A key thesis of this book is that experimental inferences can be classified in two broad categories: (1) inductions within the experiment and (2) inductions from the experiment. In Chapter 3, I introduced a further distinction borrowed from Bogen and Woodward (1988). Inductions within the experiment typically proceed in two phases: in the first phase, scientists use empirical data to identify underlying patterns or phenomena; in the second, they try to explain these phenomena by means of “deeper” theories or causal hypotheses. Both steps are nontrivial, in the sense that the data do not univocally indicate the underlying phenomena and the latter do not univocally indicate their causes. As a consequence, scientific inferences require a certain amount of ingenuity. Following an established terminology, we can say that experimental claims or hypotheses are typically underdetermined by the evidence. This chapter is devoted to discussing the attempt to solve this problem by means of the criterion of predictive success. I argue that such an attempt and other related projects fail because they ignore a key aspect of scientific inference: the background factors that determine whether the evidence confirms, refutes, or neither confirms nor refutes a hypothesis.

Like many other philosophers, I believe that the solution to underdetermination problems must lie in the area of inductive logic. Theories of inductive inference are numerous and at times highly sophisticated. Presently there is no agreement on a general theory, a fact that prima facie may seem to play in favor of the skeptics.