left hand path (which represents a side-story concerning a missing dwarf) shows an optional side path in which the player may investigate the case to a certain extent in the first Act, but cannot conclude it until the second Act. This case can also be largely side-stepped, as another chain of events in the second Act will eventually connect with the details of this case allowing it to be ignored - at least until it closes in on its terminal red circle which is an essential part of the plot.

Although this system is highly expressive, and leads to a freedom of movement within the story space which is uncommon, there are several problems which makes this approach of limited future value. Firstly, tracking changes in the plot skeleton during development is nightmarishly difficult: because of the sense of interconnectivity between events, it is very hard to implement structural changes. Furthermore, the complexity of the plot skeleton means that very few members of the development team will fully understand the structure. Given how difficult it is to fully appreciate a diagram showing a simplified version of the structure, it is easy to see why dealing with this structure in practice is no small undertaking.

Additionally, the degree of variance in the story structure makes QA much harder to perform completely (especially if there is no way to educate the QA team on the full extent of the underlying plot structure) which can lead to an extended QA period being required to ensure the stability of the code.

Of course, these problems are offset by the highly expressive nature of the structure, and the number of options for player freedom within the story. Unfortunately, much as with the parallel path structure, a great deal of this is lost on players if they do not realise that the freedom exists. Unlike the parallel path structure, however, the threaded structure has an ace up its sleeve: with so many routes through the plot supported, the player may not recognise the non-linear nature of the story, but they find that the game supports a fair degree of the actions they want to perform within the plot, which aids immersion. In essence, the advantage comes in the player not being blocked in their attempts to experiment - they are never faced with the inevitability of conforming to a linear plot (at least until the threads begin to recombine prior to the ending).

Nonetheless, what we really desire for interactive plots is a system as expressive as a threaded structure but as robust as parallel path structure.

Dynamic Object Oriented Narrative

A possible future of interactive plots is suggested by systems for dynamic object oriented narrative such as International Hobo's FreeSpeak. These use a notion from computer science (that of object orientation) with the framework used in commercial television series to provide structures that are even more expressive than the threaded structure, yet as robust as a parallel path structure.

The following diagram shows a hypothetical object oriented plot skeleton:

Before discussing the illustration, the idea of 'object orientation' must be approached. Simply put, object orientation concerns dealing with a problem by composing the data space from objects, which contain other objects, which contain other objects (like quarks in hadrons in atoms in molecules in bigger things). Everything is composed of small functional components, except for the atomic units at the base of the chain. The idea revolutionised programming languages in the 1980's, and may be about to do the same for interactive storytelling.

In the above diagram, blue squares represent an object known as an Episode (small blue squares indicate Episodes not covered in this example diagram), while all circles represent a Scene. Each episode consists of a number of Scenes which can use any structure - linear, branching, parallel path etc. Because each Episode is comprised of a small number of Scenes, the resulting set-up is robust and connectivity can be tracked on two different levels: transitions between the Scenes of an Episode (blue lines), and transitions between Episodes (red arrows).

Episodes are broadly chronological, but Episodes that refer to events that occur concurrently (i.e. that are at the same height in the case of this diagram) can occur in any sequence. In this particular example, the possibility of starting the story at different 'places' in the plot is represented by yellow circles (which might, for instance, represent playing through the story with a different character who joins the events at a different point).

Obviously if each Scene were a unique 'game level', this would suffer from many of the problems associated with multiple paths: developing many more resources than will typically be seen by the player. This is solved in FreeSpeak by the use of various other component entities, namely Personas (characters) and Locations. Every Scene takes place between certain Personas in certain Locations, and may be interactive or a cut scene. A particular Location may be used for many different game play and story related activities. For example, in Sega's Shenmue (2000) the docks are used as a stage for investigation, cut scenes, fighting, quick time events (interactive cut scenes) and even racing (although it should be noted that this reuse of location was performed in a linear fashion in Shenmue).

Any key gameplay that the designers wish every player to experience can be built into the plot skeleton as an event that must occur at some point - but the object oriented structure allows for that point to occur at many different places within the structure of the plot. The rest of the time the Locations are being used as a setting for basic repetitive game play (fighting, exploring, investigation etc.) using the core game engine. The design of a location is thus built around its uses in the game, but it can also be reused for story purposes.

