Game Development Reference
In-Depth Information
On the other hand,
we
are a little disturbed by what we just witnessed. We
turn to the door, muttering under our breath, “Wow… it was like a bad
video game or something.�
V
ARIATION BETWEEN
A
CTORS
As ludicrous as the above scenario seems, many of us have probably experienced
something like this in games. The underlying technical problem with the above
scenario (if you didn't get it already) is that all the agents were using the same algo-
rithm to determine their reaction to a stimulus. Every time we changed the world
state through our actions, we fed the 25 people the same information with which to
calculate their decisions. As a result, they all arrived at the same conclusion about
what the “best� reaction should be.
The only difference in the actions of the various people was that the employees
and the customers had different reactions at one point. The reason for that is that
they had different programmed reactions to the stimulus of us waving our gun
around. The customers dropped to the floor and began screaming; the employees
tried to placate us. The only reason for this occurring is that, at some point, either
a designer or a programmer differentiated between the two: “When the player does
[insert action], I want the customers to [insert reaction] and the employees to
[insert reaction].�
While the separation of customer vs. employee is admirable and was certainly
a logical division, there is a subtle implication in that statement. Imagine the above
statement rephrased this way: “When the player does [insert action], I want
every
single
customer to [insert reaction] and
every single
employee to [insert reaction].�
By inserting the specifier “every single…� we draw attention to the inherent weakness
in this line of thinking.
The differentiation between customer and employee was based on the reason-
able premise that customers and employees would likely react to the situation in
different ways. This is a logical assumption given the fact that their roles in a bank
robbery would be different. They enter the scenario of a bank robbery with differ-
ent goals. The customers' primary goal is “don't get killed.� The employees have
two goals: “don't get killed� and “defuse the situation by giving the really scary man
what he wants.� The problem with the artificial intelligence (AI) as theoretically
designed in our example is that every person's solution for how to reach those goals
was exactly the same as every
other
person's.
If we were to imagine how a scenario such as this one would go in real life (or
even in a movie, for those of you who can't envision yourself in the position of the
bank robber), we likely would have a different picture than the one painted above.
Simply making a list of possible customer reactions is enlightening.
