Degrees of confidence are not subjective probabilities

Abstract

We assume that 'confident' reports have their standard degree-based truth conditions: to be 'confident' that 'p' is to have a degree of confidence that 'p' which is at least as high as some contextually determined threshold; to be 'more confident' that 'p' than that 'q' is to have a degree of confidence that 'p' that is higher than one’s degree of confidence that 'q'; and so on. But what are degrees of confidence? The standard answer is that agents’ degrees of confidence are simply their Bayesian subjective probabilities: for example, how confident one is that it’s raining = how likely one thinks it is that it’s raining. We raise a number of challenges for this Bayesian account, and propose an alternative. This new account supports a pragmatic explanation of the apparent equivalence of degrees of confidence and subjective probabilities, and offers a more integrated picture of how different doxastic attitudes fit together.