Game Development Reference
In-Depth Information
we would have to model the game world in such excruciatingly fine detail that we may
as well be creating the
holodeck
from
Star Trek
. To sum up,
it ain't gonna happen
.
Of course, constructing one algorithm to tell us what we “would do� is a lot like
the normative decision theory approach of one algorithm to tell us what we “should
do.� That's what got us into this problem in the first place. To return to our bank
example, we would have to construct algorithms to replicate the life experiences of
20 typical bank customers and 5 typical bank employees. And when the cops show
up at the bank to arrest us, we would have to create three or four individual “life
experience models� of typical police officers. If constructing and using
one
such
model was prohibitive, we probably don't have a word for what the prospect of
creating
dozens
of them would be like. (I nominate “insane.�)
Never Mind All That…
Remember, however, that we really don't care about how a bank customer got to the
point in his life that would cause him to react in a particular way; we care about
the
result
of all of that information and experience-building. This is the same ratio-
nale that we used with the dentists dispensing their helpful advice. We don't know
or care
why
four of the five dentists recommended sugarless gum or
why
the fifth
did not. We just know that they did. If we were going to create our Dental Advice
Simulator, we would not have to worry about the fact that one of the dentists'
spouses recently died of complications due to diabetes and another is taking kick-
backs from the Sugar Growers of the World Association. To
simulate
the dentists
we would simply codify that 80% of dentists we encounter recommend sugarless
gum and the remaining 20% do not. We can leave it up to the players to overlay
such creative interpretations if they
really
feel the driving need to do so.
As we dealt with in Chapters 11 and 12, probability distribution and random
selection is something we can model rather well. How hard is it, after all, to recom-
mend sugarless gum 80% of the time? Using that simple procedure, our Dental
Advice Simulator could simulate a whole
convention hall
full of dentists! We have
substituted simple probability distributions for the complex (and largely unneces-
sary) process of generating the minutia of what goes into individualized human
decisions. The results are strikingly compelling, however. When we query our faux
dentists about their thoughts on sugarless gum, rather than acting in rigid unison,
80% of them (give or take a few, I'm sure) would raise their hands. A player of our
Dental Advice Simulator would see that, nod to himself, and say, “Wow… looks like
about four out of five to me. That's a reasonably accurate simulation of dentists!�
A Framework for Randomness
It's important to note that we did not make the decision completely random. We
did not flip a coin to decide. Doing so would have generated a 50/50 split on the