This parallels a TV show which has a number of sets which are reused week by week as the core places the story takes place. A Location in FreeSpeak is analogous to a set in a TV show, with the same advantages in terms of cost-effective reuse of resources.

Once the initial investment in the creation of new tools, and the corresponding skills required to operate those tools effectively, has been overcome, the advantages manifest. The approach promises an extremely expressive and robust approach to interactive plots with a degree of freedom and level of replay value (exploring different renditions of the story) which would exceed anything we have currently seen in games.

Since its inception, developments in interactive storytelling have been gradual to say the least, although it is quite possible that a dynamic object oriented narrative system such as FreeSpeak will revolutionise the approach and create a new breed of games built around an advanced interactive plot.

Interactive Characters

The principle behind interactive characters is to model personality in a manner which is simple and efficient. As with interactive plots, the underlying structure tends to be simplistic with the effect being largely illusionary, although advances in the principles of interactive narrative promise ever more complex representations in the future. Most games limit the interactivity of characters to very simplistic behaviour - often nothing more than basic, fixed conversation, but there are a number of ways in which the players actions can be allowed to affect the attitudes of other characters.

Note that the characters in a game may change attitudes to the player as a result of the plot, but if the player has no control in this change (if it is dictated as an inevitable consequence of the story progression) it cannot be considered to be a case of interactive characters.

Flags

The simplest implementation for dynamic personality is using flags. A flag is a boolean variable - a type of data which can be only true, or false. Certain player actions cause the flag to become set (true) or unset (false) which affect the apparent behaviour or attitudes of the non-player characters, perhaps affecting what they say, or where they can be found in the game world, or even whether or not they help the player.

Discworld Noir used a flag representation for some personality behaviour; certain characters would respond differently according to their prior conversations with the player; some would be more or less obstructionist in their approach, according to what the player had said to them before. The underlying behaviour was very simplistic, however, and largely affected only the dialogue. Nonetheless, the illusion that the players choices were affecting character attitudes could be seen in several places as a result of a very basic use of personality flags.

We can explore the basic principles in a hypothetical game drawing on the Star Wars mythos. Consider a game based around the events of Episode IV: A New Hope, and the question of how the finale will take place. Consider the question:

Does Han Solo bail out Luke at the Death Star?

To model the behaviour, we can look at any pivotal actions that could occur in the game space or story space that might affect Han's chances of coming to Luke's aid in the Death Star battle. This might include:

Luke and Han bond on the Millennium Falcon
Luke rescues Han in a battle
Luke accidentally hits Han in the foot with his light sabre

The first two flags would predispose Han to coming to Luke's aid, whilst the third might well discourage him from bailing out the little whiner. We would describe the logical condition that determines whether or not Han comes to the rescue as follows:

If (Bond OR Rescue) AND NOT Sabre Then
Saves Luke

Therefore, if Luke and Han bond and Luke doesn't have a clumsy accident while practicing his Jedi skills, he'll ride in like the cavalry at the end. Similarly if Luke ever rescues Han, or if there is bonding and rescuing in the early part of the story, Luke can count on not getting his buttocks burned by Vader's blasters. If Luke and Han never bond, and no rescue takes place, Han doesn't feel the motivation to come and risk his life (and his precious ship) in a rather treacherous battlezone and takes his money to go and pay off Jabba. Similarly, if Luke gets carried away against the remote and accidentally slices off Han's big toe, Han isn't going to feel inclined to fly out of the sun whatever happens and Luke will find himself facing down his dad on his own.

The resulting behaviour isn't very complex, but the player at least experiences some connection between what they did earlier in the game, and how the final scenes play out.

Affinity Systems

A more complex representation than flags uses numerical values to track the relative feelings of one character towards another. These affinity systems (or affinity-aversion systems) were created by game designer Paul Kidd in his tabletop role-playing game Albedo (1988), based upon the 'Erma Felna, EDF' comic stories by Steve Gallaci. Although the mechanics were somewhat superfluous in tabletop RPGs (since the players and GM were capable of making their own decisions regarding their character's attitudes to other characters), they have proved to have interesting applications in computer games.

