Has Laudan Killed the Demarcation Problem? Kirsten Walsh Bachelor of Arts (Hons) (Melb) Submitted in partial fulfilment of the requirements of the degree of Master of Arts (with Advanced Seminars component) History and Philosophy of Science The University of Melbourne Australia October 2009 Produced on archival quality paper.
Abstract The Demarcation Problem is to mark the boundary between things that are scientific and things that are not. Philosophers have worked on this problem for a long time, and yet there is still no consensus solution. Should we continue to hope, or must we draw a more sceptical conclusion? In his paper, The Demise of the Demarcation Problem, Larry Laudan (1983) does the latter. In this thesis, I address the three arguments he gives for this conclusion. The Pessimistic Induction: From the failure of many specific past attempts at demarcation, Laudan infers that all future attempts at demarcation will fail. For his argument to be fully convincing, Laudan needs to show that each attempt has been a complete failure, and that these failures have never led to progress in the theory of demarcation. I argue that many past attempts at demarcation have only resulted in partial failure, and many of these failures have led to some cumulative progress. So I think we can draw a more optimistic conclusion: future attempts at demarcation may be even more successful than past attempts. The Pseudo-Problem: Laudan argues that the demarcation problem presupposes an epistemic invariant : something common to all and only the sciences, which makes them epistemically special. But, says Laudan, this presumption is false so, by definition, the issue is merely a pseudo-problem. I find Laudan s argument unconvincing. I present reasons for thinking that the demarcation problem does not, in fact, presuppose an extremely simple epistemic invariant. Furthermore, there may still be a satisfactory, moderately complex epistemic invariant to be found. So I do not think any false assumption is presupposed. The New Problem: Laudan argues that we should replace the original demarcation problem with a new demarcation problem. I take this to be the problem of demarcating between well-confirmed and ill-confirmed theories. I argue that scientific status is relevant to the confirmation of theories, so the two problems are closely related. I also argue that science has other purposes; so scientific status indicates other virtues besides wellconfirmedness. Thus we do want to know which theories and activities are scientific, because this will help us to decide which theories and activities to pursue. So this new demarcation problem is not a suitable replacement for the original problem. My central question is Has Laudan killed the demarcation problem?, and my answer is No!. 1
Declaration This is to certify that (i) (ii) (iii) The thesis comprises only my original work except where indicated in the Preface; Due acknowledgement has been made in the text to all other material used; The thesis is 20,000-22,000 words in length, inclusive of footnotes but exclusive of tables, maps, appendices and bibliography. Kirsten Walsh 2
Acknowledgements This thesis is the culmination of my academic journey thus far. Starting out as a vague question, Is the demarcation problem worth solving?, it evolved into its present form. I would like to thank a number of people who have contributed to the final result in many different ways: First and foremost, I sincerely thank my long-suffering supervisor Howard Sankey, who has ploughed through various drafts, making critical suggestions and posing challenging questions. He has motivated and encouraged me every step of the way never reproaching me when the necessity of doing paid work got in the way of progress. For this, I am extremely grateful. Special thanks go to Neil Thomason for inspiring me in the early stages, encouraging me to present my work at conferences, and for reading an entire draft of my thesis during a flight to the US! I also thank Jason Grossman for his kind comments and helpful criticisms of Chapter One. This thesis evolved into its present form during a summer I spent at the University of Auckland under the supervision of Robert Nola who went above and beyond the call of duty. I thank him not only for his helpful discussions and feedback, but also for showing me around Auckland and the surrounds. I also thank Jan Crosthwaite, Rosalind Hursthouse, and the University of Auckland for providing me with the Summer Research Scholarship that made this work possible. For providing numerous distractions, as well as their unfailing support, I thank the History and Philosophy of Science (HPS) and Philosophy postgrads with whom I have shared office space, morning tea, and jugs of beer. In particular I d like to thank: Conrad Asmus for teaching me formal logic and reading a draft of my thesis; Alex Murphy for insisting that I use correct grammar and punctuation; Tama Coutts for being interested in everything philosophical; David Condylis for insisting that I should never trust a philosopher named Larry ; Ned Taylor for telling me what physicists do; Suzy Killmister for securing office space; and Kristian Camilleri and Steph Lavau for proving that theses can be finished. Special mention goes to the participants of Help! My Progress has Stalled! and Operation Endless Victory : Aaron Retz, Bryan Cooke, Chris Soeterboek, Paul Carter, Sergio Mariscal, Steph Lavau, Steven Kambouris, and Vicki Macknight I hope we are all victorious in the end! 3
A very special thank you to Erik Nyberg, my partner, colleague and friend. He always supported and encouraged me, stayed interested in the details of my interminable project, and made many helpful suggestions especially by playing the devil s advocate during the editing of my final draft. Finally, I thank my parents, Adrian and Meredith Walsh, who have offered their unconditional support and gentle counsel at every turn of the road. Their foresight and values paved the way for my privileged education. I dedicate this thesis to Larry Laudan, for all the time we ve spent together. 1 1 I haven t met Professor Laudan yet, but I hope to meet him one day! 4
