Our greatest responsibility in creating a good personality model is to simplify it. Any idiot can paint a photographically accurate portrait &emdash; but how many painters can capture the soul of a person in a portrait? Any clod can write a novel with excruciating detail and plodding thoroughness &emdash; but how many writers can capture the essence of their story in clean, clear prose? We have the same obligation to accept artistic responsibility for our model, and simplify it as much as possible. Thus, we want the smallest possible model. How small is small? I think that its size will be dictated by our estimate of the ability of the storybuilder to absorb our model. Remember, the theory behind the Erasmotron is already quite complicated; we need to guess just how much additional complexity we can saddle our storybuilder with.
It is entirely too easy to criticize a model for lacking some minor tidbit of personality, but the real test is not completeness; it is focus. If nobody criticizes our model for incompleteness, then we have surely put in too much stuff. The test of thrift is survivable poverty. Throwing too many variables at the problem is the coward's CYA approach.
What then are our criteria for including or excluding elements of the model? Here are some:
This is the notion that our personality model is not intended to be a true or correct model, but instead will be measured solely by its operational value. In other words, we really don't care whether our personality model is correct or incorrect; we care only that it is useful for our purposes in storybuilding. Stories constitute a special subset of human behavior, and while it is possible that stories could address almost any aspect of human behavior, the fact is that stories tend to concentrate on only a few aspects of human behavior. For example, I spend more time sitting on the toilet than making love, yet there are a lot more stories about lovemaking than defecation. Thus, a scientifically correct personality model would be certain to cover defecatory behaviors, but a dramatically interesting personality model really doesn't give a damn about them. We want a model that covers all the dramatically important stuff &emdash; and doesn't cover any of the dramatically insignificant stuff.
This is the emphasis on behavior rather than internal states. The idea here is to eschew long-winded theories about what's going on inside the character's head and just have the character do something. Yes, we cannot avoid some reference to mental activity &emdash; but we want to minimize such references, because they push us away from what is dramatically important. This is the interactive storytelling analogue of "Don't talk about it; show it!" My version is, "Don't make it an mental state; make it an action!" An example will help: Guy shouts at girlfriend. Girlfriend starts to cry. Wrong model: Guy feels guilty; what behaviors does guilt lead to? Right model: Guy must choose between Action#1, Action#2, and Action#3; use personality model to choose between them. If you want to have an intermediate and temporary variable that you call Guilt, fine, but just don't make it a permanent variable, OK? It's only value is in making the decision, so just make the damn decision!
A variable should be formalized into our system only if it retains ongoing operational value over a series of decisions. If its significance affects just one decision, then it should be part of an intermediate calculation, and thrown away.
Another way of viewing behaviorism is that emotions that are essentially reactions to events should not be included as state variables. In other words, if it's my party, and my boyfriend Johnny leaves with Judy, and they return with Judy wearing Johnny's ring, then am I jealous of Judy or am I jealous of Judy's wearing Johnny's ring? If the latter, then the emotional reaction is confined to the event itself and does not form a part of my permanent relationship with Judy. This is not to say that my permanent relationship does not change; we say that I am jealous of the event and accordingly less affectionate, submissive, or trustful of Judy. But do we need a permanent relationship called Jealousy?
This is really a means of maximizing simplicity. Orthogonal traits share nothing in common. Thus, affection is almost orthogonal to trust, but fear is not othogonal to respect. We must evaluate every single pair of relationships, every pair of traits, and every pair of moods, asking ourselves if this pair is truly orthogonal (or close to it). If the answer is negative, then we must look more closely to find the fundamental concept underlying these two derivative concepts.
Orthogonality implies that some factors will be built up from fundamental factors. For example, perhaps we could construct jealousy from pride and greed &emdash; but only wealth-jealousy. Perhaps we could construct another, more female kind of jealousy from a combination of pride and amorphagic. I'm not sure that these combinations are indeed correct ones, and so may not be good examples, but they illustrate the principle.
This is a set of five orthogonal attributes that supposedly span the space of human personality:
Expressive Style (quiet and restrained ---> enthusiastic and energetic)
Interpersonal Style (aloof and cruel ---> warm and generous)
Work Style (procrastination ---> dedication)
Emotional Style (moody ---> even-tempered)
Intellectual Style (traditional ---> skeptical or creative)
This doesn't impress me; where does sex come in? How about greed? How could this model be used to predict whether a character would betray a friend or remain loyal? This looks like something useful for employment tests to determine how well the applicant will fit into the workplace, not a useful model for use in interactive storytelling. Still, I wonder if there aren't some useful tidbits to extract from it?