Although it does not use them for player interactions with characters, Sick Puppies' Ghost Master (2003) used an affinity layer in its AI system for the mortals the player's ghost team spooks and scares. This layer modeled the relationships between different mortals such that they made their decisions as to where they felt safe (or who they wanted to avoid). Unfortunately, much of this behaviour was hidden by more direct behaviour - such as the mortals running screaming from the haunting, or fainting. However, it is possible to see mortals going to certain rooms that they feel safe in and a few other behaviours originating in the affinity layer.

Our previous Star Wars example can be used to demonstrate an affinity system in action. In this case we have an affinity value from Han to Luke that starts at neutral (0) and goes up or down certain amounts in response to the player's activities. For example:

Initial State:
Affinity (Han, Luke) = 0
Luke and Han bond:
+ 5 Affinity (Han, Luke)
Each time Luke rescues Han:
+ 10 Affinity (Han, Luke)
Each time Luke accidentally hits Han in the foot with his light sabre:
-3 Affinity (Han, Luke)

A possible condition to test whether or not Han acts the hero in the finale might be:

If Affinity (Han, Luke) > 5 Then
Saves Luke

Under this system, if Luke and Han bond, Luke can get away with hitting him in the foot once, but if he's reduced to eight toes, Luke is toast in the trench. And rescuing Han would cover Luke from another three accidents or so. Of course this representation is farcical, but it represents the basic principles behind an affinity system.

However, now we are using an affinity system, we do not have to focus on a single outcome. If an affinity of +5 is enough to get him to take a small risk by popping into the battle for about five seconds, a larger affinity might persuade him to play the real hero and get in the battle from the start:

    If Affinity (Han, Luke) > 20 Then
        Fights in Death Star Battle
    Else If Affinity (Han, Luke) > 5 Then
        Saves Luke

And let's face it: there has to be limit to the number of digits Han will allow Luke to accidentally sever:

    If Affinity (Han, Luke) > 20 Then
        Fights in Death Star Battle
    Else If Affinity (Han, Luke) > 5 Then
        Saves Luke
    Else If Affinity (Han, Luke) < -30 Then
        Shoots Luke Dead

Under this condition, if Luke manages to slice off ten or more of Han's digits, Han will shoot Luke dead on the spot (assuming he hasn't just lost both hands). If Luke had rescued him a couple of times, the number of amputations might have to be higher - if they bond and Luke rescues Han twice earlier in the story, Han still has one finger left to pull the trigger on his blaster and reduce the irritating farm boy to Bantha fodder.

The idea of affinity systems is being developed in FreeSpeak as a means of providing case-based dialogue, such that the dialogue for non-player characters varies in accordance to the player's actions. The approach should allow for a greater feeling of immersion at very little cost. Additionally, plot branches can occur on the basis of character affinities within the manageable framework that the FreeSpeak system provides, a technique which has not yet been seen in games.

Algorithms

The basis of algorithmic interactive characters is to use a mathematical model to represent character attitudes and emotions. A distinction should be made between AI algorithms for the behaviour of characters, and algorithms for representing their personality and attitudes. The former is ubiquitous in current games, but the behaviour is usually based entirely around goal-oriented activities (find the player, defeat the player in battle). What we are concerned about in this section is means of reflecting the emotional world of characters mathematically.

The simplest algorithmic representation may be a probabilistic approach. Concluding our comparisons in the Star Wars milieu, consider this formula:

    P(Save) =             Rescues
                    _______________
                    Rescues + Wounds

In this formula, the probability that Han will save Luke at the end of the story is equal to the number of times Luke rescues Han divided by the number of rescues plus the number of light sabre wounds. So if Luke never rescues Han, the probability is 0 (it won't happen). If he rescues Han and never injures him, the probability is 1 (it will definitely happen). If there are two rescues, but one wound, the probability is 0.666 (66% chance); if there is one rescue and nine wounds, the probability is 0.1 (10% chance). Again, this is vastly simplistic, but it demonstrates the basic principle.

The trouble with probabilistic behaviour in games is that if it only happens once or twice, it seems to be essentially random. For example, in the last example, a fingerless Han might still come to rescue Luke because there is still a random chance - but the player has no appreciation of the hidden random system that gave that result. For this reason, probabilistic behaviour is best used when it is going to be spread across a large number of events or entities, where the probabilities can be seen by the player as proportions of a reacting group. This also makes it largely unsuitable for the behaviour of individuals.

