Rationalizations and Promises in Games
Johan van Benthem

Abstract:
Understanding human behaviour involves "why"'s as well as
"how"'s. Rational people have good reasons for acting, but it can be
hard to find out what these were and how they worked. In this Note, we
discuss a few ways in which our actions, preferences, and expectations
are intermingled. This mixture is especially clear with the well-known
solution procedure for extensive games called 'Backward Induction'. In
particular, we discuss three scenarios for analyzing behaviour in a
game. One can rationalize given moves as *revealing agents'
preferences*, one can also rationalize them as *revealing agents'
beliefs* about others, or one can change a predicted pattern of
behaviour by *making promises*. All three scenarios transform given
games to new ones, and we prove some results about their scope. A more
general view of relevant game transformations would involve dynamic
and epistemic game logics.