Interactive Storytelling Tools for Writers

The starting point for any dramatic interaction is a personality model. This is the Object (data) side of the Processes (algorithms) that determine dramatic behavior. I’ll discuss those algorithms in the next lesson

What is a model?
Modeling is the mathematical version of artistic expression. To put it another way, the Mona Lisa is a painted model of a human character. It does not show the entirety of that woman’s personality; instead, it beautifully expresses one facet of human personality. In the same fashion, a novel like Huckleberry Finn does not give us a complete representation of Huckleberry Finn or of the society in which he lived; instead, it presents a model of that society and of the reaction of an innocent boy to the evils of that society. A movie like Star Wars offers a model of the path of a young man as he faces the challenges of life. All artistic expressions are models.

A model is a simplification of reality that neatly communicates one important aspect of reality. A model does not claim to be completely accurate; its goal is to clearly communicate some important idea, and clarity is obtained only by stripping away details that are not part of the idea. 

Let’s model the earth. We start by saying that the earth is round; more technically, we say that it is a sphere. That’s our first model. But it turns out that, because the earth is rotating, the centrifugal force causes the equatorial regions to stick out a bit, so the earth is actually an oblate spheroid. That’s a better model of the earth. But we can make a more accurate model by tracking the motions of satellites; it turns out that, because the mass of the earth deep inside is unevenly distributed, the earth is actually kind of lumpy. So that’s a more accurate model.

These images are greatly exaggerated; from space, the earth looks like a sphere. Of course, we could make a model that takes into account the mountain ranges and valleys — that’s even more accurate. 

I suppose that I could have included the ancient model in which the earth is flat. That’s a model, too. It’s wrong, but then, isn’t the spherical model wrong, too? The oblate spheroid model is also wrong. Even the lumpy pear model is wrong; it misses some fine details, like the fact that the shape of the earth changes slightly because of the gravitational force of the moon. 

Here’s a Big Idea: truth is infinite. Nothing you can say is ever completely true; every statement you make is a simplification of the truth. In other words, EVERYTHING we know about reality is a model! Completeness is not the ideal of communication; that honor goes to conciseness. The quality of the truth we communicate is entirely dependent upon the goal of the communication. The only situations in which you need to know that the earth is not a sphere are those involving sending rockets into orbit; for all other purposes, it’s best to say that the earth is a sphere. 

What is the value of π? Well, it’s 3.14, right? Well, no, it’s actually 3.14159. Uh, really, it’s 3.141592653589793238. To be strictly correct, it’s infinitely long. So what’s the correct value of π?

There is no “correct value” for anything! That’s because there is nothing that is completely, perfectly knowable. EVERYTHING we know is an approximation, a simplification— a model. 

Here’s where Object and Process combine to create meaning. Just as a noun (Object) and a verb (Process) combine to make a meaningful expression (a sentence), models as objects combine with algorithms as processes to create meaning. It is the union of the two that generates meaning. Hence, the “correctness” of a model depends entirely upon the algorithms applied to the model. 

For example, if you’re calculating the surface area of a circular cake because you need to know how much icing you’ll need, you can just use 3 and get close enough. On the other hand, if you’re calculating the surface area of a lens, you might want to use 3.14. And for very precise calculations of, say, the volume of the sun, it might be prudent to use 3.1416. Here’s the important idea: the model needs to be only as good as the algorithms that will be applied to it. You can never achieve perfection, so you must always decide how much precision you need for the task at hand.

Human beings are immensely complex creatures, so we apply all sorts of different models in working with them. For example, let’s consider Mr. Barack Obama. Who is he? That’s a stupid question, because it has no operational significance. Instead, we must ask questions that reflect our models. Photographs, for example, present different models of the man:

Which is he? He’s ALL of these things. No single model, be it President, Husband, Thinker, or Father captures the complexity of the man. We can also apply other models: biometric models, such as height, weight, and so on. Or we could use a financial model listing his assets and his liabilities. We could use an educational model presenting the various degrees he has earned. Or a career model presenting his job history. A biographer would attempt to capture all of the most important models describing the man. Yet we know that even a thousand-page biography would never capture the truth of the man.

