Abstract

Peter Lipton famously argued that we want our scientific theories to be lovely and likely, that is, we want them to provide good explanations and to be very probable (if not true). Unfortunately, there is a tension between these two epistemic virtues, and it is not clear how they are related. Thus, the question arises whether the Bayesian (who prefers likely theories) and the explanationist (who prefers lovely theories) can be friends, as Lipton claims. Although much ink has been spilled over this question, in this talk I want to take a fresh look at it and make two points: First, I argue that successfully providing an explanation is an example of non-empirical evidence in favour of the theory in question. This point can be made more precise by a simple Bayesian model, which also provides (as a bonus point, so to speak) a justification for the bonus point approach to explanationism inspired by van Fraassen and championed by Douven – at least if certain conditions are met. Second, I investigate how the strength of an explanation – its explanatory power – can be measured in Bayesian terms, and show how this all fits nicely into a coherentist epistemology of science.