Variation in Choice - Behavioral Mathematics for Game AI - page 425

In the above criteria for what we do and don't want, we were coming at the
problem from both ends. We want more than just the best choice. We acknowl-
edged that there are other choices that might be acceptable. On the other hand, we
certainly don't want the worst choice, and there are likely a lot of choices that aren't
the worst but are pretty darn bad as well. We want to avoid those. But how do we
separate the good from the bad?

R ANDOM FROM T OP n C HOICES

If we list our choices ranked from optimum down to downright silly, someplace in
the middle is a meeting point between the good ones and the bad ones. The prob-
lem is that we don't know where that point happens to be. This causes us to get a
little arbitrary at first.
In our Dudes example, we scored each combination of Dude and weapon. By
selecting the best score (in this case, the lowest number), we chose which weapon
to use and which Dude to target. However, that means that every time we encounter
that exact arrangement of Dudes, we will do exactly the same thing. While this may
not be startling in and of itself, what if there are two of us… or five… or a whole
bank full of us all running the same algorithm. We would all respond exactly the
same way. We would all pull out the same weapon and fire at the same Dude.
Never mind that this collective response is a tactical nightmare; it just looks silly.
It looks even more ludicrous when we translate it back out of our Dude scenario
and into something like our bank example. Even if everyone was processing the
same inputs, would everyone react exactly the same way? Not likely. Therefore,
despite the fact that there may be a single best Dude to attack with a single best
weapon, it probably does not behoove us to insist that we follow that model rigidly.

Opening the Playbook
Thankfully, one of the advantages of ranking the scores was that we learned more
than simply which one was best . We also are aware that there were others that were
decent choices as well… just not the best. To confirm this, let's examine the results
of our final Dude example from Chapter 14. (Note that in Chapter 14, we had lim-
ited our list to only the best weapon for each Dude. Here, we include them all. We'll
see why in a moment.)

Name Weapon Score

Evil Genius R/L 7.7

Boss Man R/L 8.5

Baddie 3 R/L 9.7

Evilmeister R/L 9.8