There are more complex approaches, such as International Hobo's PsiScape, which models the emotional relationships between characters using a field model (as if the emotional connections were like a magnetic or gravitational field). Under this system, one looks for attractive and repulsive forces and overlays the different fields in order to determine behaviour - individuals therefore move like iron filings in an ever-changing magnetic field.

A basic tech demo for PsiScape mapped the behaviour of warriors on a battlefield using a field which represent the balance between an individual's fear and duty. Anything that an individual would find intimidating was a repulsive force, anything that an individual would feel safe near was an attractive force, whilst an attractive force representing their sense of duty pulled them towards the enemy. The scale at which an individual was affected by these forces was specified by a number of variables which represented the personality of that individual. The numbers varied as a result of actions in the battle - so a hero that slew a powerful foe would suddenly be a source of fear for the allies of the slain leader.

The result was a system in which the warriors moved in battle based on their emotional state. Cowards grouped with their allies, and avoided tough enemies, whilst heroic characters dived into the fray and - provided they were able to live long enough - eventually terrified the enemy enough to force them to withdraw. The effects of landscape could be added by a static vector field that would be superimposed (although complex battlefields require some additional mechanics to become tenable).

In principle, dialogue could be generated by considering the strongest forces acting on an individual: if their direction of movement was governed by fear of a strong foe, their dialogue could reflect their cowardly attitude (or reference the person they were running from), whilst a character who was trapped between two strong forces - fear pushing away, but duty just pulling them forward - could make dashing gung ho comments as they bury themselves in the fray.

The PsiScape principle could be applied to interactive characters at a different scale. Imagine a world in which different characters could be affected by a number of different fields. Greed, fear, loyalty, fraternity, love; territoriality; all could be represented by a PsiScape field, and each character would respond to some of the fields. Their movements in the world could be governed by the balance of fields, just as the position on the battlefield was determined in the PsiScape tech demo. Imagine a merchant motivated by greed and fear alone. If you encounter them in a road, she can comment that she is heading to the town ahead because of they've struck gold (which would place a source of greed in the town, which would be the strongest force acting on the merchant). Similarly, a refugee might be found in a shanty town because invaders have taken over his village. Since fear of the invaders was the force that overpowered the strong territorial force to stay in the village, he might comment about his hatred for the invading force.

This field-driven approach to interactive characters has considerable potential, but is not yet at a state whereby we can expect it to appear in computer games any time soon. Nonetheless, it shows the extent to which algorithmic character behaviour might impact on interactive storytelling in the future. It is likely that lessons learnt in interactive plots and affinity-based dialogue systems will strongly influence the future of algorithmic characters.

Making Interactive Stories Manageable

In order to safely undertake interactive stories, the game designer (or story designer) must take into account the needs of everyone in the development team with respect to the story design. There are specific vulnerabilities introduced by choosing to go beyond a linear plot and cardboard characters, and the team must be prepared for the problems and pitfalls, as well as steps needed to maximise the benefits of introducing an interactive story.

The production process needs to take into account several shifts in the usual needs of a project:

More Pre-production: It is important that the plot skeletons for the story are set down on paper before the programmers and artists break ground on anything other than research or prototyping. Similarly, if interactive characters are to be used, the parametric design should be paper-tested or prototyped before the main game is constructed.
This adds to the pre-production phase, although it need not add much cost of development as only the design (and dialogue scripting) team is needed for this pre-production.
Early Script Delivery: The usual system for script delivery is to only write the story materials after the game is already 50-90% complete). This is bad practice for most games, since the story is best developed in parallel with the design, but it is not acceptable for interactive stories. The dialogue script should be delivered as soon as is humanly possible.
QA Should Start Earlier: Ideally, the plot skeleton should be implementable so that the Quality Assurance team can begin testing it as soon as possible. If the plot skeleton exists only on paper until most of the game has been implemented, there may be insufficient time to debug the plot. This will add to the cost of development slightly, but only a small QA team is needed to test a plot skeleton.
Tweaking Time Desirable: If there are any parametric elements to the interactive story (especially if there are algorithmic characters) the team should have an opportunity to adjust parametrics in the latter stages of development. Every project would like to have this 'tweaking time', but it is more or less essential for interactive stories.

