Game Theory

1 Introduction
Game theory is the branch of mathematics that tries to model conflict or competition between two
(or more) people. With the exception of Statistics, this is the branch of mathematics that is most
used in the social sciences like Sociology, Psychology, Political Science, and also Economics. Game
Theory had its roots in 1917, but began to pick up steam with John von Neumann in 1928 and
John Nash in the 1950s.
The basic setup is that there are some number of players, each player has some number of strategies
to choose from, and associated payoffs for each player for every combination of strategies. In general,
this can get very complicated, but we will be focusing on the simple case where there are two players,
and each player has two strategies. This would mean that there are four possible outcomes, with
payoffs associated to each player. This can be shown in a matrix like the following:

Player B
B1 B2
Player A A1 2, 2 2, 0
A2 3, 0 0, 9

This means that if Player A chooses strategy A1 and Player B chooses strategy B2 (denoted (A1 , B2 )
for short), then Player A would win 2 and Player B would win 0. The goal of each player is to win
the most (or lose the least). The only control each player has is in which strategy they themselves
choose.

One basic situation and example matrix would be a situation many of us have been in, catching
a bus. More specifically, suppose you want to cross a street to get to a bus stop, but the light is
red. You see the bus coming down the road, and you know if you do not cross at this moment,
you will miss the bus. Your choices (strategies): Cross the street or Don’t cross the street. What
can happen is in the next few seconds, either a car goes down the road, or there is nor car. This
situation can be viewed by the following matrix:

Nature
Car No Car
You Cross −1000 10
Don’t Cross −5 −5

The numbers are somewhat arbitrary, but the idea is that if you don’t cross, you will be slightly
annoyed that you missed the bus and have to wait for the next one; if you cross and there is no car,
you would be quite happy you made it onto the bus and don’t have to wait; if you cross and there
is a car, you get hit, which is obviously very bad. This sounds all rather silly, but many of us have
played this very game. We look at how busy the road is to tell how likely is it a car will be driving
by. That is, we try to determine which strategy Nature is going to play, and we react accordingly.
This highlights one of the main issues with game theory: If we knew what the opponent would do,
we know what we would then do. This can quickly degenerate to the classic ”If I know that she
knows that I know that she knows...” kind of thinking.

2 Zero Sum Games

A zero sum game is one where any winnings for one player counts as a loss for the other player.
We will generally simplify the diagram so it will be easier to read, where each number represents
winnings for Player A and losses for Player B.

B1 B2 B1 B2
A1 5, −5 −8, 8 A1 5 −8
A2 3, −3 1, −1 A2 3 1
Figure 1: A zero-sum game matrix and its simplification.

Suppose you are Player A is the above situation. How do you choose which strategy to use? We
want to have the best thing to happen to us, that is, win the most money. So, A1 seems tempting.
However, we would also have the potential to lose 8. In game theory, we generally take the pes-
simistic view of things. That is, for each of the strategies, what is the worst thing that can happen?
Out of those, which is the least bad? This is the same as finding the minimum of each strategy,
and then the maximum of the minimums. This is called the maximin.

Let us look at a specific example:

B1 B2
A1 6 5
A2 5 4

Notice that the minimums are 5 for A1 and 4 for A2 . So, to maximize these minimums, Player A
would choose A1 with worst case scenario of 5 if Player B chooses B2 . Now, these values represent
losses for Player B, and so, would big numbers are bad. That is, Player B would mind the maximums
of each strategy, and the minimums of those maximums (called the minimax). The maximums are
6 for B1 and 5 for B2 . So, Player B would choose B2 with worst case scenario of 5 if Player A
chooses A1 . Notice how both players ended up at the outcome of (A1 , B2 ). This is called the saddle
point for this game.
We could have come at this another way, somewhat through trial and error. Again, consider yourself
as Player A. Ask yourself, ”If Player B chose B1 , what would I do?” ”What if B2 was chosen?” No
matter what Player B does, you would want to choose A1 . Similarly, Player B would always want
to choose B2 . If Player B were to deviate from this, it could only help you, and vice versa. So,
(A1 , B2 ) is the only ”logical” choice.

