What makes a strong inference to the best explanation? Two major challenges arise when evaluating the strength of an inference to the best explanation. First, there is the difficulty of determining which candidate explanation is “best.” Second, we face the problem that the true explanation might not be among the candidates we are considering, simply because we haven’t thought of all possible explanations. Let’s consider each of these challenges in turn.
To decide which explanation is the “best” explanation of some phenomenon, we can assess the candidates in terms of explanatory virtues—characteristics that we believe make an explanation more likely to be true, such as simplicity, breadth, consistency with other things we believe, etc. Scientists frequently employ this approach when considering competing theories or hypotheses. Unfortunately, there is no definitive, universally accepted list of explanatory virtues, even in the sciences.Historian and philosopher of science Michael Newton Keas identifies twelve important virtues of good scientific theories, and he develops a helpful classification schema that illuminates the relations between different virtues. His list of virtues may be the most comprehensive in the literature, but he does not claim it is exhaustive. See Keas (2018), “Systematizing the theoretical virtues,” Synthese 195:2761–2793. To complicate the matter, explanatory virtues sometimes conflict with each other: a complex explanation might have greater breadth (i.e., it may explain a broader range of phenomena) than a simpler explanation, for example. In cases like that, we must make the difficult decision which virtue to prioritize. Moreover, there is no fully objective way of comparing degrees of simplicity, breadth, and so on. For these reasons, philosophers and scientists often must rely—at least to some extent—on their own intuitions and common sense when comparing the virtues of competing explanations.
The second problem seems inescapable: we can never be absolutely certain that the true explanation is among the candidates we are considering. The history of science is rife with examples of seemingly “best” explanations that were later rejected when a new, previously unconsidered, explanation was devised. Nonetheless, it is often reasonable to believe the conclusion of an inference to the best explanation with a high degree of confidence, especially when one candidate explanation exhibits many virtues and is clearly superior to its rivals.
For example, the best explanation for my experience of seeing a desk in front of me is surely that there really is a desk here, even if I cannot rule out other possibilities with absolute certainty. I might be dreaming, or perhaps I am living in an illusory world like The Matrix, and maybe there are other possible explanations for my experiences that haven’t even crossed my mind. Even as I acknowledge those other possibilities, the common sense explanation (that there really is a desk here) is undoubtedly the best explanation for my experience of seeing a desk.
What makes a good argument by analogy? Though it has played many important roles in philosophical discourse, argument by analogy is perhaps the most challenging type of non-deductive argument to evaluate. For an argument by analogy to be compelling, the similarities to which it appeals must be intuitively relevant, somehow. In an example given previously, I inferred that my cat has conscious experiences because the cat and I have relevant similarities. However, I also share many similarities with my office desk. The desk and I both have legs, we both are made of matter, we have similar masses, we both exist during the same time period and in the same geographic location, and so on. Can I conclude, on the basis of these similarities, that the desk is also conscious? No, because the similarities in question are intuitively irrelevant to the property of consciousness. But how can we determine, in general, which properties are relevant and which are not?
Unfortunately, it is difficult to formulate precise—yet sufficiently general—rules for determining what features of a comparison are relevant. In some cases, perhaps, we can rely on background information regarding which kinds of properties tend to be correlated, thereby supporting the analogy with an inductive inference or statistical syllogism. Alternatively, we might support the analogy using an inference to the best explanation. For example, if I think the best explanation of my consciousness has something to do with neural processes in my brain, but nothing to do with how many legs I have, then I may regard neurological similarities as relevant—and number of legs as irrelevant—to the question whether my cat or my desk has the property of consciousness. Thus, the strength of an argument by analogy often rests on the strengths of the other non-deductive inferences that support it. In many cases, however, it seems we must rely on our own intuitions and common sense to decide what is relevant and what is not.