Because of the extra costs accrued in pre-production, QA and in tweaking time, interactive stories will for the time being be constrained to big-budget projects (which can afford extra expense) and 'art house games' (those with production values far below the mass market expectations). For most projects, which are on a tight budget for both time and money, any introduction of interactive story elements must be handled with great care.

Once the decision to include these elements, however, certain steps will minimise some of the typical problems:

Ensure multiple people know plot skeleton: It is vital that more than one person in the development team is familiar with the plot skeleton of the story. Not only will this improve the team's overall flexibility within the framework of the story, but it will limit the risk of severe delays in the event of loss of key personnel (if only one person knows the plot skeleton and they quit, the game will be in big trouble!)
Make sure all uses of locations known: The game designer (or story designer) must ensure that the producer (or game designer) provides a summary of all the times a location is used, and for what reason. Interactive stories are likely to reuse locations, and the artists and programmers must know everything that a location will be used for if they are to do their jobs to the best of their abilities.

In addition to avoiding the pitfalls, certain steps will aid in maximising the benefits of an interactive story:

Record and report all incidentally generated sub-locations: When artists work at implementing what is required for a location, they usually create some 'sub-locations' from their own artistic sensibilities in order to make the location look right. These extra locations are effectively 'free', and it may be worth taking advantage of them. It might be possible to use these instead of a different location, or to create a new scene.
Record dialogue as late as possible: To provide maximum freedom for minor plot restructuring and additional scenes, translate and record dialogue at a late point in the project. A good rule of thumb for all computer games scripts is 'write early, record late'. Try to get the script written as soon as the main details have been locked down, but hold off recording with voice actors (and translating to other languages) until the latest, safe point in the development cycle.

Finally, it is worth looking at a few issues that affect marketing:

Hype new technology: Take advantage of any non-linearity or interactive narrative elements as PR opportunities. Few games have them, so make sure that marketing know that this game has genuine unique selling points (USPs), rather than fairly familiar selling points.
Educate the Audience: It is important to ensure that the appropriate target demographics are aware of their freedom to choose in the game. As above, this will act as a selling point, but also it will improve word of mouth promotion after game is released which can make a big difference in the long term sales potential of the game, and of any brand that it creates.

With a little forethought and care, interactive stories can be integrated into a game project with minimal overall cost or risk, but as ever the games designer must pay attention to the many different participants in the game development process if they are going to do their job to the best of their ability.

Conclusion

We have seen how the field of interactive storytelling has grown out of the table top role-playing games of the 1970's and onwards, and discussed two ways in which interactive storytelling is likely to become more widespread over the next few decades.

Interactive plots offer the potential for genuinely non-linear narratives, although care must be taken that the design work that goes into creating the interactive plot is felt by the player. It is easy to put in details that only the games designer will appreciate, and this is especially true if an interactive plot is used in a long game: few players will replay it, so the interactive elements will be largely invisible (although the player may enjoy the game more because more of their wishes and desires are covered by the game as allowable courses of action).

Dynamic object oriented narratives may provide a means to explore interactive plots in a robust and cost-effective manner. If they deliver all that is hoped, players will experience a new breed of games in which their choices are reflected in the development of the story without the game requiring a lot of 'unseen' material being developed.

Interactive characters are still at a simplistic stage, but affinity systems promise interesting dialogue design possibilities in the near future, and field-based approaches (or other procedural algorithms) may in themselves present new ways of structuring game stories. It may be possible to create a dynamic world in which stories create themselves from the interactive materials.

The first games to successfully implement and market high-order interactive storytelling will almost certainly found new genres and acquire brand dominance of that genre - this is the mass market prize for interactive storytelling. For the time being, however, we are likely to see interactive stories emerging from the fringes of the games industry in niche products and 'art house games'.

Given the attachment to stories that humans display, and the sense of satisfaction engendered by a feeling of involvement in a story as it develops, it seems likely that interactive storytelling will become an increasingly important aspect of narrative game design over the next two decades. The time may come when even the mass market appreciates the advantage of giving players the ability to make meaningful choices within a game world.

Based upon the presentation 'Interactive Storytelling in Games: What it is, Why we need it, and How you do it' presented by International Hobo at Digital Media World 2001, November 13th 2001.