Game Development Reference
In-Depth Information
Amazed Dad: “So how do we know what price is right?�
Aaron: “Well… we should raise it a little at a time and keep watching to see how many
people are coming in. If less people come in, then we know that it was too much.�
Proud Dad: “Son, you just expressed a concept that many 40-year-olds don't
understand.�
Aaron had made the observation that he wasn't operating in a static environ-
ment. His prospective customers may react to his decisions. Therefore, to make the
correct decision, he needed to be able to somewhat predict what their behavior
would be. What Aaron needed to determine was what the interface was between
his
decision and the reaction of the guests. There must be a relationship that can be fig-
ured out that would predict, within reason, their behavior. Of course, we all know
that what he was expressing was the connection between price and demand. If
something costs little, it is more attractive; as the price increases, it is less attractive.
If that is put into a consumer model (and additionally considering that differ-
ence in consumers' ideas as to what the “right price� is), you will find that a rising
price will satisfy fewer and fewer people. Once the number of willing buyers falls to
a certain point, it doesn't matter what we charge them; we won't be making as
much money as before. Taken to the extreme, you could charge one million dollars
for a zoo pass, but you probably won't get many takers. Alternatively, if we try to get
more customers by cutting the price, we could reach a point where we are getting
lots of business but not making any money. Again, the extreme end would be mas-
sive numbers of guests for free. There is a businessman's joke that says, “We'll make
it up in volume.� Uh… no you won't.
Turning Human Decisions about AI into AI Decisions
Interestingly, as this was a computer game, and the park guests in
Zoo Tycoon
did
react appropriately to changes in price, we know that some clever designer had
actually created a formula for that behavior. As humans playing the game, our job
was to determine what this formula was. We know that attitudes would move in a
certain direction, but we didn't know how much. As I mentioned, my son's approach
was to move the price a little at a time and observe the changes in behavior.
Eventually, we would be able to determine the right price for our zoo admission.
(Of course, since our zoo was constantly changing in what it offered, the right price
was a moving target—a sign of a good algorithm design. More on that later.)
However, in game AI, the roles are slightly different. The decision that we hu-
mans were making is one that could be applied to an AI agent such as a shopkeeper
in a role-playing game (RPG). What is the right price to charge for a particular
item? If all else was equal, the role of zookeeper would have to do just as we did—
take into account the unknown formula that went into crafting the decision model