Table of Contents Abstract... 1 Declaration... 2 Acknowledgements... 3 Table of Contents... 5 Table of Figures... 8 0 Introduction... 9 0.1 What is Demarcation?... 9 0.2 Laudan s Three Sceptical Arguments... 11 1 The Pessimistic Induction... 13 1.1 Laudan s Pessimistic Induction... 13 1.2 Inductive Inference... 16 1.3 The Resemblance Assumption should be Rejected... 19 1.4 Progress in the Philosophy of Method... 20 1.5 Reply: Scientific Method Changes... 23 1.6 Reply: The New Tradition is not Epistemic... 27 1.7 Rejoinder: Testability is Epistemic... 28 1.8 Rejoinder: The New Tradition is Progressive... 29 1.9 Conclusion... 32 2 The Pseudo-Problem... 34 2.1 Laudan s Requirements and Pseudo-Problem... 34 2.2 Requirement One: Accuracy... 36 2.2.1 Objection: Demarcations can be legislative... 36 2.2.2 Sub-conclusion: Reasonable accuracy is sufficient... 37 2.3 Requirement Two: Precision... 37 2.3.1 Objection: Precise enough for the specific purpose... 38 2.3.2 Objection: Ordinary vagueness could be replicated... 39 2.3.3 Sub-conclusion: Moderate precision is sufficient... 40 2.4 Requirement Three: Epistemic Superiority... 40 2.4.1 Objection: Only epistemic significance... 41 5
2.4.2 Objection: Indirect epistemic virtues... 42 2.4.3 Reply: Too many virtues... 43 2.4.4 Reply: Too unreliable... 44 2.4.5 Sub-conclusion: Indirect epistemic virtues are sufficient... 45 2.5 Requirement Four: Invariance... 45 2.5.1 Variance and complexity... 45 2.5.2 Objection: An extremely complex invariant... 48 2.5.3 Reply: The demarcation criterion must be simple... 49 2.5.4 Rejoinder: Demarcation needn t be extremely simple... 50 2.5.5 Sub-conclusion: Moderate invariance is sufficient... 52 2.6 The Moderate Demarcation Criterion... 53 2.7 The Existence of the Epistemic Invariant... 53 2.7.1 Objection: Rules don t vary that much!... 56 2.7.2 Objection: Ultimate goals might not vary... 57 2.7.3 Sub-conclusion: Deep epistemic homogeneity... 58 2.8 Conclusion... 58 3 The New Problem... 61 3.1 Laudan s New Problem... 61 3.2 Accounts of Confirmation... 62 3.2.1 Laudan s views... 62 3.2.2 A simple account... 63 3.2.3 Sophisticated accounts... 64 3.2.4 Overcoming underdetermination... 65 3.3 Objection: Science is Relevant to Confirmation... 66 3.3.1 Science is more than confirmation... 66 3.3.2 Types of relevance... 66 3.3.3 No exhaustive relevance... 67 3.3.4 Causal relevance... 68 3.3.5 Statistical relevance... 69 3.3.6 Logical relevance... 71 3.3.7 Sub-Conclusion: Scientific status is strongly relevant... 71 3.4 Objection: Science has Other Purposes and Other Virtues... 71 3.4.1 Improving our understanding requires novelty... 72 3.4.2 Improving our living standards requires usefulness... 72 3.4.3 Achieving confirmation requires testability... 72 3.4.4 Pre-selection before confirmation... 73 3.4.5 Other virtues can outweigh confirmation... 74 3.4.6 Reply: Pursuit versus acceptance... 74 6
3.4.7 Rejoinder: Demarcation is still useful for pursuit... 75 3.4.8 Rejoinder: Other virtues may affect acceptance... 76 3.4.9 Reply: Science is all about confirmation... 76 3.4.10 Rejoinder: Then science is relevant to confirmation!... 77 3.4.11 Sub-conclusion: Being scientific indicates other virtues... 77 3.5 Conclusion... 78 4 Conclusion... 79 Appendix A Some Non-Ideal Definitions... 80 A.1 Ideal Definitions... 80 A.2 Non-Ideal Definitions... 81 A.3 Gold... 82 A.4 Science... 83 A.5 Diamonds... 84 Appendix B Some Complex Demarcations... 86 B.1 Thagard... 86 B.2 Lugg... 87 B.3 Derksen... 88 Appendix C Other Virtues versus Confirmation... 90 C.1 Wonten versus Newton... 90 Bibliography... 93 7
Table of Figures Figure 1.1: Three possible views of demarcation.... 20 Figure 1.2: Candidate rules in plausible levels of generality.... 22 Figure 1.3: Two inadequacies of the eschew proposal.... 22 Figure 1.4: A hierarchy of goals and methods.... 26 Figure 1.5: Laudan s Old and New Traditions of demarcation.... 28 Figure 1.6: Some methodological progress since Popper.... 31 Figure 2.1: Two epistemic virtues in a positive feedback loop.... 45 Figure 2.2: My alternative requirements for a demarcation criterion.... 53 Figure 2.3: The implications of heterogeneity for the invariant.... 56 Figure 2.4: Some accepted rules and their opposites.... 57 Figure 2.5: My alternative interpretation and claim.... 59 Figure 2.6: Two interpretations and my corresponding truth values.... 59 Figure B.1 Thagard s two conceptual profiles.... 86 Figure C.1 Two Laws describing the behaviour of apples.... 90 8
0 Introduction In 1983, Larry Laudan published a paper that he called The Demise of the Demarcation Problem (Laudan, 1983), in which he argued that the problem is unsolvable. This is my reply to Laudan s paper. 0.1 What is Demarcation? To demarcate is, literally, to mark a boundary (Sykes, 1989). So a demarcation tries to sort things into two mutually exclusive groups: things inside the boundary, and things outside the boundary. If the boundary is not perfectly precise, then there may also be some borderline things that are not clearly in or out. In the philosophy of science, the problem of demarcation is to mark the boundary between things that are scientific and things that are not. 2 Several different terms are commonly used to describe the contrasting things that are not scientific, with slightly different connotations: non-scientific, pseudo-scientific, and unscientific. I do not think the differences here are important to Laudan s argument or my reply. Science has many aspects, including (but not limited to): (a) basic elements such as theories, predictions, experiments, and results; (b) technical refinements such as mathematical models and formulae, measurement tools, and statistical analyses; (c) general virtues of theories such as confirmation, novelty, simplicity, explanatory breadth and usefulness; and (d) social arrangements such as qualified experts, published journals, peer review, large institutions, and competition for funding. We can try to demarcate between what is scientific and what is not with respect to areas of knowledge, or with respect to any of these particular aspects. I take it that all these demarcations should be closely related, e.g. once we have decided that an area of knowledge is a science, we would probably describe most of its theories, methods, instruments, and experts as scientific. 