Darwinian-Selectionist Explanation, Radical Theory Change, and the Observable-Unobservable Dichotomy

ABSTRACT

In his recent 2018 book, Resisting Scientific Realism, K. Brad Wray provides a detailed, full-fledged defense of anti-realism about science. In this paper, I argue against the two main claims that constitute Wray’s positive and novel argument for his position, viz., his suggested Darwinian-selectionist explanation of the success of science and his skepticism about unobservables based on radical theory change. My goal is not wholly negative though. Instead, I aim to identify the type of work that an anti-realist like Wray would need to undertake in order to further substantiate their position, viz., taking a stance on inductive inference and support, and the type of realist and anti-realist positions that seem viable.

View addendum:

Induction, Rationality, and the Realism/Anti-realism Debate: A Reply to Shech

Acknowledgements

I am grateful to Guido Bacciagaluppi, Michael Watkins, and Thomas Lockhart for comments on an earlier version of this paper, written while I was a Senior Fellow at Descartes Centre at Utrecht University in the summer of 2019. Thanks also to students in my ‘Challenging Physicalism & Realism’ (Fall 2020) for helpful conversations.

Notes

1 I will not defend Norton’s theory or claims here. He dedicates two books (Norton 2021; Manuscript) and many papers to the matter (e.g., Norton 2003, 2014). That a formal approach to induction is ostensibly misguided is also identified in Goodman’s (1955) ‘new riddle of induction.’

2 For worries about an infinite regress problem see Norton (2014) and Norton (Manuscript).

3 For instance, see Lipton (1991/2004), Leplin (1997), Psillos (1999), and Schindler (2018).

4 See Van Fraassen (1980, 39–40) for his original presentation.

5 I’m setting aside a well-known worry to the extent that such ‘explanations’ are tautological. In the context of the biological sciences, there is usually a deeper story to tell in terms of traits, fitness, etc., with a corresponding debate in the philosophy of biology literature such as how to characterize fitness (see, e.g., Rosenberg and Bouchard 2015; Heine and Shech 2021). Charitably, I take it that the Darwinian-selectionist explanation suggested by Wray (2018) is part of the deeper sociological story to be told about rational and scientific practices.

6 See, for instance, Shapiro (2007) for a discussion of linguistic-semantic, modal, and epistemic accounts of deductive inference.

7 Note the emphasis on ‘generally.’ It is not claimed that ‘all bismuth samples are uniform’ since this would turn the inductive argument into a deductive one. Some elements like sulfur have different allotropic forms with different melting points (Norton 2003, n1).

8 Another approach is to admit ‘ … that there might be lucky flukes where false theories have (for mysterious reasons) managed to get things right, but that in the majority of cases the success of theories is best explained by their approximate truth’ (Schindler 2018, 47).

9 The following includes past criticisms (Wray 2018, 161–163; original emphasis), wherein it is claimed that the selectionist explanation for the success of science:

(1) can only explain past successes and gives us no reason to think that theories that have been empirically successful in the past will continue to be successful in the future. (Blackburn 2005, 178; Lipton 1991/2004, 194; Psillos 1999, 97)

(2) is compatible with a realist explanation for the success of science, so it is not a threat to the realist explanation. (Kulka 1996, S299; Lipton 1991/2004, 193)

(3) is not sufficiently deep, for it does not explain what is common to all empirically successful theories. (Kitcher 1993, 156; Psillos 1999, 96)

(4) does not explain why any particular successful theory is successful. (Leplin 1997, 9; Lipton 1991/2004, 194)

(5) cannot account for the fact that we have any successful theories at all. (Blackburn 2005, 179)

In my view, a lot of the criticism is misguided. For example, consider (5). Wray’s account for why we can have any successful theories in the first place consists of noting that since predictive success is relative to accepted standard, ‘it is more or less guaranteed that the theories that scientists accept will be successful to some degree.’ Although I find this line of reply unsatisfying, it is not clear to me that a realist explanation will do better. That approximately true theory affords (or even best explains) successful prediction does not imply (or even make likely) that we will be able to discover and construct predictively successful theories in the first place. What best explains that we have predictively successful theories at all? The fact that we happen, fortuitously, to live in a world that is conducive to inductive inferences. This need not have been the case. It is at least conceivable that our world was so extremely chaotic that most prediction would fail, even if we somehow had true theories.

10 I set aside stubborn, fictional characters such as the Tortoise in Carroll’s (1895) parable.

11 Of course, there will be some counterexamples since there are occasions in which the interests of scientist shift away from predictive success. However, looking to modern science as a whole, successful prediction and retrodiction, or, empirical adequacy and strength, is necessary for the acceptance of (let alone belief in) theories.

