Talk by Joseph Berkovitz: On evidence, induction, and probability
On evidence, induction, and probability
Institute for the History and Philosophy of Science and Technology
University of Toronto
This is based on joint work with Aaron Kenna. The various conceptions of probability can be divided into two broad groups: objective and ‘epistemic’ probabilities. Intuitively, objective probabilities are supposed to reflect objective features of the world, such as frequencies, symmetries, and laws, whereas epistemic probabilities are supposed to reflect states of knowledge/ignorance that agents or groups have, or should have, in conditions of uncertainty.
Epistemic probabilities are related to evidence, and their value might differ relative to different evidence. We argue that there are two main conceptions of the relation between evidence and probability: the ‘logical’ and the ‘psychological’. In the ‘logical’ conception, which is the dominant view, probabilities should reflect the evidence on which they are based. Ideally, the evidence, jointly with rational principles and premises, should uniquely determine epistemic probabilities. In the ‘psychological’ conception, probabilities neither reflect the evidence, nor are determined by it. The relation between evidence and probabilities is open-ended and its exact nature depends on the context and the reasoning agent. The reasoning from evidence to probability is governed by instincts, intuitions, and know-how, which are the outcome of experience and training. Accordingly, different agents could rationally infer different probabilities from the same evidence.
We argue that the logical conception is inadequate. In particular, this conception fails to reflect the important role that instincts, intuitions, and know-how play in scientific reasoning in probabilistic reasoning. We also suggest that the psychological conception is on the right track.