Why phylogeneticists should care less about P opper's falsificationism

Sir, Again and again, biologists have based their argumentation on Popper's falsificationism to defend their position—especially in the context of phylogenetics and biological classification. For instance, during the debate regarding whether phylogenetic systematics or evolutionary classification represents the best method of classification, proponents of either side based their argumentation on Popper (e.g. Bock, 1973; Wiley et al., 1975; Kitts, 1977; Cracraft, 1978; Platnick and Gaffney, 1978). Popper has also been referred to in the discussion about the justification of the best method of reconstructing phylogeny by proponents of both parsimony and likelihood alike (e.g. Farris, 1983; Kluge, 1997; Siddall and Kluge, 1997; Faith and Trueman, 2001; de Queiroz and Poe, 2001; Vogt, 2007). And now biologists refer to Popper again, this time in the discussion regarding the philosophical justification of the Bayesian approach in phylogenetics (Helfenbein and DeSalle, 2005; Randle and Pickett, 2010; Farris, 2013). I have already argued elsewhere that with respect to character state distribution statements, cladograms represent unfalsifiable hypotheses (Vogt, 2008). If cladograms would be falsifiable by character statements, we would have some characters that could not be mapped and optimized for some particular cladogram topologies. This is not the case, because any given character state distribution can be explained by apomorphy and homoplasy alike, and both explanations are consistent with the relevant background knowledge of descent with modification. Only if a cladogram together with relevant background knowledge would necessarily prohibit homoplasy, would characters would represent putative falsifiers of cladograms. Since this is not the case, no character state distribution necessarily (i.e. in a deductive sense) contradicts any given cladogram topology, and therefore Popperian falsificationism is irrelevant in the context of cladogram hypothesis testing. Moreover, since likelihood, parsimony, and Bayesian approaches in phylogenetics attempt to test competing cladogram hypotheses, they cannot be based on Popperian falsificationism. Whereas this argumentation was focused on cladogram testing, one can question the relevance of Popperian falsificationism for scientific methodology in general and for historical sciences in particular. When discussing falsificationism, it should be in the context of its own premises and whether it meets its own claims. In the Logic of Scientific Discovery, Popper (1959) introduces his approach by distinguishing between the psychological question of how we discover hypotheses (context of discovery) and the logical–philosophical question of how we justify our hypotheses (context of justification). Popper argues that only the latter is of relevance to the scientific method. Thereby, Popper claims that empirical science must be grounded in experience and that justification is at the heart of a demarcation criterion of empirical science against all kinds of pseudosciences, such as Marxism or Freudian psychoanalysis. Any attempt to justify induction leads to a trilemma (see also Hume's problem of induction): (i) circular reasoning if induction is justified by induction; (ii) infinite regress if it is justified by a higher-order principle of induction, because the higher-order principle must be justified itself; or (iii) a priorism/conventionalism/dogmatism if a principle of induction is considered to be justifiable independently of experience. As a consequence, one cannot prove the truth of any empirical scientific hypothesis—verification of empirical scientific hypotheses is not possible. Therefore Popper claimed that a method of justification of empirical scientific hypotheses must be free of induction, and he argued that making induction part of the foundation of the scientific method would be unscientific. This is one of the most important premises of Popper! Popper suggested to restrict justification to deduction, and proposed falsifiability as the demarcation criterion of science from pseudoscience, which rests on a specific form of deduction called modus tollens. According to modus tollens, I can conclude that if my hypothesis “all things in this bag are white beans” is true, I can predict that if I take something out of this bag, it will necessarily be some white bean. However, if I happen to take something out of this bag that is not a white bean, I know that my initial hypothesis is necessarily false, because my experience contradicts the prediction. Thus Popper claimed that only those hypotheses that can be contradicted by experience are scientific hypotheses, because they are falsifiable (Popper, 1959). And Popper claimed that this approach does not imply any inductive elements. Much has been criticized about Popper's approach, for instance that falsificationism itself not falsifiable. Popper himself admits this, but he also argues clearly that it does not have to be falsifiable, because it is not an empirical hypothesis but a philosophical or logical theory of the scientific method (Ladyman, 2002). So this accusation misses its target. But there are two general problems with his notion of falsification and a major problem with his notion of corroboration that I want to address in the following. Popper claims that empirical scientific hypotheses must be tested against experience and that this test should be based on modus tollens and thus exclusively on deduction. Popper argues that hypotheses represent general/universal statements, whereas experience is represented in the form of singular existential or basic statements. The former have the form of “all-sentences”, whereas the latter have the form of “some-exist-sentences”. However, Popper also claims that basic statements are not infallible but themselves theory laden, because perception is an active process that involves interpretation of the observer and therewith reference to some theoretical framework (Popper, 1959). As a consequence, basic statements cannot be verified by experience but must be tested themselves, or as Popper (1959) suggested, they must be accepted by an act of free decision. This, however, poses a serious problem for Popper's theory because hypotheses are no longer testable if one cannot decide whether a basic proposition that would, if true, falsify a hypothesis is actually true (Thornton, 2009). As a consequence, just like induction, falsification leads to either infinite regress or some sort of conventionalism, and therefore does not provide a method of justification that is superior to induction. Popper (1959) argues that if a hypothesis has been tested critically, either it will be falsified or it will gain corroboration. The degree of corroboration gained through a test depends on the test's severity—the amount of falsifiers the test potentially accredits. The problem is that the disjunction of falsification and corroboration is not as strict as Popper suggests, as non-falsification is not necessarily corroboration, and vice versa (Thornton, 2009). Moreover, scientific theories are usually very complex and cannot be falsified by a single isolated observation (which would be naïve falsificationism). Neither can they themselves be tested in isolation from other theories, because each test rests on additional background assumptions