Chris Crawford is one of the great curmudgeons in the game industry. He is one of the notable pioneers in the early Atari days, and is known for his approach to interaction in games. More recently, he has been interested in expanding interactivity into territory that, in his opinion, reaches out and beyond the scope of games. The new model that he proposes is interactive storytelling, and he has been working in relative isolation for quite some time to develop Storytron, which is an interactive storytelling platform. Crawford’s work is alternately hailed as messianic, derided as unusable, or simply considered impossible. This text describes many of the ideas that would later go into Storytron. My intention in reading it is to relate his findings with my own discoveries on situated models of behavior.
Fundamentally his work on interactive storytelling is very different from my project. My project is on adaptation of narrative worlds into games. While my desire to hang onto the term “game” may be considered unusual, after all, many consider The Sims to not be a game, I do depart from the position of focusing on stories and storytelling. The process of storytelling (or being told a story) is an interesting and rich process, it is not the same as being in a world. This is a small difference, a shift in emphasis, but it is fundamental and informs the disagreements that I have with some of his conclusions.
Story
The opening poses stories as important because of their cognitive role. Crawford poses the idea of cognitive meshes, which are collections of ideas bound together through associations. This is a reduced and somewhat vague portrayal of cognitive science, but it is not inaccurate. These meshes represent an individual’s understanding of the world, what I would call a model. The concern is how learning occurs. Gradually the structure of these meshes is revised and integrated into “cleaner” forms as learning occurs. An “Aha!” moment will occur when new ideas are introduced and suddenly connect to multiple associations within the network. This sort of moment is characteristic of what happens during interaction. Stories are seen as these idea meshes. The act of reading views a mesh (the model of the story world), but does not engage with it. Interaction presents the mesh in many ways in attempts to elicit that moment of integration.
According to Crawford, stories are about: people, conflict, and choices. In these last two elements, stories are very much like games. Stories however are not about puzzles, visual thinking, or spatial thinking. These simply are not relevant. Both stories and games use spectacle, but stories need more than spectacle. Stories occur on stages, not on maps. An interesting thing to note about this set of claims is that the model being presented strongly resembles Aristotelian drama. I think that in general, stories are looser than drama (Aristotelian especially), but this is an interesting point, and relates back to interaction later (via Laurel).
Interactivity
Crawford’s definition of interactivity is a key point: “A cyclic process between two or more active agents in which each agent alternately listens, thinks, and speaks.” (p. 29) This definition is actually somewhat subtle, though. The key is not the extent of any one of these three phases, but rather how integrated they are. Interactivity is powerful because it compares multiple different networks of meaning.
The role of interactivity within artistic products (especially storytelling) leads to a conflict over artistic agency. The matter is who has control over the experience. An artist will want to have lots of agency over the experience and the final product, but that comes directly at the cost of the user to interact with it. This sort of dilemma is expounded in Mateas’s Semiotic Considerations.
In terms of aesthetics of interactivity, Crawford considers three factors: speed, depth, and choice. The first two are not all that interesting. Speed represents the degree of responsiveness. Depth suggests using subject material that is important and meaningful to the user. The real point of fruition in interactivity comes in the matter of choice. This has two elements: the significance of the choices, and the perceived completeness. Essentially, the choices available must satisfy the needs of the user, be meaningful within the story world, and also be seen as complete. The matter of choice is dead on. However, it also applies to the AI controlled agents themselves. They have choices of what to do which fall under the same categories of constraints.
Interactive Storytelling
There are some general but worthwhile observations. Crawford claims that interactive storytelling is not simply games with stories, or interactivized movies. The class of games described is fairly limited: “A form of interactive entertainment involving simple and/or violent themes, relying heavily on cosmetic factors, in which players must exercise precise hand-eye coordination, puzzle solution, and resource management skills.” (p. 46) This description is meant to describe games that are commonly played. This isn’t quite complete, because games such as The Sims, the best selling series of PC titles of all time, does not exactly fall within this category. However, the category does describe the wide range of mainstream games. Games may have stories as what Jesper Juul calls fictions, but the interaction in the game is not interaction with the story itself.
Similarly, with film and other linear media, it does not make sense to overlay interactivity on top. Crawford gives the example of Star Wars, where the hero is presented with six choices (for example, “Rescue Princess Leia?” “Run away from Darth Vader?” “Trust the foce to blow up the Death Star?”), but all of the choices are dramatically required. An experience giving the “player” options to refuse these decisions would not be enriching the possible story outcomes. In order to have stories that are interactive a different approach must be used. There is a conflict between plot and interaction. This conflict is dramatically compared with the conflict between free will and determinism in theology.