2 It is possible to propose logically weaker demarcation criteria that are only one-way: they either include things as scientific, or they exclude things as non-scientific, but not both. One might also define other weaker types of criterion. But traditionally philosophers have been concerned with strong criteria that are two-way, applicable in all areas, and so on. Laudan clearly states that he is concerned with strong criteria of this kind (Laudan, 1983: 119), and strong criteria will also be my concern. Note that if a strong criterion were discovered, then this could also function as a logically weaker criterion (e.g. exclude non-sciences). So if my defence of the possibility of demarcation is successful, then it holds for strong and weak criteria alike. 9
We can sort things adequately for a specific purpose and context, but this sorting may not be adequate for another purpose or context. What purposes could a demarcation of science serve? Let us suppose that the successful demarcation is based on fundamental features that make science epistemically superior, which gives us both a deeper theoretical understanding of science and the practical ability to tell whether something is scientific. Then it could be useful in several ways. Firstly, it could be of theoretical interest to philosophers, e.g. to help them explain why science is epistemically superior. Secondly, it could be of practical interest to non-scientists, e.g. to help them decide what research to fund or who to trust. Thirdly, it could even be of practical interest to scientists themselves, e.g. to help them improve their practices. 3 Demarcation is described as a problem because it has proved to be very difficult and perhaps impossible to achieve. Philosophers of science have worked on it for many years. While many solutions have been suggested, and many philosophers think they have solved it, no solution has been accepted by all or most philosophers of science. Therefore, I shall assume, as Laudan does, that the demarcation problem is unsolved. In claiming that the demarcation problem cannot be solved, Laudan is rejecting all forms of the problem. He is not distinguishing between demarcations that contrast nonscientific, pseudo-scientific, or unscientific. He is not distinguishing between demarcations that apply specifically to areas, theories, methods, people, etc. He is not distinguishing between demarcations for one purpose or another. He is asserting that the solutions to all of these variations on the demarcation problem are either impossible or unimportant. Like Laudan, I will not focus exclusively on any of these particular variations on the problem. I take it that they are all related. However, this variety does make it harder for Laudan to establish that no useful version of the problem can be solved. Is there any difference between a demarcation and a definition? A definition of what is scientific also tries to sort things into two groups. Things are scientific iff they fit the definition; things are not scientific iff they do not fit the definition. Again, there may be borderline cases. Definitions can apply to any aspect of science, and be designed for any purpose, whether theoretical or practical. I will not assume that the projects of demarcating and defining what is scientific are exactly the same, but they are clearly similar I will argue that similar issues can arise. 3 For more detailed discussion of the purposes of demarcation see for example (Cioffi, 1970), (Gardner, 1957), (Kuhn, 1996), (Resnik, 2000), (Ruse, 1982), and even (Laudan, 1983: 111). 10
0.2 Laudan s Three Sceptical Arguments There are two possible reasons for the failure to solve the demarcation problem: 1. There is a solution out there the reason we haven t found it (or accepted it) yet is because philosophers have not been imaginative (or perceptive) enough; or 2. There is no solution to the demarcation problem that s the reason we haven t found one. Laudan favours the second diagnosis, arguing that the demarcation problem hasn t been solved because it is unsolvable. Laudan gives three main sceptical arguments against the original demarcation problem: A Pessimistic Induction None of the many and varied criteria offered so far have successfully demarcated science from non-science. Therefore, it is unlikely that there will be any future success. A Pseudo-Problem The demarcation problem presupposes an accurate, precise epistemic invariant in all and only science. There is no such feature. Therefore, the demarcation problem is a pseudo-problem. A New Problem A better alternative to the demarcation problem is to identify theories that are well-confirmed. We can (and should) evaluate confirmation without considering scientific status. Laudan concludes that terms such as pseudo-science and non-science do nothing but rhetorical work. He recommends that philosophers and scientists trade-in their jargon and rhetoric for sound argument and strong evidence. If we are serious in our quest to identify superior theories, then we should evaluate theories solely on the basis of their empirical and conceptual credentials, and their scientific status should be irrelevant (Laudan, 1983: 125). It is important to recognise that Laudan does not claim that science doesn t exist. He agrees that the terms science and non-science identify a genuine distinction, but he argues that this distinction has no philosophical and epistemological significance (Laudan, 1983: 125). In this thesis I reply to each of these three arguments in turn. My aim is modest: refuting Laudan s arguments, rather than putting forward my own criterion. Other critics of Demise have been more ambitious. They have attempted to refute Laudan s arguments by counterexample: offering their own criterion as a new solution to the problem of 11
