Game Development Reference
In-Depth Information
16
Variation in Choice
In the past few chapters, all of our examples have been solved by finding “the
best� answer to the problem. We built all of our utility functions to rate and
score the aspects of the decision, sorted the options by the score, and selected
the one with “the best� score (be that highest or lowest). Naturally, this sounds
strikingly like normative (or prescriptive) decision theory. In fact, if we recast it in
the vernacular we used in Chapter 4, we were determining what we
should
do.
Remember that normative decision theory assumes that our agent:
Has
all
of the relevant information available
Is able to
perceive
the information with the accuracy needed
Is able to perfectly
perform
all the calculations necessary to apply those facts
Is perfectly
rational
We spent a good deal of time in Part II of this topic showing examples of how
humans just don't do what is
perfectly rational.
On the other side of the spectrum,
descriptive decision theory tells us what people
elect
to do. More specifically, it reports
what they have
already
done. Because our data for how demons from the under-
world will counter the assault of a single space marine is somewhat lacking, we have
to construct our own models—which defies the survey-based definition of descrip-
tive decision theory.
Our goal, however, is to create behaviors that only
look
like they are born out
of the wide palette of descriptive decision theory's data. The first step in this direc-
tion is
away
from the single “best answer� that normative decision theory gives.
That is, we are looking for
variation
in our decisions. How do we create this varia-
tion without data from which to work? To find the answer, let's first examine why
we want variation in the first place.
417