Game Development Reference
In-Depth Information
If you put that mentality on the other side of the table from you in the form of
a human player (assume for the sake of example continuity that it is the same guy
with the dark glasses and stupid hat), it fails to provide any sort of challenge over
and above playing against the roll of an Official Rock-Paper-Scissors Decision Die
(ORPSDD). You are playing against a random selection machine. The only differ-
ence is that the human player may be more likely to react to your victory dance than
would the ORPSDD. In fact, as long as the taunting was not included in the test, I
would suggest that an ORPSDD would admirably pass the
Turing Test
(see sidebar).
TURING TEST
The oft-cited Turing Test was conceived by famed AI researcher Alan Turing. He sug-
gested that a machine entity could not be deemed “intelligent� unless it performed
the action being tested in a manner indistinguishable from a human.
Currently, the test is used in a far more loose sense than what Alan Turing originally
proposed. That leads to its being used in situations for which it was not intended—
much as the term “litmus test� is used in far more situations than the chemical test
that it truly is.
In a broad sense, the term “Turing Test� has become the equivalent to saying
“believable.� If one says that an AI entity “passes the Turing Test,� it generally means
that it looks as good as a human player. This believability is always going to be limited
in scope, however. Usually, only a specific behavior or set of behaviors is being judged.
This scope is, by necessity, also limited to game-relevant situations. For example, an
agent may run, hide, and shoot in a manner that looks like a real player. However, if
you try to engage this same agent in a conversation about the weather, the agent is
going to be exposed for the narrowly designed entity that it is.
That being said, the phrase “it passes Turing� is a colloquially understood badge
of honor that means “it looks and acts pretty darn well.� As such, it is a goal to strive
for as an AI designer and programmer.
Tic-Tac-Toe has a few more options, as I mentioned earlier. However, the state
space of the game is so small that at any given moment the choices we have before
us can be reduced to “correct,� “incorrect,� or “delaying the inevitable.� If the point
is to win the game, there really is no choice involved. At least not much more than
answering the question, “Do you want to win the game?� So, while there are more
choices in play for both you and your opponent, victory comes down to a failure on
either your part or that of Mr. Glasses & Hat to correctly answer the question, “Do
you want to win the game?� (Figure 1.3). Assuming a static level of continual lucid-
ity for both of you, the only way someone is going to lose is if he
allows
the other
