Game Development Reference
In-Depth Information
… none of those are at all human, and therefore any decision based on that
model is bound to be flawed in its efforts to
look
human (or animal, etc.).
If we swap a few things into those four items, however, we can start to sense a
model that is a bit more like what we encounter in reality.
Has
some of
the relevant information available
Is able to perceive the information—albeit with
some inaccuracies
Is able to perform the calculations necessary
within a margin of error
Makes decisions that involve factors
other than
perfect rationality
In each of the italicized areas above, we have washed out some of the perfect
computational ability that computers just happen to be good at. In its place, we
now have some more fuzzy ideas and nebulous concepts. How do we know
how
much
information to make available? How do we construct
some inaccuracies
in the
perception? How do we insert a
degree of error
into calculations? How much error?
And what sorts of factors do we include other than perfect rationality? For that
matter, what other factors
are there?
This is where we can put
positive
or
descriptive
decision theory to work. By
starting with the raw logic of what they
should
do, applying an analysis of what
people, animals, or even orcs
tend to
do, we can build models that guide what they
will
do in our games (Figure 4.3).
FIGURE 4.3
Combining normative and positive decision theory takes the
limited world model perceived by an agent, adds the potential for errors, and
creates a belief about the world. When combined in a moderately rational fashion
with a behavior model constructed from observations of what people tend to do,
it yields varied, yet believable realistic behaviors.