demarcation (e.g. (Butts, 1993), (Lugg, 1987)). The problem is that unless such a criterion is accepted, the refutation also fails. In contrast to these critics, the strength of my refutation does not rely on any particular criterion being the final solution to the problem of demarcation. My central question is: Has Laudan killed the demarcation problem? 12
1 The Pessimistic Induction 1.1 Laudan s Pessimistic Induction Laudan wonders if a solution to the demarcation problem is still feasible, so he considers past attempts at demarcation in order to shed some light on this question. He initially considers the candidates proposed by Aristotle. Laudan tells us that the task of identifying genuine knowledge had been attempted even earlier. But Aristotle s focus was on scientific knowledge, and the solution he proposed was extremely influential. Laudan says that Aristotle demarcated science from craft with his criterion of knowledge of first causes. This distinguished between know-how (the kind of knowledge a craftsman has about how to build a boat that floats) and know-why (the kind of knowledge a scientist has about why the boat floats). We can only arrive at scientific knowledge about an event or behaviour (know why it has occurred) by tracing its causes back to first principles (Laudan, 1983: 112-113). According to Laudan, this candidate had a varied career. Initially, it served as grounds for dismissing certain fields of inquiry as unscientific for example, early mathematical astronomy failed to qualify as a science because it didn t yield knowledge of first causes. Instead, astronomers offered hypothetical models, which they sought to test by comparing predictions made by their models with the observed positions of the planets. Laudan tells us that it wasn t until the beginning of the seventeenth century that scholars started to question this position on the scientific status of astronomy. Galileo, Huygens and Newton wanted to give scientific status to many systems of belief that laid no claim to understanding underlying principles or knowledge of first causes. Thus, knowledge of first causes failed to become the accepted solution to the demarcation problem (Laudan, 1983: 113-114). Laudan tells us that Aristotle also identified a second, complementary demarcation criterion: he demarcated science from superstition with his criterion of apodictic certainty. He claimed that the product of scientific inquiry was demonstrably certain, i.e. infallible (Laudan, 1983: 112). Infallibility and knowledge of first causes worked together as a twopronged demarcation: science can be distinguished from non-science both by the certainty of its knowledge and by the basis of this knowledge in first principles (Laudan, 1983: 113). But after the latter was rejected, infallibility became the sole criterion of demarcation. Laudan notes that, for a while, infallibility was a great success: despite their disagreement in 13
other areas, scholars of the seventeenth and eighteenth centuries widely agreed that scientific knowledge was apodictically certain (Laudan, 1983: 114). Laudan tells us that this criterion was finally rejected when scholars noticed that existing theories were often amended or replaced by better theories this could only occur if the existing theories were false. It now seemed that few (if any) scientific theories were infallible, and philosophers were forced to conclude that scientific knowledge was fallible after all. Thus, Aristotle s apodictic certainty also failed to solve the demarcation problem (Laudan, 1983: 114-115). Laudan tells us that after the final defeat of Aristotle s twin criteria for demarcation, philosophers considered methodology as a possible replacement. They aimed to show that the scientific method, although fallible, was a better way of testing empirical claims than any other method. And if it did make mistakes, it was sufficiently self-corrective that it would soon discover them and put them right (Laudan, 1983: 115). Furthermore, this superior method was the thing that distinguished science from non-science, and made scientific knowledge epistemically superior (Laudan, 1983: 115). According to Laudan, while many philosophers believed in the methodological criterion, they could not agree on the details of this method. The candidates for the scientific method were diverse: some thought that scientists reasoned by induction; others thought that scientists restricted their theories to what could be directly observed; and still others thought that scientists preferred theories that successfully predicted novel facts. Without agreement about the details of scientific methodology, philosophers were unable to argue persuasively that methodology is what demarcates science from non-science. Laudan also notes that most of the proposed methods failed to resemble the methods actually used by scientists. So, for these two reasons, this approach failed before it got very far. Firstly, philosophers were unable to identify the scientific method (that all and only scientists used). Secondly, philosophers were unable to establish the superior epistemic credentials of any of the methods considered (Laudan, 1983: 115-116). According to Laudan, an entirely new approach to demarcation emerged early in the twentieth century. This new approach equated science with meaningfulness: scientific statements are those that have a determinate meaning. Philosophers argued that we can establish whether or not a statement has a determinate meaning by deciding whether or not the statement can be exhaustively verified. This approach became known as Verificationism: a claim is scientific iff it can be exhaustively verified (or confirmed) by empirical testing (Laudan, 1983: 120). Another candidate from the same period is 14
