Game Development Reference
In-Depth Information
First, any changes Laurie and I make to our preferences for food are going to
change our personal desire utilities. (I actually do really like Chinese food most of
the time!) We can change how much we desire various foods. Additionally, we
could change the weights of who gets to decide. There may be times when we would
weight her preference equal to—or even
greater than
—mine. However, that is only
the first step in the process.
Any changes we make to process that determines the utilities of time or price
for a specific meal are going to propagate from the first layer to the second just as
the changes to our personal desire weights do. However, the weight of the factor
where we made the change throttles the magnitude of the changes somewhat. It is
entirely possible that a major change in one factor will have very little effect in the
overall decision because that factor's weight is minimal. On the other hand, if a fac-
tor weighs heavily in the decision, even a minor change can cause a significant shift
in the final number.
The last layer is the weights themselves. We must remember to think of the
weights as their own utility function. Naturally, changes we make to those weights
have a direct effect on the outcome.
After passing through all layers of the filter, we arrive at a single value that is a
composite of all the processes and weights that are in place above it. We can then
compare these final utility values to determine which selection has passed through
“best.�
C
OMPARTMENTALIZED
C
ONFIDENCE
There are no limits to the depth or breadth of the multi-layered weighting process.
We can have as many levels as we need. Each layer can have many different compo-
nents as well. Our only limit is the information that we have available in our design.
(From a design standpoint, an alternate mentality is to say, “This is the decision I
want my characters to make… what information should my game track to facilitate
it?�) Of course, the complexity increases as we add more factors and layers.
There are methods for managing this complexity, however. Most of the meth-
ods involve making sure
we
don't confuse
ourselves
through our own process. One
such method is a topic we already covered in this chapter. By ensuring the homo-
geneity of our data by establishing limits and practicing normalization, we keep the
relationships between factors relatively simple.
As a quick example, if we were to have scored Laurie's dinner preferences on
one scale and mine on another, the process of weighting them appropriately would
have been more complicated. By scoring all preferences on a normalized scale of 0
to 1, we ensured that the 2:1 weight ratio between my desires and hers was, indeed,