Many times the algorithms we apply when using a model are unconscious. For example, in assessing the photographs, we use complex algorithms about human nature to process the meaning of a face. Look at the big, broad grin in his “President” image. Your facial analysis algorithms tell you that this is a friendly guy, confident and good-hearted. The Husband image shows his love for Michele through the expression on her face. Our facial analysis algorithms tell us that she is confident of his love; therefore, he must love her very much. The Thinker image is easily recognized by our facial analysis algorithms to show that he is a serious and intense thinker. Lastly, the Father image shows, through the face of his daughter, that she knows how much he loves her. 

We don’t just randomly pile on models. Instead, we start with a question, and then find the model that best answers the question. If you’re a bank manager considering an application for a loan, you’ll look at his financial model. If you want to learn what he knows, you’ll look at his educational history and his personal experience. If you want to learn how he thinks, you’ll want to read books that he has written. Each of these is one model of a complex phenomenon. 

The key idea here is that models are not mean to capture truth in its entirety. A model is something that we USE to answer a question, and we answer that question by applying processes (algorithms) to the data in the model.

Modeling Personality
The models you’ll be creating in the computer will consist of a group of personality traits. Those personality traits are variables: numbers that are different for each character. The task in creating a personality model is to create the ideal set of personality traits. The creation of that set is governed by three ideals:

1. Operational utility
The biggest mistake that people make in designing personality models is to think in terms of Object rather than Process. They concentrate on what a personality IS rather than what a personality DOES. For example, some people will rightly judge that lustful feelings are a part of every personality, so they’ll include a trait for this in their personality model. For some storyworlds, this would be a useful addition, but there are also plenty of story worlds in which no sexual activity is intended to take place. For such storyworlds, a trait for sexual desire is just excess baggage. 

The inclusion of a trait for lust is an obvious example of poor model design, but there are more subtle ways in which a personality trait might seem right and true, but in the actual operation of the story world, won’t be of much value. The purpose of a personality model is not to realistically define human personality; it is to be USED in making behavioral decisions for the characters. In other words, you want to include personality traits that you will likely use in calculating the decisions that your actors will be making. Thus, the personality model is defined by the verb list for the storyworld: you need personality traits that help your actors discriminate between verb options they face. Ergo, after first defining the verb list, you must ask yourself what kind of personality traits would be most useful in making choices between different verbs.

For example, suppose that your verb list includes these verbs that you judge will play an important role in your storyworld:

Express Affection Directly (“I like you.”)
Express Disaffection Directly (“I dislike you.”)
Express Affection Indirectly (“I like him.”)
Express Disaffection Indirectly (“I don’t like her.”)

What personality traits would be necessary to permit you to direct the actors to select the appropriate verb for any given situation? Obviously you need a personality trait expressing the notion of liking or disliking somebody. Let’s call it ‘affection’. 

But now we encounter one of the many problems plaguing us when we create personality models: is affection a component of romance, or friendship, or both? Isn't romantic love different from friendly affection? Brutal simplification is called for here. We must treat romantic love and friendly affection as essentially the same thing. You will, of course, object that romantic love is not at all the same as friendly affection. That’s irrelevant; what matters is whether the trait ‘affection’ will help actors decide among the verbs in the verb list. 

Now, if you wanted to include romantic interaction in your storyworld, then you would have populated the verb list with verbs just as ‘kiss’, ‘hold hands’, and ‘make love’. In that case, you truly would need a personality trait for romantic love as opposed to friendly affection. The key point here is that the personality model serves the verb list, NOT your notions of what comprises a realistic representation of human personality. Different storyworlds will need different personality models. 

2. Orthogonality
“Orthogonal” means is “mathematically independent”. It applies to a set of parameters and means that, if you change one of the parameters, none of the other parameters will change. A simple example comes from the directions of the compass. There are two orthogonal directions: north-south and east-west. If you take two steps to the east, that changes your east-west position but has no effect at all on your north-south position. 