Falsificationism: a claim is scientific iff it can be falsified (or refuted) by empirical testing (Laudan, 1983: 121). Laudan tells us that these candidates failed in two ways. Firstly, it wasn t the case that all and only scientific statements were verifiable or falsifiable. Laudan remarks that (a) there are clear examples of scientific claims that are not exhaustively verifiable or falsifiable, (b) all non-sciences contain at least some claims that are verifiable (Laudan, 1983: 120), and (c) falsificationism appears to give scientific status to any crank theory, as long as it makes ascertainably false claims (Laudan, 1983: 120-121). Secondly, Laudan argues that the notion of testability fails to identify what is important about science: being testable-inprinciple does not make a theory worthy of belief (Laudan, 1983: 122). Laudan argues that these two candidates mark a significant shift in the approach to demarcation. Where the earlier candidates attempted to identify an epistemological demarcation, these later candidates attempted to identify a syntactic or semantic demarcation (Laudan, 1983: 121-122). Finally, Laudan wonders whether there are any promising candidates waiting in the wings (Laudan, 1983: 122). He considers the following: 4 Scientific claims are well-tested; Scientific theories exhibit progress or growth; Scientific theories make surprising predictions that turn out to be true; Science is the only form of intellectual system building that proceeds cumulatively; and Science is the sole repository of useful and reliable knowledge. He argues that none of these candidates is promising. Some scientific theories are highly speculative, and therefore untested and unreliable. Some established scientific theories do not progress rapidly or make lots of surprising, successful predictions. Finally, some scientific theories do not contain their predecessors as special cases, therefore not all scientific progress is cumulative. He concludes that none of these candidates identifies a feature that is always or only displayed by science (Laudan, 1983: 122-124). 4 Laudan does not mention his own demarcation attempt, in which he offers problem-solving effectiveness as the primary aim of science (Laudan, 1977). This is surprising, since he continues to develop this idea in later years (e.g. (Laudan, 1990b)). 15
After considering these past demarcation attempts, Laudan argues that all future attempts at demarcation will probably fail. This argument looks like a pessimistic induction, 5 which contains at least this premise and conclusion: 6 P1. All past attempts at demarcation have failed. C. All attempts at demarcation will fail. 1.2 Inductive Inference Inductive inference can be represented in the following form: 7 P1. All observed xs are Q. C. All xs are Q. Inductive arguments are not deductively valid, in that P1 does not entail C by the normal rules of deductive logic. Rather, the conclusion is ampliative: it expands on what is contained in the premises. Specifically, C has the same logical form as P1, but it has a broader range of application, because it also includes unobserved xs. 8 Hence, unlike deductive inferences, inductive inferences are fallible: P1 may be true while C is false. Although all inductive inferences are fallible, some seem more likely to fail than others. Some inductions seem good, and hence more convincing. Others seem bad, and hence less convincing. It is difficult to give precise, general rules for what constitutes a good or bad inductive inference, but I take it that sometimes we can tell the difference. 5 This pessimistic induction about demarcation attempts should not be confused with Laudan s well-known pessimistic induction to the conclusion that all scientific theories are false (e.g. (Laudan, 1981a: 121-124), (Laudan, 1984: 121), (Psillos, 1999: 101)). 6 I have three reasons for interpreting this as a pessimistic induction. Firstly, Laudan does not make the logical structure of his argument entirely explicit and clear, so to some extent I must address arguments that are implicitly suggested by what he says. Secondly, Laudan does devote a large amount of space to describing this historical sequence of failed demarcation attempts, and does conclude that all future attempts to identify an epistemic version of a demarcation criterion will probably fail (Laudan, 1983: 124). This certainly suggests a pessimistic induction, which I am entitled to discuss as an explicit argument. Thirdly, Laudan has also advanced a similar argument in the past, namely his pessimistic induction about the truth of scientific theories, so it is not unreasonable to attribute this kind of argument to him. 7 One might call Laudan s pessimistic induction a meta-induction, because it is about Philosophy rather than nature. However, I take it that meta-inductions, pessimistic inductions, and inductions about nature all have a similar form. 8 So, C does entail P1, and hence is logically stronger than P1. 16
