Factor Weighting - Behavioral Mathematics for Game AI - page 348

If we do need to make a change, we must be careful to only change components
where there are problems. Of course, that necessitates finding where the problem
begins in the first place. We must resist the temptation to begin tweaking one com-
ponent to solve problems in another. We must locate the problem first and then
change only that component until it generates the results we want or expect (which
are not always the same thing). The process can be involved, but the end result is
that our decision models can handle many widely disparate pieces of information
in a manageable fashion.
E VERYTHING I S R ELATIVE
If the prior chapters have given us the tools with which we can build our AI algorithms,
this chapter gave us the rules for drawing maps and blueprints and measuring our
construction materials. As with maps and blueprints, through the right application
of the techniques above we can put the entire world into perspective.
I once had a social studies teacher who repeatedly proclaimed, “Everything is
relative.� During class discussions, he would often respond to a student's unqualified
claim with “Compared to what?� For example, the word hot is only meaningful in
context. A hot day may be 90 degrees (F). “Billy's head feels hot� may mean a 100-
degree fever. On the other hand, both 90 and 100 degrees make for comfortable , not
hot , bath water. I'm sure that cooking your steak over “hot coals� would suggest
something hotter than what you would be willing to bathe in.
His point was that information needs to be processed together with other related
information. Additionally, we can combine information with other less strongly
related information if we can find some sort of common ground as an intermediary.
While this was neither his forte nor his intent, his assertion is prophetic in the
realm of behavioral AI. Because we are often trying to find that one decision factor,
we must combine many different pieces and parts. Often the information that has
common ground is only meaningful when we put it in context. And for those con-
cepts that aren't so obviously related, we define that common ground. The tools
above allow us to do this. Through the iterative and layered process, we can take
innumerable bits of information and construct one final expression that, in the
proper context, represents to us what should be done.
Of course, as we shall soon revisit… just because it should be done doesn't
mean that we will do it! After all, who among us is perfectly rational?