The reconciliation in this problem is to abstract out the space of possibilities. Some elements of dramatic structure maybe must be preserved, but other elements may be flexible. The essence of this, although Crawford does not make a big deal of the idea, is that this requires a view of the problem as constructing a story world. The example he gives is of the world of Arthurian myths. This is not a single narrative, but rather, it is a space where many narratives may occur. The Arthurian world has many different dramatic themes, so it may be possible to have one story in this world that focuses on one theme but not another. “This is the key to creating interactive storyworlds: multiple but connected themes. An interactive storyworld must present the possibility of romance, betrayal, battle, spiritual growth, and many other possibilities.” (p. 56) Crawford explicitly advises against considering a specific story as a subject for interactivity, because the themes and possibilities are limited. This is an interesting point in comparison to adaptation of Pride and Prejudice, because it reflects the fact that a storyworld is necessarily underspecified by a single story. One way around this critique is to compare aesthetic goals: story richness versus sense of presence. It may also be argued that the storyworld for Austen extends through the rest of her works, which is consistent with the culture of adaptation.
Two Cultures, No Hits, No Runs
Crawford argues that one problem with the development of interactive storytelling is the cultural divide between those who produce technology and those who would compose stories. There is a problem with the technological culture which is actively hostile to literature, and the artistic community is hostile to procedural thinking. The picture he paints is general and stereotypical, but is indicative of a widespread cultural gap. Gaps exist between theory and practitioners, as well as between artists and programmers. I believe that this difference is artificial, and is created by the values and practices of these communities. Programmers are taught the value of conciseness, correctness, and have the aesthetic of completeness and mastery, this is informed by the entire discipline of practice. Writers and artists develop values of ambiguity, incompleteness, and interconnectedness. These values are oppositional, compounding problems between the cultures. My approach I think reconciles the two: both code and expressive works are systematic, and these systems themselves have properties and embedded values. They may be observed functionally, but also artistically.
Verb Thinking
Crawford is known for his perspective on verbs. Systems may be understood in terms of verbs and nouns. Data and content is a matter of diversity of nouns, but the range of expressivity is dependent on the verbs in the system. The cardinal question of game development is “what does the player do?” and any answer to this question takes the form of verbs. Processes require verb thinking, and interactivity is dependent on processes. Verb thinking when applied to objects perceives them in the context of affordances, rather than properties. At one level, these are isomorphic, but choosing to look through from the perspective of verbs is important in the process of thinking about games.
Above all, try to think about things in terms of what things do, not what they are. A window is not glass;Â it’s something that blocks air movement while permitting light to pass freely. A car is not an engine, a body, seats, and so forth. It’s something that moves; everything else is subsidiary. A computer is not a box with a whirring fan; it’s a processing machine. A pill is not a bundle of exotic chemicals; it’s something that alters the biochemistry of your living processes. (p. 101)
Verbs tie into systems of causality. The goal for the player is to understand that system of causality. This connects to the discussion of models discussed earlier. The act of reading allows the reader to witness and observe the model described in the text. This model is witnessed through causal relations, and verbs must be used to elicit these causal processes. The language for understanding these relationships must be mathematical in nature.
Crawford gives an example of pseudocode which would be used to represent how agents might make choices within a storyworld. These are presented only as examples, not as a correct model. I would argue that these sorts of situations require world-based meaning derived from the context, that meaning is situationally derived. However, this model does not preclude that condition, but it seems more awkward.
competingForce1 = Loyalty[Darth, Emperor] + SelfInterest[Darth] – Idealism[Darth]
competingForce2 = Love[Darth, Luke] + Empathy[Darth]
if(competingForce1 > competingForce2)
then WatchLukeDie
else TurnAgainstEmperor
(p. 108)
One reason why I find this problematic is that the magnitude of the competing forces is important to the character. The plot does depend on the final decision, but what is important from the perspective of character is why that decision was made. In the Star Wars example, the moment is also dramatically significant because it is the first visible moment of Darth actually expressing care for Luke. (I am not that familiar with the lore, so that claim may be off) It is a dramatic reversal of what we have come to expect from his character. Eventually, the characters do need to decide, but the moment before that decision, where the audience can see visible signs of anxiety and hesitation are significant as well.
