Evidence and Inference
Dissertation, University of California, Los Angeles (
1980)
Copy
BIBTEX
Abstract
I have attempted to evaluate both Reichenbach's and Harman's views with respect to one such traditional problem of induction, namely the problem of general rules which has come to be known as the 'Grue Paradox' and I argue that that problem arises for inference to the best prediction but not for inference to the best explanation. ;If it can be shown that inductive inferences may in fact be correct for different kinds of reasons, then we will no longer have any reason to think that any one theory for the justification of principles of inductive correctness has to solve all of the problems which have been raised so far concerning inductive inference. Perhaps thinking that one theory must be comprehensive in this sense has kept inductive logic in its rather low state. However, once it has finally been realized that at least two distinct sets of principles are required to provide a criterion of inductive correctness, we may then find that certain problems encountered by one set of principles are avoided altogether by the other. Perhaps only when we have realized that we don't have to solve certain problems with respect to particular principles of inductive correctness, will we then be able to complete the task of constructing a full-fledged inductive logic. ;Specifically, the way in which I have shown that inductive inferences may be correct for different kinds of reasons is by showing that the rule of inference to the best prediction proposed by Reichenbach and the rule of inference to the best explanation proposed by Harman yield concepts of inductive evidence which are incompatible. Reichenbach's rule of induction yields the confidence concept of statistical evidence and Harman's rule of induction yields the sufficiency concept of statistical evidence. The confidence concept and the sufficiency concept have been shown by both Birnbaum and Hacking to yield incompatible results, hence if those concepts of evidence are incompatible so are the rules proposed by Reichenbach and Harman. It would then appear, and indeed I want to claim that it is the case, that inference to the best explanation and inference to the best prediction should be treated as distinct goals of inductive inference, which call for separate and distinct justifications. ;Essentially I have tried to show that there is a deep divergence between two of our most important bases of reasoning--explanation and prediction--which call for separate and distinct justifications. In particular, I have tried to show that Reichenbach has so far the best method for justifying inferences to the best prediction and that Harman has so far the best method for justifying inferences which are inferences to the best explanation. For too long philosophers have thought that all correct inductive inferences are correct for the same reason either because all correct inductive inferences are ultimately reducible to inference to the best prediction or because all correct inductive inferences are ultimately reducible to inference to the best explanation. Hempel in fact has argued that inference to the best prediction and inference to the best explanation have the same logical structure and are hence functionally equivalent. But to think that explanation and prediction are symmetric, or to think that there is but one criterion of correctness which every correct inductive inference must satisfy is, I contend, a mistake. Philosophers have long since abandoned the misconception that inductive inferences are inferences which procede from the specific to the general and it is now time to abandon another misconception, namely, the misconception of thinking that correct inductive inferences are all correct for the same reason and that there is ultimately only one kind of correct non-demonstrative inference