2. Peirce on Induction and Abduction
Abduction, on Peirce’s mature view, differs in kind from induction. Induction, for Pierce, includes any use of empirical evidence to test or support a hypothesis. Abduction, by contrast, is reasoning which introduces hypotheses into inquiry in the first place: “abduction commits us to nothing. It merely causes a hypothesis to be set down upon our docket of cases to be tried” (Pierce 1932-58, §5.602).1 Abduction, in other words, aims to identify hypotheses which are worth pursuing further (McKaughan 2008). As Peirce stresses, the pursuit-worthiness of a hypothesis cannot be reduced to how plausible or probable it is; in fact: “Sometimes the very fact that a hypothesis is improbable recommends it for provisional acceptance on probation [i.e. for further testing]” (Peirce 1932-58, §6.533). For instance, if a hypothesis can be reliably falsified and if it would be valuable (given the goals of inquiry) to rule it out conclusively, increasing its improbability can make it more pursuit-worthy (cf. Nyrup 2015).
While Peirce thus distinguishes abduction from induction, he is somewhat ambiguous as to what it means for abduction to “introduce” or “suggest” hypotheses. Does it mean generating hypotheses or merely to select already generated hypotheses for further pursuit?2 Sometimes it is characterized in terms of generation. For instance: “Abduction consists in studying facts anddevising a theory to explain them”; it is “the process of forming an explanatory hypothesis” (1932-58, §5.145, §5.171, emphases added). However, abduction is also described as “the operation of adopting an explanatory hypothesis” (§5.189, emphasis added), through the characteristic inference schema (ibid. ):
The surprising fact, C ¸ is observed;
But if A were true, C would be a matter of course,
Hence, there is reason to suspect that A is true.
In order to employ this inference schema (as Peirce himself notices) a reasoner must already have formulated the explanatory hypothesisA mentioned in the second step, seemingly contradicting the idea that abduction is the process through which A is generated.
This tension can be resolved by interpreting abduction as encompassing both the generation and the selection of hypotheses to be pursued.3 The reason it makes sense to include both under the label ‘abduction’ is that they share a common goal, namely to identify hypotheses worth pursuing further. As Fann (1970, 41-43) argues, simply formulating new hypotheses is not difficult. Rather, since we cannot examine every conceivable hypothesis, the problem is to generate hypotheses it is worth spending time and resources considering further. Generative reasoning should thus be evaluated along the same dimensions as reasoning concerned with selecting hypotheses for pursuit. The above schema outlines one salient criterion for pursuit-worthiness, namely whether the proposed hypothesis accounts for the phenomena we seek to explain. But other criteria—e.g. testability and economy—are relevant too; the schema does not exhaust what Peirce meant by abduction (Pietarinen and Bellucci 2014, 355-6).
This interpretation also allows us to answer two objections, often raised against normative accounts of generative reasoning.4 First, since we do not control which ideas occur to us, how can anything normative be said about hypothesis generation? The answer is that one can still choose when ,whether and how to generate new ideas, and these choices can be normatively evaluated in terms of how efficiently they generate pursuit-worthy hypotheses. For example, a rather ineffective strategy would be to flip through a medical lexicon, hoping to chance upon diseases with symptoms similar to the ones observed in the patient. Experienced physicians will (one hopes) be able to deploy better strategies for generating hypotheses. We discuss some of these in the clinical case in Section 4.
Second, why assume that there is any particular kind of reasoning (viz. “abduction”) through which physicians generate hypotheses? The answer is that we do not assume this. For us, the term ‘abduction’ does not refer to any particular process of reasoning. Rather, the distinction between abduction and induction refers to a difference in goals: abduction aims to identify hypotheses worth pursuing further; induction aims to determine the likelihood of a hypothesis. Similarly, although the generation and selection of hypotheses for pursuit share a normative foundation, the concrete (e.g. verbal, mental, symbolic) reasoning processes involved in generating or formulating a hypothesis of course differ from those involved in choosing between already formulated hypotheses. Trying to overcome the problem of having too fewhypotheses is still different from trying to overcome the problem of having too many . The two can still be distinguished.
A particularly useful, unified interpretation of abduction, which has been proposed in recent scholarship (Hintikka 1998; Paavola 2004; Pietarinen and Bellucci 2014), construes abduction in terms ofstrategic reasoning . On this interpretation, individual choices about hypothesis generation and pursuit should be evaluated in terms of whether they contribute to an overall investigative strategy. We will explain this interpretation in more detail and show how it helps make sense of our clinical case study in Section 6.
In summary, we draw the following lessons from these Peircean ideas. First, we distinguish three general types of reasoning in medical diagnosis: (1) generating plausible diagnostic hypotheses; (2) selecting and prioritizing these for pursuit, i.e. for further testing and consideration; (3) accepting or rejecting diagnoses, or more broadly, evaluating the how likely different diagnoses are in light of the available evidence. Second, we take (1) and (2) to share the same goal, viz. pursuit-worthiness. Third, thinking about diagnosis in terms of strategic reasoning provides a useful unified framework for evaluating the generation and selection of hypotheses. We will now describe in more detail how these lessons apply to medical diagnosis.