Crawford does address a few arguments about the quantification of characters. There is a somewhat natural reaction in the literary community about the problems with quantization of human characteristics. Crawford puts three counterpoints to this argument. The first is that in drama, simplification is necessary. Dramatic characters are simplified from real characters. The second argument is that characteristic attributes, such as love or loyalty are things that one can have more or less of. These are not normally considered to be numeric, but they are comparable at least at some level. The last argument is that while love and loyalty might me multidimensional concepts, the expression of those concepts must fall under some specific dimensions for the purpose of the story world. The dimensions that are relevant are related to the artistic content of the story world.
Language-Based Strategies
In the preceding chapters, Crawford discusses some approaches to interactive storytelling. These examples are branching structures, environmental strategies, data driven strategies (grammars). Finally, he settles on language based strategies as the ideal solution for these problems. This is the key element where I disagree with Crawford’s approach, but the strategies are particular to his aesthetic goals. Essentially, interaction takes the form of writing- composing sentences within a computer based language. Effective interactivity does require a robust system of communication. The idea is to create a language, a sort of creole which is capable of effectively being constructed and understood on a computer. An early approach was the game Siboot, which uses a visual 2d language. Crawford argues that the primary way for enabling depth of interaction is to provide player’s with access to the language of what they can do. I actually think that games like The Sims, which provide menu and object driven approaches to interaction is quite effective, as an alternative to this.
Personality Models
Crawford’s goal is to create a model of personality which is small and tight. Essentially, if the character is tight, then the plot may be flexible. The important elements in developing character models are completeness, conciseness, orthogonality, and connection to behavior. These depend on the dramatic and expressive structure of the world. The model must be complete to represent everything that needs to be represented, concise enough to keep in mind, orthogonal in the sense that the variables have correct dimensionality.
The personality model that Crawford uses is broken into several variables: intrinsic attributes, mood, volatility, accordances, and relationships. The intrinsic variables are listed as follows: (p. 190-192)
- Integrity: Characters with high integrity will keep one’s word, not lie or reveal secrets. Low value characters will do the opposite.
- Virtue: A high virtue character will take others’ needs and desires into account while making decisions. Low virtue characters are selfish.
- Power: This is the ability of the character to wreak injury upon others in physical, financial, social senses.
- Intelligence: High intelligence enables characters to make effective and pragmatic choices.
- Attractiveness: This is an actor’s appearance, and essentially their charisma and desirability.
Mood variables. These seem to be based on basic emotions, which I have found to be problematic. The justification for these binaries is given in behavioral terms, but I disagree with the sense of opposition between anger and fear, for instance. The absolute magnitudes of these is important.
- Anger/Fear. I disagree with the opposition of these, but Crawford claims that they are sides of the same coin. These are equivalent to “fight or flight” in the sense of behavioral responses. Anger is different from hostility, and is presented here as a global emotional variable (of an individual), rather than something that is directed toward a particular individual.
- Joy/Sadness
- Arousal/Disgust. The consideration of arousal and disgust is tricky, since in other models, emotional arousal is associated with all moods. Crawford’s definition of arousal relates more strongly to the sexual dimension of arousal or heightened sensitivity in a positive fashion.
Volatility variables. These affect the rate at which the moods change, high values let characters moods change dramatically, while lower values mean that these values change more slowly. This relates to the stability of moods, so characters with low volatility values will seem to be relatively set in their moods. This does not seem accurate in the case of some fictional characters who are generally predisposed to a particular mood- self disgust, depression, perpetual anger. These characters could reach their alternate extremes, but fictionally they tend to return back to their predispositions. Although, this could be ameliorated and incorporated into the model by inclusion of “predisposition” variables.
- Adrenaline: Controls the rate of change for anger/fear. I find this problematic because anger is something which dissipates and tends to fluctuate even when it is active.
- Manic/depressive: Controls the rate of change for joy and sadness.
- Sensuality: Controls the rate of change for arousal and disgust.
Accordance and relationship variables. These variables control how characters interpret the intrinsic variables of other characters. Accordance variables are used as the default for a character’s impression of a stranger’s intrinsic variables, while the relationship variables control how the character interprets the intrinsic variables of specific other actors. If character A needs to make a decision based on the Integrity of character B, then A’s interpretation of B’s Integrity = A.AccordIntegrity + A.PerIntegrity[B]. These follow the same templates of the intrinsic variables:
- AccordIntegrity is something like gullibility versus suspiciousness. PerIntegrity[other] is similar to trust. Characters with high values of these will tend to trust and confide in others, believing them to keep their word.