Now, a more interesting example:

B1 B2
A1 3 6
A2 5 4

Let’s follow the same reasoning we just used in the last example. If you are Player A, what would
you do if Player B chose B1 ? Since (A1 , B1 ) = 3 < 5(A2 , B1 ), you would want to choose A2 . However,
if you were to choose A2 , Player B would want to choose B2 . But then, you would rather choose
A1 . This would, in turn, make Player B want to choose B1 . So you would like to.... As you can see,
this degenerates quickly to a case of ”If she knows that I know that she knows that...”.
We can also try to find the minimax and maximin. Notice that if Player A were to play conser-
vatively, strategy A2 would be chosen, with the worst case scenario of 4. Similarly, Player B’s
conservative move would be B1 with worst case scenario of 5. Since these two values do not agree,
there should be some sort of strategy that would balance things out.
This problem is not so cut-and-dry as the last one. The main difference is issue of prior knowledge.
If you could spy on the other person, or if they gave some cue as to which strategy they will choose,
it would have made no difference in the prior example since you would choose the same strategy
regardless. Here, you would tailor your strategy based on what your opponent is planning. The
only solution to this is to effectively get rid of the planning, in some sense. That is, we want to
institute some chaos by randomly picking between the strategies. The question is now, how often
should you pick A1 versus A2 ? To figure this out, we will follow a strange method, which hides a
lot of math in the background. First, subtract one column from another (say, the second minus the
first). We will also do similarly for Player B by subtracting the rows.

B1 B2
A1 3
2 −2
A2 −1

To average each of these to be zero, we would want to choose A2 3 times as often as A1 . That is,
75% of the time A2 should be chosen, and 25% of the time, A1 should be chosen. If we played with
this mixed strategy, what do we expect would happen? First, suppose Player B chooses B1 . Then,
75% of the time we would win 5, and 25% of the time we would win 3. That is, on the average,
we would win .75(5) + .25(3) = 4.5. Similarly, If Player B chooses B2 , then we would expect to
win .75(4) + .25(6) = 4.5. This shows that regardless how Player B plays, we will win the same
amount on the average. So, Player B cannot outsmart us or trick us into getting a smaller amount.
However, this also says if we play this way, we can’t do any better than 4.5 no matter how ”bad”
Player B plays. Also, you can check that the mixed strategy for Player B should be to randomly
choose between strategies, but pick each one equally often. Also, this would cause the expected
value of the game to be, again, 4.5.

3 Non Zero Sum Games

Now, we shall (quickly) look at some games that are not zero sum. For example, consider the
following situation:

Stag Hare
Stag 10, 10 0, 4
Hare 4, 0 4, 4
Figure 2: ”Stag Hunt”

In this situation, consider two hunters trying to decide if they are going after big game or small
game (”Stag” or ”Hare”). They can get the Hare by themselves, but would need to work together
in order to get a Stag, but each would see a large benefit. The minimax/maximin strategy would
say they should go after Hares, but it would be in both their best interests to work together to get
a Stag.
Another example (Warning: Gender stereotypes ahead!):

Woman
Ballet Football
Man Ballet 6, 10 0, 0
Football 0, 0 10, 6

Suppose a man and a woman decided to go out, the man wanting to go to a football game, and
the women to the ballet. Further, suppose they know when they are going to meet, but forgot
what they decided to do. If on the day of the event they couldn’t contact each other, how should
they decide where to go? If they end up in different places, no one would be happy. If they end up
in the same place, they both would be happy, but whoever is doing the activity they prefer is happier.
 Silent Talk
Silent −1, −1 −10, 0
Talk 0, −10 −7, −7

Finally, let’s finish with the Prisoner’s Dilemma:

The situation here is that two people are arrested by the police for a crime, and each are offered a
deal to rat out the other person. If they both stay silent, they are both convicted for a minor crime.
If they both rat each other out, the both go to jail for a significant amount of time. However, if
one talk and the other does not, the silent one goes to jail for a long time, while the other walks
free. What is interesting here is that if both players follow the minimzx principle, they both get -7,
when they could have agreed to both get -1. However, there is always the temptation to go for the
0 and the fear to stay away from the -10. This gets even more interesting when playing the game
multiple times, as players can react to prior betrayals. A good resource to look at simulations of
this is http://www.iterated-prisoners-dilemma.net/.