Suppose, however, in a moment of bold perversity, you decide to use north-south and northwest-southeast as your compass directions. Under that model, if you take two steps north, you’ll change your north-south position but you will ALSO change your northwest-southeast position. That’s a stupid way to model space! We model space with 3 orthogonal parameters: latitude, longitude, and altitude. Moving in only one of these directions doesn’t change your location in either of the other two directions. Of course, moving in a combination of these directions, like an airplane, will produce changes in all those directions.

It’s difficult to achieve perfect othogonality in personality models, because so many personality traits overlap. Suppose, for example, that we had two personality traits: timid and honest. Those are fairly orthogonal. But would a timid person lie as readily as a dominating person? Here’s how to consider the problem:

1. How many people might plausibly be extremely timid and extremely honest?
2. How many people might plausibly be extremely dominating and extremely honest?
3. How many people might plausibly be extremely timid and extremely dishonest?
4. How many people might plausibly be extremely dominating and extremely dishonest?

If all four questions yield the same answer, then the two traits are orthogonal. Again, it is unlikely that you will ever find two traits that are perfectly orthogonal; human personality is nowhere near as neat and straight as space. You must instead strive for maximum orthogonality.

Orthogonality is important because it makes the process of designing behavioral algorithms easier. For example, suppose that your personality model includes two halfway-related traits: pride and dominance. You have a verb “insult”, and you have as possible consequences of that verb the verbs “remain silent”, “insult”, or “punch”. You must create algorithms that allow an actor to choose among these three options based on the actor’s personality traits. So, do you write the algorithm using pride or dominance? Is the reaction to an insult determined by a person’s pride or a person’s dominance? 

You have an easy answer: both. You’ll design an algorithm that simply adds both pride and dominance together to make the decision. That sounds good, but it won’t be long before you face a similar situation with another verb, and once again you’ll use the sum of both traits. After a while, you’ll notice that you seldom use either pride or dominance separately; most of your algorithms use the sum of the two. Doesn’t that suggest that they’re really pretty much the same thing? 

A counterargument would arise if you sometimes use them against each other. That is, if you’re always using the sum of the two, that’s bad, but if you also use the difference of the two, that’s good. In other words, if they always get used the same way in relation to each other, then they’re really parallel traits and should be collapsed into a single trait. If they are never used in the same way in relation to each other, that’s good. 

As always, most real problems are never so clearly black and white; in many cases you’ll end up with a personality model whose traits are mostly orthogonal but sometimes parallel. Therefore, you must design your personality model to achieve the maximum amount of orthogonality, admitting to yourself that you’ll never, ever hit upon a perfect model. 

Slippery semantics
Another problem that trips up many people is the confusion between what a word really means and you use it to mean in your algorithms. This would not be a problem if you had studied physics for years, because physicists have gotten good at using words to mean only what physicists use them to mean. Let’s talk about the word “force”, for example. You know what force is; you can look it up in the dictionary. Here’s the definition I found on the Internet:

 

For physicists, however, force means just one thing: the physical phenomenon that causes acceleration according the equation F = ma. They’re not saying that this is the only correct definition of force; rather, this is what they mean when they’re talking physics. When they’re talking about anything else, they use the normal definitions of force.

In the same manner, you must be able to define your own technical terms that do not mean what they normally mean, but instead mean something very specific inside your storyworld. “Love” means a lot of things in the real world, but inside your storyworld, “love” will have a narrow meaning that is defined by the way it is used in your mathematical algorithms. 

My Preferred Model
I have played around with personality models for over 30 years. If mistakes are the foundation of all learning, then I am undoubtedly the World’s Leading Authority on Personality Models. I boiled it down to three basic personality traits, but of late I have added a fourth. I have found it useful to label a personality trait by a pairing of its negative value and its positive value. This makes clearer what the variable quantifies. 

It is important to realize that these are operational variables, not words. You can argue all day long about the best words to use, but if you rely on your own interpretation of the meanings of the words, you’ll end up confused. I shall define how these personality traits are to be used. If you don’t like my words, call them whatever you want: rutabaga, orphanage, symbiosis, whatever word you like. Just make certain that you understand how they work in the personality model!

