Game Development Reference
In-Depth Information
2 to 1. If, instead, we score my ratings between 0 and 3 and hers between 0 and 5,
we have to perform one extra step to say that, at the moment,
my
desires are worth
twice as much as hers.
The goal is to isolate each individual component as much as possible. If we can
ensure the integrity of a particular component, we do not need to know the process
that generates the inputs to it, or what is going to happen to the output it generates.
We have established
compartmentalized confidence
in that single portion of the
decision.
If we trust that the decision model for each of our components is perfectly
valid—
in its own scope
—then we can subsequently trust that they will all work
together well (Figure 13.8). That is, if we believe that each step is correct, the entire
decision model will be correct as well. Of course, if the outcome of the decision
model doesn't seem to make sense to us, we need to go back through each layer of
the model and question our premises.
FIGURE 13.8
If we ensure the integrity of each stage of a layered weighting process, we
can feel confident that the output is correct as well.
For example, if we are comfortable with the idea that time, price, and desire are
the three components to our dinner decision,
and
we believe that the weights of 5,
3, and 2 (respectively) were correct for a night when time is tight and money is
scarce, we should be happy with the results that these weights generate. We trust
that the processes that determine the utility scores for time, price, and desire are
doing their jobs. We also trust that whatever score we generate at this step will be
used correctly down the road. (Theoretically, we could use the final result of
the “dinner decision� example in a larger decision model—such as “plans for the
evening.�)