Making this particular inductive inference is equivalent to assuming that unobserved xs will resemble observed xs, at least with respect to the property Q. I shall call this the resemblance assumption (a resemblance which is specific to each inductive argument). We can replace any application of a rule of inductive inference with an additional premise such as this, and thereby make the argument deductive. Of course, this does not make the inference any less fallible; we have merely replaced an unreliable rule with an unreliable premise. There is some dispute over what additional premises, if any, should be present to make a good inductive inference. I have used only P1 and C because they capture the basic idea, on which everybody agrees: induction extrapolates from the observed to the unobserved. Some philosophers, e.g. (Chalmers, 1999: 45-49), claim that (generally speaking) good inductive inference requires this additional premise: A large number of xs have been observed under a wide variety of conditions. Certainly, an inductive inference is unlikely to be good if we have observed only one x! If all of these many and varied xs (without exception) are Q, then one usually has good reason to assume resemblance. 9 In some cases, it is clear that the resemblance assumption is inappropriate. Consider the following two scenarios: Scenario 1: You and I are playing a game. There are four cups lined up on a table. You turn your back while I hide a coin under one of them. The object of the game is for you to guess which cup is hiding the coin. You point to the first cup and say, Is it this cup? and I say No. You point to the second cup and say, Is it this cup? and I say No. You point to the third cup and say, Is it this cup? and I say No. At this point, you throw your hands up in the air in frustration and say, I m never going to get it! Show me which cup it is. 9 P1 demands that all of the xs are Q. But statistical inferences can be described as inductive, because they extrapolate from the observed sample to the unobserved population. In such inferences, the requirement that the observed xs are many and varied is replaced by an assumption that the observed xs are a random sample from the population. The requirement that all the observed xs are Q is usually replaced by a premise that some specific proportion of observed xs are Q. The conclusion then concerns the proportion of xs in the population. But I can leave aside all the complications involved in statistical inferences, because Laudan s induction seems to take the traditional, non-statistical form. 17
You have inferred that since all of your past attempts at identifying the correct cup have failed, all of your future attempts at identifying the correct cup will fail. This inference is a bad one! By a process of elimination, on your next guess you would have identified the correct cup. Scenario 2: You are helping a child who is learning how to tie his shoelaces. On the first day, he tries and fails. On the second day, he tries and fails. On the third day, he tries and fails. At this point, you conclude that he s always going to fail, so you go out and buy him shoes with Velcro fasteners instead. You have inferred that since each of the child s past attempts to tie his shoelaces has failed, all of his future attempts to tie his shoelaces will fail. This inference is a bad one! When a child is learning to tie shoelaces, he will fail many times before he succeeds. If he keeps trying, probably the child will eventually succeed. While I do not have a general rule to tell me which inductions are good ones, I can still tell in each of these cases that the resemblance assumption is inappropriate. In both of the above scenarios, we have good reason to believe that the unobserved cases have a significant chance of being different to those already observed. In the first scenario, we know that one of the cups definitely has the coin in it. So, as each cup is eliminated, the probability that the next guess will be correct rises. Thus, each new unobserved case is more promising than the last, and the fourth guess is completely certain to be correct. In the second scenario, although the child fails to tie his shoes each morning, we hope that he is learning and improving each day. Indeed, it is reasonable to expect that over time the child will learn how to tie his shoes. One could even argue that the use of the term failed is inappropriate here. It is not nuanced enough to convey the notion that each day the child comes a bit closer to success: each day the failure is partial, not total. So, in both of the above cases, we have good reason to reject the resemblance assumption. These examples seem to illustrate a general principle. If relevant progress is being made towards success, then there is often a good chance that success will eventually arrive, despite a sequence of failures. Hence, in such cases we have good reason to doubt a pessimistic induction. In summary, inductive inference is equivalent to the assumption that unobserved xs will resemble observed xs with respect to the property Q. There are appropriate and inappropriate resemblance assumptions and sometimes we can tell the difference. Evidence of progress is one good reason to doubt a pessimistic resemblance assumption. 18
1.3 The Resemblance Assumption should be Rejected Those who believe that the demarcation problem is solvable do not believe that all attempts at demarcation will succeed; rather, they believe that an attempt will succeed. The fact that all observed attempts at demarcation have failed is disappointing, but it does not contradict this optimistic belief in eventual success. In fact, the history of demarcation attempts may be encouraging if these failures were only partial, and there is some cumulative progress. Laudan, in contrast, is arguing that the demarcation problem is unsolvable and that all attempts at demarcation will fail. This resemblance assumption is dubious if we have good reason to believe that philosophers are making relevant progress. So, to make his pessimistic conclusion compelling, Laudan must assume that each attempt at demarcation is a complete failure and that these failures have never led to progress. I shall argue in the following sections that Laudan s bleak assessment of philosophical progress is not correct. Many past attempts at demarcation have resulted in failures that are only partial, and many of these failures have led to some cumulative progress. In fact, one might make an optimistic induction: since many observed attempts have resulted in significant progress, then many future attempts will