- AccordVirtue is how well the character might believe others are intrinsically good. Low values would cause the actor to believe the world is full of people who have its worst wishes at heart. PerVirtue[other] is how much the character believes the other is virtuous. Crawford suggests that this has a similar operation to the Virtue variable, and it may be effective to add Virtue and PerVirtue in calculations.
- AccordPower is how the character believes the rest of the world to be powerful, essentially a measure of timidity. A low value would yield overconfidence, essentially an overestimate of it’s own Power. PerPower[other] is something like awe or fear. Fear is inspired by PerPower minus PerVirtue, which would be how much the agent suspects the other of being both evil and powerful.
- AccordIntelligence represents how readily a character will defer to other character’s judgements, while low values will yield another sort of overconfidence in decisions. PerIntelligence[other] reflects something like respect, which can be positive or negative accordingly.
- AccordAttractive causes a character to see the world as either beautiful or disgusting. PerAttractive[other] causes the character to believe another is attractive, which may mutually affect PerVirtue. A character who is attracted to another may overestimate the virtue of the object of affection.
Drama Managers
There are some useful bits here: Crawford supports the idea of drama management. The emphasis is that the process of playing through an interactive story is a dramatic experience. Crawford does not build on theory of performance and drama all that strongly, but raises a few important points. The first is that the drama manager is a solid interactor in the sense that it listens to the actions of the player, thinks of potential responses, and manipulates the world for the phase of speaking. The second is that the drama manager must work to encourage the dramatic flow of the experience, rather than working to thwart the player.
There is a strong impulse in games and artistic works to force the player down the path that the designer thinks is the “best,” rather than allowing the player to interact effectively with the world. The problem with this is that it enables players to behave “unreasonably” within the world, working to thwart the game world. The response Crawford proposes is to allow the player to do any reasonable thing within the game world, but impose consequences (ostracism, death) for irresponsible behavior. An important example of this is in Mateas’s Facade, which will eject the player from the drama (in the form of Trip escorting him or her from the apartment) when not acting according to the dramatic rules.
The final idea proposed is to implement a scoring system to measure the effectiveness of a story. This seems problematic from the literary perspective, in the sense that it involves quantization of narrative value, but it does make sense at some level. There can be good or bad stories, in the sense that a good story might have positive results. The example given is in the Arthurian storyworld, where the score will be damaged if Mordred becomes king, improved if characters have mutual good feelings toward each other, and improved if Excalibur is returned. A legitimate critique of this is that it impedes the player’s ability to make their own meaning from narratives, restricting their openness in the Eco sense. If a story will necessarily be judged, then that means that the player “ought” to reach a good score. That means that there will necessarily be some ideal solutions, and as such, the story world comes pre-evaluated.
A good counterexample for this, though, is that many literary works are tragedies. There is a concern about players committing imaginary suicide, however the dramatic value for tragedy is more significant. This interesting approach turns the normal survival and scoring instinct for gameplay on its head. Most games value self preservation, but tragedy, pre-evaluated though it may be, would value self sacrifice. The means for encouraging the player to commit virtual suicide is to shift the player’s perception of his or her role to that of a dramatic performer.
Verbs and Events
Crawford poses a model of interaction that works by the player constructing sentences. Verbs transform into events when executed. These interactions have a specific structure: It has a name, it has some import-how significant or noteworthy the event is, it requires some time to prepare, some time to execute, and has some sort of audience. Crawford also lists types of audiences that these events may have:
- MentalState, the action is entirely mental
- AnyAudience, the action does not have any requirements on the audience
- RequireWitness, the action must be witnessed by someone
- SubjectOnly, the action must be carried out in secret
- SubjectAndDirectObjectOnly, both must have privacy and secrecy
- AllAudience, the event is global, and when it occurs, everyone knows
- etc…
HistoryBooks and Gossip
The key issue with the mechanic of these interactive stories is that characters must communicate, and NPCs must communicate with each other. Crawford poses a model of gossiping, where agents will communicate according to some algorithms. The model of communication relies on sharing events. The gossip model provided is dramatically motivated, rather than socially motivated. A key element of this is a mechanism for causing characters to lie and keep and share secrets. This section is somewhat underspecified, though.
