Robotics Reference
In-Depth Information
ers still in the pot (the more opponents there are, the less likely it is that
the program holds the strongest hand), and what betting has taken place
thus far (if the betting has suggested that one or more of the opponents
have strong hole cards then the program downgrades its assessment of its
own hand strength).
In other versions of the program, betting decisions are computed in
a different way, using Monte Carlo simulations. At a given point in the
hand, a simulation is conducted to the end of the hand in much the same
way as Ginsberg did in Bridge. One thousand simulations are run, both
for the scenario where the program's initial betting decision is a call, and
for when it is a raise. (A fold decision is easy to compute—it has an
expected value of zero.) The program then chooses the fold, call or raise
decision
52
that has the highest expected value.
In order to increase the accuracy of evaluation of its own hand, the
program assessed the likelihood of each of an opponent's possible hands
being played to the current point in the pot. In effect, the program
asked the question: “Given the way that the betting has gone thus far,
what is the probability, for each of the various possible holdings, that my
opponent does hold that hand?” This idea had been suggested by Findler
as a way of placing a program in its opponents' seats at the table and
reversing its reasoning processes by inferring from the opponents' actions
their status in the game. The stronger the opponents' hands appear to
be, the less strong Poki believed its own hand to be. Of course, the
opponents can bluff, but when assessing the strength of an opponent's
hand on the basis of his bets, Poki takes into account whatever is known
about the frequency with which that particular opponent bluffs.
53
Another important feature in Poker is the potential of a hand, which
is not the same thing as its strength because the arrival of new commu-
nal cards can change everything. Hand potential is another important
evaluation feature—one that assesses the probability of a hand improv-
ing or being overtaken when additional communal cards arrive on the
table. For example, if a player holds the following two-card hand
♣
♦
Q, then at the moment his hand is low in strength (not even a pair),
but it offers excellent potential for improvement to a flush (five cards of
♣
♣
♣
2,
4, and if the communal cards showing thus far are
7,
9,
52
If the program can check, it is treated as equivalent to being able to call. If it can bet, it is
treated as equivalent to being able to raise.
53
It is of course only possible to discover that an opponent has bluffed when he stays in the pot
until the showdown, after all the betting is finished, and is then forced to show his cards.
