Changing a Decision - Behavioral Mathematics for Game AI - page 406

I NTHE G AME Flotilla of Futility

Consider the following example (Figure 15.1):

In a tile-based strategy game, we have assembled our naval fleet for a
bombardment and invasion. We analyze the coastal cities of our enemy to
determine which one is the most vulnerable. After using our complicated
utility algorithm, we decide to attack Dudetown (I just had to use the
Dudes again!), an ill-defended city on the western side of the continent.
Our fleet sets out in the direction of the soon-to-be-ravaged city (1).
When the fleet has traveled a significant portion of the distance (2), we
reevaluate the situation by once again analyzing the coastal cities. Much to
our surprise, we find that another city, Suckerville, is completely unpro-
tected! Our trusty attack algorithm determines that you can't get much
better than an undefended city. We issue new orders to our fleet to head in
the direction of Suckerville. Unfortunately, our new target happens to be
on the eastern side of the continent. Oh well, for the easy spoils, it is worth
the trip.
All the way around the continent (3), we continue to check our status.
Every time, the still-undefended Suckerville comes up on top of our prior-
ity list. Upon closing to attack range of Suckerville (4), we are astounded to
discover that, at the last moment, our enemy has moved large numbers of
defensive units into the city. Strangely (and almost simultaneously, it
seems), the heretofore well-fortified Dudetown is now devoid of military
occupation!
Well aware that an attack on Suckerville would be met with fierce resistance,
we see Dudetown once again as a far more attractive target. Our ships turn
around (5) and begin the long trip back around the continent toward what
must be inevitable victory on the western rim of this vast land (6).
As we once again spot the doomed city of Dudetown on the horizon (7), we
check our intelligence reports one last time and find… wait a minute… are
you kidding me?!?
Anyone who has played strategy games probably uttered a knowing chuckle
about two paragraphs in. It turns out that Suckerville was appropriately named.
Our decision model, and by association, our fleet, was exploited by the enemy.
This is a common technique to confuse the AI. It is most effective in tile-based
games where there is a distinct grid space that defines “in the city.� Moving all the
military units one space outside of a city causes the AI that is analyzing the poten-
tial targets to see the city as being completely undefended.