Consider, for example, the game depicted in Figure 1, which shows the
possible strategy choices for each player beside the rows and above the columns,
and the payoff to each player inside the cells, with player 1's payoff coming
first. We assume that player 1 chooses a strategy that maximizes expected
payoff given some subjective assessment over player 2's choice. Does this
assumption allow us to reach any conclusions concerning player 1's choice? The
answer is evidently no: either U or D might be optimal for player 1, depending
on that player's precise assessment. What about player 2? Simply knowing that
player 2 maximizes, given some assessment over player 1's choice, is sufficient
to rule out player 2 choosing R. If player 2 assesses probability greater than
1/2 to player 1 choosing U, then player 2's best choice is L; if player 2 assesses
probability less than 1/2 to player 1 choosing U, then player 2's best choice is
C; finally, either L or C is best if player 2 assesses exactly probability 1/2 to
player 1 choosing U.
Now comes an important assumption: suppose that the conclusion just
reached concerning player 2 is available to player 1. What exactly is entailed by
this assumption will be discussed in a moment. First, however, let us see where
this line of reasoning leads. If player 1 can indeed eliminate the possibility of
player 2 choosing R then player 1's subjective assessment should assign probability
0 to this event, in which case U is a better choice for player 1 than is D. If
this latter conclusion is in turn available to player 2 then player 2 will choose L.
Our reasoning has led to the conclusion that player 1 will choose U and player
2 will choose L.
What informational assumptions are implicit in this line of reasoning? First,
player 1 is assumed to know that player 2 is rational, which means that the
player chooses a strategy that maximizes expected payoff, given some probabilistic
assessment over the other players'—in this case, player 1's—strategy
choices. Second, player 1 is also assumed to know player 2's payoffs. These two
assumptions together allow player 1 to conclude that player 2 will not choose
R. Similarly, player 2 is assumed to know that player 1 is rational and to know
player 1's payoffs, and to know that player 1 knows player 2 is rational and that