that have to be tested themselves as well. Moreover, tests usually involve some ceteris paribus clause (i.e. there are no disturbing uncontrolled factors) as an additional premise that is necessarily open and thus cannot itself be tested (see also Lakatos, 1970). As a consequence, we can test only the hypothetico-deductive setting as a whole (Vogt, 2008), not single components of it: whenever we experience a falsifying experimental result, we can conclude only that the entire set that has been tested, consisting of hypothesis, background knowledge, ceteris paribus clause, and observations, as a whole cannot be true. While we can conclude that at least one of the above-mentioned components is necessarily false, we cannot decide which one. It could be any single one, several, or all of them (see sophisticated methodological falsificationism; Lakatos, 1970). One could also argue that, as neither background knowledge nor observations and ceteris paribus clauses are verifiable in principle, actual falsification of a hypothesis is not possible. This problem is known as the Duhem–Quine thesis (Duhem, 1906; Quine, 1951). Popper argues that although actual falsification is not possible (Popper, 1983), his demarcation criterion of falsifiability is not affected because falsifiability refers to the logical and not the epistemological status of a hypothesis. However, it affects his method of justification, which is also at the heart of the methodological discussions in biology. According to Popper (1959), the corroborability of a hypothesis indicates the highest possible degree of corroboration a hypothesis can gain. Because the corroborability of a hypothesis is inversely proportional to its absolute logical probability, it equals the testability of the hypothesis, which equals its empirical content measured by the amount of potential falsifiers and thus its degree of falsifiability (Popper, 1959, 1983). The idea of logical probabilities takes a central position in Popper's theory: highly improbable hypotheses have the potential to gain high degrees of corroboration. Unfortunately, even Popper himself never gave a consistent operational account of how to quantify logical probabilities in actual research practice. This, however, makes their application for evaluating the amount of accredited falsifiers of a test impossible, and thus also the quantification of degrees of corroboration gained by successfully passing a test. Hitherto, every attempt in actual research practice to provide a method for evaluating logical probabilities has failed (Schroeder-Heister, 1998), and no objective procedures have been proposed for assigning logical probabilities to propositions (Mahner and Bunge, 1997). As a consequence, Popper's theory of corroboration is not quantifiable, which makes its application in actual research practice problematic, to say the least. Popper (1959) held the view that repeating a test that a hypothesis already successfully passed beforehand does not increase its degree of corroboration to the same extent as it did when passing the test the first time. In other words, repeating tests yields diminishing gains in corroboration. Whereas this seems to be reasonable, because otherwise one could take the most severe test and repeat it endlessly to gain maximum corroboration, it entails the assumption that multiple repetitions of a test will probably yield identical results. This, however, represents an implicit inductive assumption of a general regularity underlying nature, and points to an inductive element in Popper's theory of corroboration. But Popper claimed that empirical science must be grounded in experience and that the scientific method must include a theory of justification that is free of inductive elements. Obviously, Popper's theory of corroboration does not meet his own premises and claims. Without his theory of corroboration, however, scientists would have no method for justifying their choice of a currently best hypothesis that can be used in practice and that is consistent with Popperian falsificationism, thus rendering falsificationism not only inconsistent but also inapplicable in science in general. Apparently Popperian falsificationism does not live up to its own claims. Popper's major criticism against the use of induction for a method of justification of empirical scientific hypotheses also applies to his falsificationism. In its strict doctrine, falsificationism is inapplicable not only in the context of historical sciences such as phylogenetics, but in science in general. As a consequence, phylogeneticists should stop referring to Popperian falsificationism in defending their position, at least if they do not want to erect a purely rhetorical argument from authoritarianism. If phylogeneticists want only to appeal to the ethical maxim implicit in Popper's work—that scientists should always strive for the refutation of a hypothesis as a result of testing it against experience, instead of striving for an increase of its degree of confirmation and support—this concerns less the choice of method for cladogram analysis (parsimony vs. likelihood vs. Bayesian inference), but rather a choice of character sampling and character coding. Moreover, if phylogeneticists want to emphasize the importance of recognizing the general impossibility of verifying empirical scientific hypotheses or the importance of testing them empirically, they do not necessarily have to refer to Popper and dedicate the methodological foundation of an entire discipline exclusively to his theory. Awareness that the verification of universal empirical statements (X exist and all X are q) is impossible in principle is already entailed in Hume's problem of induction. Also, the role of hypothesis testing in a hypothetico-deductive setting (see hypothetico-deductive method; Hempel, 1966), by deductively inferring necessary predictions that can be tested in an experimental setting, has been proposed before by Peirce (“a question put to nature”; Peirce, 1934/1935)—not to throw in another authority. Peirce also acknowledged the important role of this sort of hypothesis testing and the knowledge gained if Nature replies “No!”. Peirce also made a clear separation between hypothesis generation on the one hand and hypothesis evaluation (justification) on the other, However, contrary to Popper, he did not conceive this evaluation step as an exclusively deductive procedure, but rather as an inductive step (Peirce, 1934/1935). This seems to be more in concordance with what scientists actually do. Apparently not everything that is part of Popper's theory has been proposed by him. There have been theories of justification and hypothesis testing before Popper. Neither does everything that is part of Popper's theory depend on his main idea of falsificationism. Many general ideas can be used outside his doctrine. Therefore it is time for phylogeneticists to stop hiding behind the names of famous philosophers and their isms, thereby often erecting rather rhetorical arguments from authoritarianism, and instead to start taking the lead again in developing their own specific methods of justification and hypothesis testing—methods that meet the particular demands and limitations of their discipline. Looking for inspiration within philosophy is reasonable, for sure, but following a single doctrine without keeping the mind open for alternatives is definitely not.