Bad_Good
This is easiest personality trait to understand. Synonyms include Nasty_Nice, Evil_Virtuous, BlackHat_WhiteHat, Villainous_Heroic, Contemptible_Noble. You use this variable to make decisions that hinge upon how nasty or nice a person is. For example, suppose that Fred slaps Tom. Tom has two possible options: he can pull out his Magnum .45 and blow Fred’s head off, or he can turn the other cheek. You would use Bad_Good to make the choice between options. A high value of Bad_Good would favor turning the other cheek; a low value would favor blowing Fred’s head off. 

Bad_Good is not a measure of competence or ability, speed or strength.

You’ll use this trait a great deal.

Faithless_Honest
This trait is also easy to understand: it’s just how much integrity a person has. This affects whether they tell the truth, honor their promises, and behave with integrity. This personality trait was important in some storyworlds I built, but I now recommend that beginning storyworld builders refrain from including behavior involving honesty versus dishonesty. The player will have difficulty discerning dishonest behavior. The other behavioral area of application of this trait is honoring promises made. Inasmuch as this requires deferred execution (“If you loan me the money to buy a hamburger today, I’ll repay you tomorrow.”), this is also something for beginning storyworld builders to avoid. 

Faithless_Honest is not a measure of knowledge, expertise, or intelligence.

Timid_Dominant
This is a more useful trait than Faithless_Honest, but it can be tricky to understand. Synonyms would be Submissive_Assertive, Compliant_ Overbearing, Servile_Pushy, Acquiescent_Self-Assured, or Obedient_Willful. This trait is used when some kind of conflict of wills arises between two characters. When a confrontation occurs, the character with the higher value of Timid_Dominant will prevail, all other factors being the same. 

Timid_Dominant does not include any component of courage or bravery.

Ascetic_Hedonistic
This reflects the degree to which a character pursues any form of sensual gratification. A person with a high value of this trait enjoys good food, likes to wear fancy clothes, desires to accumulate nice things, and is lustful. This trait is completely orthogonal to Bad_Good; there is no reason why a person with a high value of Ascetic_Hedonistic cannot also have a high value of Bad_Good. Similarly, there is no reason why a person with a low value of Ascetic_Hedonistic cannot also have a low value of Bad_Good; such a person might be, for example, an intolerant religious fanatic. Synonyms would be Austere_Sensual or Puritanical_Pleasure-Seeking. 

These four numbers define the personality of a character.

“Just four numbers?!?!!?? Humans are much more complex than that!!!!”

Yes, and the shape of the earth is much more complicated than a sphere, but you call it a sphere, don’t you? Remember, the important thing is that the complexity of the model must be appropriate for the task. I can assure you that this simple four-variable model will give you all the complexity you can handle. My experience has been that it is too complicated for many people to work with comfortably. With time and experience, we’ll be able to move on to more complex models. For now, stick with the tricycle. You can learn to ride the bicycle after you’ve gotten some experience.

In fact, the model does have more than just four variables, but all the other variables are in some way associated with these four. Once you learn to think of dramatic characters in terms of these four fundamental traits, the other variables come easily.

Perceived values
The first group of associated variables are other people’s perceived values of the personality traits. In other words, Bob may have a Bad_Good value of 0.5, but other people have different perceptions of how bad or good he is. Some people will think that he’s nicer than he really is; their perceived value for his Bad_Good might be 0.6 or 0.8. Other people will think that he’s worse than he really is; their perceived value for his Bad_Good might be 0.1 or 0.3. I call these the “perceived values”, or p-values for short. For example, Nancy’s pBad_Good for Bob might be 0.67, while Tom’s pBad_Good for Bob might 0.34.

These perceived values translate into relationships. The perceived value of Bad_Good represents the affection that the perceiver holds for the perceived. In other words, if Sally’s pBad_Good for Bob is 0.8, then Sally really likes Bob. On the other hand, if Joe’s pBad_Good for Bob is -0.8, then Joe hates Bob.

“But affection is not the same thing as perceived Bad_Good.”

