CIRJE-F-34. Matsushima, Hitoshi, "Learning as the Dynamics of Emotions and Strength of Confidence", December 1998.

This paper investigate a Bandit problem in which a decision maker chooses between the safe action and the uncertain action infinitely many times. She does not know the expected payoff for the uncertain action. Her decision on whether to experiment is influenced by the atmosphere in the society. An adaptive learning rule is introduced which regulates her unconscious dynamics of emotions. We argue that realistic decision makers mostly fail to choose the efficient action in the long run, even though they experiment infinitely often. We present a necessary and sufficient condition for efficient learning which requires a very restrictive psychological nature that the confidence which a decision maker has in the uncertain action being better is the strongest.