also result in significant progress. Since sufficient progress must eventually lead to success, one might then infer that philosophers will eventually succeed. This optimistic resemblance assumption leads to the opposite conclusion! However, I do not need to defend such an optimistic view in order to refute Laudan s pessimism. I only need to establish that some progress is evident. This leaves the future of the demarcation problem unclear (at worst), with a significant chance that philosophers will eventually succeed. This is sufficient to establish that Laudan s inductive argument is unconvincing. I summarise these three possible views of demarcation in Figure 1.1. I argue that Laudan s pessimistic assumptions are untrue (so I place an F for false next to these items), and therefore it is not clear that we will never succeed (so I place a question mark here). I claim that the optimist s assumptions are true (so I place a T for true next to these items), but it is still not clear that we should accept the optimistic prediction that we will eventually succeed (so I place another question mark here). I argue that a mixed assessment of the situation is warranted, and therefore we should conclude (at worst) that the outcome is uncertain (so I place a T next to all these items). 19
Assessment Local Result Global Result Future Prediction Pessimistic Complete Failure F No Cumulative Progress F Never Succeed? Optimistic Significant Success T Cumulative Progress T Eventual Success? Mixed Mixed Results T Mixed Results T Outcome Uncertain T Figure 1.1: Three possible views of demarcation. 1.4 Progress in the Philosophy of Method Laudan tells us that prior to the nineteenth century, Aristotle s Apodictic Certainty was considered to be the definition of science. However, the replacement of several welldeveloped scientific theories by new theories left philosophers with no choice but to conclude that scientific knowledge was fallible after all. From my point of view, this failure of Apodictic Certainty was progressive: it led directly (by elimination) to a correct belief about science. In any case, nineteenth-century philosophers turned to the scientific method to do the job. They thought that the method used by scientists was fallible, but nonetheless superior to methods used by non-scientists; and hence was an adequate demarcation criterion. Laudan says that for this approach to succeed, philosophers needed to complete two tasks (Laudan, 1983: 115): 1. Identify a method that all and only scientists follow; and 2. Justify this method by appealing to its superior epistemic status. According to Laudan, philosophers could never deliver on either of these tasks because of their lack of agreement about the basic tenets of the scientific method (Laudan, 1983: 115-116). Without such agreement, philosophers were unable to argue persuasively that superior method is what demarcates science from non-science. Furthermore, proposed methodological rules were either incomprehensible or too complicated to follow: Laudan identifies a rule instructing one to eschew theoretical entities (Laudan, 1983: 116) as typical of this era. Finally, Laudan tells us that the scientific method was never adequately 20
justified because philosophers had no good reason to prefer one proposed scientific method over another or to any unscientific method. Laudan is surely correct in saying that the rule one ought to eschew theoretical entities is a failure. However, he seems to assume that this kind of rule is symptomatic of the failure of the entire approach. Moreover, he appears to think that the philosophical theory of scientific methodology as an approach to demarcation ( methodology for short) was a dead-end, and that the astute philosopher should have realised this in advance (Laudan, 1983: 115). He notes that several of his contemporaries, who he regards as respectable philosophers (Laudan, 1983: 115), approach demarcation in this way; but he doesn t discuss any of their developments. Presumably, he doesn t see them as progressive. I claim that Laudan s dismissal of methodology is premature. In fact, recent developments in methodology have shed some light on the failures of nineteenth-century attempts to demarcate between science and non-science. One might wonder which less-than-astute philosophers continued to work on methodology. In fact, one of them was Laudan. One year after Demise, Laudan published a book called Science and Values (Laudan, 1984) in which he proposed a theory of scientific methodology. He developed his methodology in subsequent papers (e.g. (Laudan, 1987)). He argued convincingly that scientific method is goal-directed. Laudan claimed that methodological rules make no sense as isolated statements of the form: One ought to do x. Rather, they should be regarded as conditional statements of the form: If one s goal is y, then one ought to do x. So, particular methods can be justified by a rule such as this: As long as one s goal is y, and one believes that doing x is more likely than any alternative method to produce y, then one is justified in doing x (Laudan, 1987: 203). These points have been accepted by many philosophers of science (e.g. (Sankey, 2000), (Worrall, 1988)), and may be regarded as developments in the field. Even if philosophers didn t know what was wrong with the eschew rule at the time, we can now identify at least one problem: it s not explicitly related to a goal, and it s not clear what kind of goal would really justify this rule. Another development in methodology was made by Rosenberg (1985), who argued that there are levels of method: some rules are more general than others. We can distinguish between, say, rules that give general advice about how to proceed, and rules that specify particular actions that ought to be taken. For example, in Figure 1.2 I have set out some candidate rules arranged in plausible levels of generality. This shows us something 21