12 See my Shech (Forthcoming) for further discussion of what powers logical inference in the context of the MTI.

13 Or, if you believe that (say) Bayesianism is the one true formal theory of induction and confirmation, one can identify the Bayesian machinery as warranting inductive inferences.

14 Psillos (2020), for instance, disagrees; while Stanford (2020) (following Leplin (1997)) claims that the selectionist explanation provides only part of the answer, viz., why theories survive.

15 One may object that the scientific realism debate isn’t about what one should rationally endorse. After all, Van Fraassen and Wray don’t argue that it's irrational to be a scientific realist. Nor need a scientific realist argue that it's irrational to be a constructive empiricist. Both sides can recognise the rationality of one another's positions while nonetheless disagreeing about which is superior. In reply, and first, all that matters for my argument is that realists think that one ought to believe in claims like ‘electrons exist’ because they are likely to be approximately true given the evidence that supports our best science, while anti-realist hold either that one need not believe such claims or that such claims are not likely to be true. Second, looking to Van Fraassen (1980, 2004), I maintain that the issue at heart is fundamentally about whether it is rational for an anti-realist to remain agnostic about unobservables.

16 Thank you to an anonymous reviewer for raising this objection.

17 In other words, what I am saying here does not imply taking a stance on the syntactic versus semantic interpretation of theories. As long as theories imply a set of propositions that can be true/false then we can put logic, including Norton’s inductive logic, to work.

18 For details, see Norton (2021), especially chapters 12-16. It is worthwhile to note that there are various contexts where Bayesian confirmation theory will not be applicable because it isn’t licensed by the relevant background facts.

19 It may be worth noting that as a deductive argument that starts from radical theory change and concludes that we ought to be anti-realists about unobservables, the argument isn’t valid. Wray is correct to note that if ‘theories are only partial representation of the world’ and if ‘scientist will be led to investigate phenomena that the accepted theories are not fit to account for’ this will result in the emergence of new theories (Wray 2018, 186). Analogously, maps are also partial representations and our changing interest in, say, different regions implies that there will be many maps. However, none of this implies that our theories or maps will be discarded since they may form a consistent whole. Worse, if theories are discarded it isn’t clear why it is only the unobservable parts that are discarded.

However, Wray’s (2018) conclusion (on p. 202) suggests that radical theory change is problematic for the realist in a similar manner to the (in)famous pessimistic meta-induction, which holds that since the history of science is a history of discarded theories then, by induction, current theories are likely to be discarded too. Crucially, he seems to think that it is the responsibility of the realist to ‘identify some significant difference between today’s theories and past theories,’ or between today’s scientists and past scientists, in order to block the pessimistic inference (Wray 2018, 93; but also see 96-97). In contrast, I maintain that, insofar as such arguments are inductive then it is the anti-realist who has to identify background facts having to do with similarities between today’s theories (or scientists) and past theories (or scientists) that warrant the pessimistic inference (Shech 2019). Since I don’t think that such background facts can be found, I submit that Wray’s argument isn’t cogent.

20 The observable-unobservable distinction is vague, but it is still categorical:

That ‘observable’ is a vague predicate. There are many puzzles about vague predicates, and many sophisms designed to show that, in the presence of vagueness, no distinction can be drawn at all. In Sextus Empiricus, we find the argument that incest is not immoral, for touching your mother's big toe with your little finger is not immoral, and all the rest differs only by degree. But predicates in natural language are almost all vague, and there is no problem in their use; only in formulating the logic that governs them. A vague predicate is usable provided it has clear cases and clear counter-cases. Seeing with the unaided eye is a clear case of observation. Van Fraassen (1980, 16)

21 To be clear, the inference is warranted for all actual (past, present, and future) observed samples of bismuth, and it is also warranted for all possible samples of bismuth (even those, say, that lie far away in regions of the universe that will never be epistemically accessible to us). The inference is not warranted for a bismuth atom since single atoms do not ‘melt,’ but it is warranted for a sample of bismuth too small for the naked eye to see and such a sample for Van Fraassen (and, I can only assume, also Wray) is ‘unobservable.’ Generally, phase transitions (like ‘melting’) do exist in small systems that cannot be seen by the naked eye. Similar comments apply to the example in Section 1 from crystallography.

22 For instance: ‘On page 58, for example, he equates 'theoretical knowledge' with 'knowledge of unobservable entities and processes'. Yet it's commonplace to think that theories needn't involve unobservable entities, which will make the discussion confusing … ’ (Rowbottom 2019a).

