Between Sophistication and Resolution: Wise Choice

In this chapter, which draws on my earlier work on the subject, I focus on agents whose preferences violate axioms of expected utility, with violations of the Independence Axiom being my prime example. This exposes them to the threat of dynamic inconsistency in sequential choice. I consider three different policies such agents could use to secure consistency. After presenting two standard policies, sophisticated choice and resolute choice, I move on to wise choice - the policy I favor. Unlike sophisticated choice, wise choice does not assume diachronic Separability: it allows that what would be rational to do at the later stages of a sequential decision problem might depend on what the agent has previously done or might have done or on the plan she has previously adopted. On the other hand, unlike resolute choice, which also rejects Separability, wise choice does not assume Reduction to Normal Form, according to which the sequential character of a decision problem is irrelevant to its solution. Like sophisticated choice, wise choice makes use of backward induction but applies it differently: in predicting future choices, a wise chooser does not simply rely on her current unconditional preferences. Instead, she goes by her current preferences conditioned on the hypotheses regarding her future choice situations. In the final section, however, I consider whether wise choice can after all be reduced to sophisticated choice if the agent’s decision problem is appropriately re-described. While such reduction is possible, it has its costs and limits.