That’s true, but it’s close enough. Sure, you could make all sorts of complicated relationships that are far more accurate.  But when you tried to actually use such a complicated model in your algorithms, you’d quickly get lost. I know, because I’ve seen this with my own work and with the work of others. Keep it simple! This is a model, not infinite truth. The way to evaluate it is to answer the question, “How well can I create algorithms with this model?” 

In the same fashion, pFaithless_Honest is trust. We have more difficulties figuring out what pTimid_Dominant means. My best guess is that it represents the degree of intimidation that the perceiver feels for the perceived. I must caution you that the Timid_Dominant personality trait is the most difficult trait to work with. If you want to be cautious, use just Bad_Good and Faithless_Honest.

Here’s a visual representation of pValues for Bad_Good:

These are the pBad_Good values that different actors have for the actor in the center: pBad_Good[somebody][You]

Circumferential Values
My experience is that we occasionally need to go one step further in indirection. Consider, for example, this statement: 

“You invited Tom and Liz to the party?!?! Don’t you know they hate each other?” 

This doesn’t occur often, but this kind of knowledge — how other people regard each other — does become significant at times. In keeping with the geometry of perceived relationships shown above, I call such relationships circumferential, because they circle around the actor at the center. Here’s a diagram showing what the actor “You” believes to be Liz’s perceptions of other actors:

I use this label to specify a particular relationship:

cBad_Good[You][Liz][Tom]

This number specifies what You believe to be pBad_Good[Amy][Tom]. Here’s a diagram I used in my 2013 version of Gossip:

This shows both the values of pBad_Good[You][Somebody] (the arrows radiating outwards from You) and the cBad_Good[You][Somebody][SomebodyElse] (all the other arrows). Note that this includes the arrows pointing inward towards You. These can be particularly important; it’s nice to know how other people think about you. The color of the arrow designates the nature of the feeling. Bluer means a negative value and redder means a positive value. In this diagram, Max and You like each other, as do Ella and Mort. Max and Mort hate each other, as do Max and Zoe. 

Accordance Values
There’s one more spinoff from the four basic parameters in this personality model: accordance, designated with the prefix ‘a’. This is the degree to which an actor tends to perceive higher values of a parameter. The easiest to understand is aFalse_Honest. This is the degree to which a person tends to trust other people. It corresponds to gullibility. An actor with a negative value of aFalse_Honest is suspicious of everybody. 

Similarly, an actor with a high value of aBad_Good likes everybody, while a person a negative value of aBad_Good is a misanthrope who sees everybody as an enemy. As always, aTimid_Dominant is a bit clumsy; it is not orthogonal to Timid_Dominant itself. That is, an actor who has a negative intrinsic value of Timid_Dominant will necessarily have a positive value of aTimid_Dominant. There’s no way around this; it’s a flaw in my personality model, plain and simple. Accordance for Ascetic_Hedonistic is .Perhaps the accordance of Ascetic_Hedonistic could be considered to be judgmental: that is, a person with negative Ascetic_Hedonistic would view everybody else as hedonistic, and a hedonistic person could view everybody else as ascetic. I don’t know. You decide.

Moods
I use three moods; they take both negative and positive values. They are:

Sad_Joyful

Fearful_Angry

Disgusted_Aroused

These are not arbitrarily chosen moods; they correspond (roughly) to hormones in the body. The clearest of these associations is the role of epinephrine and norepinephrine in the fight-or-flight response, although this is just the beginning of a long chain of hormonal responses. 

Serotonin is associated with happiness and its absence with depression. Disgusted_Aroused might seem like a weird combination but they do cancel each other out.

The processing of moods is straightforward. Moods normally have a value of 0.0. An event can suddenly push a mood away from zero. The magnitude of the mood change will be calculated from the event itself as well as the volatility of the character — a personality attribute of each character. Afterwards, the mood decays to zero. I have used both an exponential decay function and a linear decay function. I found that the linear decay function was a bit more appropriate in most dramatic situations, because the exponential decay fell off too quickly, but that may have been because I was using too large an exponent.