Some Incoherencies Resulting from Minimal Information Methods

The technique of minimizing information (henceforth, infomin) has been commonly employed as a general method for updating one’s prior probabilities. In the past, infomin updating has been criticized on the grounds that the updating behavior it recommends cannot be represented within a higher-order Bayesian framework. We argue that, in a wide class of cases, infomin can be faulted on more fundamental grounds without presupposing that an agent is always committed to a single, determinate higher-order probability distribution.