Abstract
Scientific Realism stands or falls with Inference to the Best Explanation. Realism cannot be accepted if one has reason to think that Inference to the Best Explanation cannot lead to the truth, or is unlikely to. Peter Lipton raises three important problems for his model of Inference to the Best Explanation: Voltaire’s objection, Hungerford’s objection, and the problem of Underconsideration. In this paper I show that Lipton’s own solutions do not fully answer those problems. I argue that what is required to solve these problems is for our conception of explanatory goodness to be truth-conducive because it is sensitive to the way the world actually is. I suggest that the cognitive psychology of exemplars, as described by Kuhn, may provide an answer.