Abstract

We extend a result by Gallow concerning the impossibility of following two epistemic masters, so that it covers a larger class of pooling methods. We also investigate a few ways of avoiding the issue, such as using non-convex pooling methods, employing the notion of imperfect trust or moving to higher-order probability spaces. Along the way we suggest a conceptual issue with the conditions used by Gallow: whenever two experts are considered, whether we can trust one of them is decided by the features of the other!