Rational vs. Irrational Behavior - Behavioral Mathematics for Game AI - page 109

If we are then to assume that the enemy forces are a cornucopia of varying unit
types and strengths, how do we go about building our own forces to adequately
match theirs? For example, if we were to discover that the enemy is building 60%
melee units, is our decision going to be different from if they were building
70% melee units? Or 75%? What if they were building 60% melee units but also
30% mounted units? Is that different than 60% melee and 25% mounted? How
much different? If there were five or six different unit types, the complexity mounts
considerably. What if our archers are 80% effective against melee units but only
60% effective against mounted units? Is that much different from if the efficiencies
were 70 and 50%, respectively? How much different?
Of course, this all comes with the caveat that we are able to perceive the num-
bers exactly. While this may be a simple task for a computer to do, it is somewhat
more difficult for our senses to process. Humans can only handle counting so many
objects at one time before we start to get a little bleary eyed. Unless the enemy is
polite enough to line his army up in neat rows of 10 units each, simply getting an
accurate count of the various units by type is a significant challenge. Oh… don't
forget that this number is also likely to be continually changing as well.
The best we can hope to come up with is that the enemy is creating “more of
A than of B and maybe twice as many A as C.� Calculating with any greater accuracy
than that may not only difficult, but, given the transient nature of any instanta-
neous count, it could also be a degree of specificity that is irrelevant. So what do we
do? While we are still interested in the general makeup of the enemy army, we
choose to be rationally ignorant of the exact order of battle.
C OMBINING I T A LL
So what is the logical, rational conclusion of all of this? At the beginning of the
chapter, we suggested that the human behavior that is presented by descriptive
decision theory (i.e., what people tend to do) lies somewhere between the rigid per-
fection of normative decision theory (i.e., what people should do) and complete
randomness. Let's review what we have explored.
Perfect Rationality Is Flawed: While we can admit that acting under perfect
rationality leads to decisions that are “right� in a theoretical sense, they often
fail to yield realistic-looking decisions.
Not Everyone Is Rational: The reason acting perfectly rational often does not
yield good results is that we cannot expect superrationality—the perfect ratio-
nality of all other players.