23 Early on in his book, Wray (2018, 49) cites Van Fraassen to explain how he views anti-realism and the observable-unobservable distinction: ‘[A]nti-realists are not thoroughgoing skeptics. Anti-realists are skeptical, but only in a circumscribed manner. Specifically, they are skeptical about: (I) the claims our theories make about unobservable entities and processes (see, for example Van Fraassen 1980) … ’ Later, he cites Stanford (2006, 3) and says that ‘[b]y the term ‘unobservable,’ I merely mean to capture the range of entities that [are] ‘too fast or too slow or too rare or take place on too grand a scale for us to engage with in ordinary ways … ’’ (Wray 2018, 100). The ‘too fast or too slow’ and ‘too grand a scale for us to engage with in ordinary ways’ remarks suggest Van Fraassen’s notion of the observable-unobservable distinction. But the ‘too rare’ remark allows for a modified interpretation. Vickers’ (2020, 15) argues that there is indeed a slight difference between Wray and Van Fraassen on the observable-unobservable distinction, and that this ‘ … has dramatic consequences for one’s degree of scientific scepticism.’ Still, his suspicion is that ‘Wray does not actually wish to depart from van Fraassen so dramatically; if he really did, then he surely would not have referenced van Fraassen in the way he does … ’ Ultimately, my point about the observable-unobservable distinction not making a difference for evidential-inductive support and inference holds on either interpretation (and, in fact, it is likely stronger if Wray moves away from Van Fraassen as Vickers suggests).

24 Objects greater than 0.1 mm in size are typically visible to the naked eye, but bismuth nanoparticles can be as small as 40–50 nm (cf. Zhao, Zhangb, and Dang (2004)).

25 The concept of thick epistemic access is credited to Jody Azzouni (1997, 474–477; 2004) and is contrasted with thin epistemic access, which is the access that one has to an object through a theory that has five virtues: (i) simplicity, (ii) familiarity, (iii) scope, (iv) fecundity, and (v) success under testing (479).

26 Interestingly, Wray (2020) seems to recognize that the problem with certain scientific claims is a lack of evidence and an abundance of inductive risk. For example, after noting Wray’s lack of clarity on the observable-unobservable distinction, Vickers (2020, 14) asks: ‘‘Does Wray think we know that the outer core of the Earth is liquid metal?’’ In reply, Wray (2020, 37) notes recent geological developments regarding the Earth’s outer core and says: ‘I am led to conclude that geologists are finding new and surprising things about what lies beneath the Earth’s surface … Hence, Vickers’ confidence that our theory of the Earth’s core is not likely subject to change in the future is unwarranted. Where he sees settled knowledge, I see uncertainty.’ Fair enough, perhaps we shouldn’t be realists about such issues but, of course, it doesn’t follow from this that there is a categorical difference between an ‘observable’ and an ‘unobservable’ (as Wray, following Van Fraassen, maintains), whereby one is rational in being a realist about the former and an anti-realist about the latter.

27 This is so since, for many aspiring anti-realist/empiricists, epistemic voluntarism may be too radical of a position to embrace, and it isn’t clear how the type of inductive skepticism and blatant relativism that lurks in the background can accommodate the scientific enterprise.

28 See Salmon (1984, 213–227) for Jean Perrin’s argument for the existence of atoms and Norton (2000) for a similar explication with the existence of the electron in the work of J. J. Thomson (what Norton calls the overdetermination of physical constants) and Niels Bohr (what Norton calls demonstrative induction).

29 For example, Alan Baker (2005) and Mark Colyvan (2010) use abduction to argue that the scientific realists ought to also be a mathematical Platonist, and Baker (2016) appeals to theoretical virtues such as ‘scope generality’ and ‘topic generality’ in countenancing realists-nominalist objections.

30 Cf. Van Fraassen (2001, 163): ‘The point of constructive empiricism is not lost if the line is drawn in a somewhat different way from the way I draw it. The point would be lost only if no such line drawing is considered relevant to our understanding of science.’

31 Of course, one may frustratingly object: ‘but this is all that Wray and Van Fraassen want!’ My reply is that their writings clearly indicate that they want more—they want to rationally maintain agnosticism (or skepticism in Wray’s case) about unobservable aspects of reality (like atoms and electrons) for which the inductive-empirical evidence is overwhelmingly strong. I don’t see how this can be done unless one also embraces a more radical skepticism than either Wray or Van Fraassen wish to, and this is so even in sprit of Wray’s (2018) novel contributions, which is what I have argued for up until now.

32 Admittedly, there will be concerns, for instance, that ‘going local’ will merely reduce the debate to questions that only scientists, not philosophers, can address (Dickens 2016).