else that was wrong with the eschew rule: it is not clear what particular actions should follow from this general rule, i.e. how we could possibly implement it. Avoid forming inaccurate conclusions Avoid making inaccurate measurements Avoid experimenter and placebo effects Avoid incorrectly rejecting the null hypothesis Always calibrate a ph-meter against distilled water Always perform double-blind tests Always select a low p-value Figure 1.2: Candidate rules in plausible levels of generality. Laudan attributes the failure of the eschew rule to the fact that it involved complex conceptions which neither scientists nor philosophers of the period were willing to explicate (Laudan, 1983: 116). This is a good enough reason to reject the rule: if a methodological rule is such that we cannot tell when it is being followed and when it is being flouted, then it is of little use to philosophers or scientists. However, this third criticism (while valid) only tells us that good rules need to be spelled out more clearly. Compared to the previous two criticisms, it doesn t provide us with much new information about what methodological rules should look like. Taken on its own, the failure of the eschew rule is scarcely progressive. It eliminates one candidate rule, but this still leaves many other possibilities. There does not seem to be any good bit that we can take from it! However, philosophers have now learned some general lessons that would avoid such rules. These serve to identify at least two reasons why the eschew rule was a failure (as depicted in Figure 1.3): 1. It s not clear what kind of goal would justify this rule; and 2. It s not clear what particular actions should follow from this rule. GOAL? One ought to eschew? theoretical entities. ACTIONS Figure 1.3: Two inadequacies of the eschew proposal. 22
Presumably, Laudan selected the eschew rule because it is one of the worst rules that methodology has ever offered. Its flaws are supposed to illustrate the failure of the entire nineteenth-century methodological approach. But, as I have demonstrated, even this rule can be used to illustrate my point: since the nineteenth century, progress has been made in methodology. 1.5 Reply: Scientific Method Changes I have presented the history of change in methodology as a positive i.e. philosophical accounts of the scientific method are progressing towards a better understanding of it. However, one implication of the philosophical developments I have discussed is that scientific methods themselves may well change over time. This should occur when scientists develop better techniques for achieving their goals, and does occur frequently with statistical and experimental techniques. Laudan sees this kind of methodological change as a negative. He argues that if everything about science changes, then we cannot give an enduring definition of science (Laudan, 1987: 214). 10 If the change is completely pervasive (so that not even the nature of the changes is constant), then this is surely correct. So, if we hope to demarcate between science and non-science, then we would like our account of science to: a) Account for the variation (over time, between disciplines, etc); and b) Explain the common aspects (over time, between disciplines, etc). In (b), we would be identifying some characteristic general things that don t change. A plausible response to Laudan s objection is that the most general things don t change, even if the details do. Specifically, one might argue that we can identify an appropriate overall goal (or set of goals) for science. I shall call these ultimate goals. If the ultimate goals don t change over time and across disciplines, then this might be sufficient. These goals would need to be very general in both their applicability and attractiveness, since they would need to be goals to which all scientists are at least superficially committed, despite the disparity of their particular practices. There is one obvious candidate: a complete, true theory of the world (and preferably, one that is easy to understand!). The goal of truth is applicable to all areas of science, because regardless of which aspect of the world is under investigation, scientists can develop theories attempting 10 This is not an argument he makes in Demise. However, because it is relevant to his pessimistic comments on methodology, I shall digress briefly to examine this argument. 23
to describe this aspect (which are then true or false). Truth should also be very attractive in any area, for two reasons. Firstly, a true theory offers genuine insight and understanding of the world, which is often desired for its own sake. Secondly, a true theory allows practical predictions and manipulations of the world for our benefit. It should be accurate not only about phenomena we have already observed, but also about any other phenomena described by the theory (which we have never observed). For the moment, let us suppose that truth is the enduring ultimate goal of science. As previously discussed, Laudan agrees that goals can constrain and regulate method. To this end, he identifies a naturalist justification of method (Laudan, 1987: 206-207): If actions of a particular sort, m, have consistently promoted certain cognitive ends, e, in the past, and rival actions, n, have failed to do so, then assume that future actions following the rule if your aim is e, you ought to do m are more likely to promote those ends than actions based on the rule if your aim is e, you ought to do n. This justification requires that we can tell when e is achieved. But Laudan argues that some goals are such that we cannot tell when they have been achieved. He describes these goals as transcendent, 11 and he argues that truth is one such goal (Laudan, 1984: 50-55). If e is transcendent, then we cannot know which actions can promote or achieve it. Therefore, a complete, true theory of the world cannot be the e in Laudan s rule, and cannot be the enduring, ultimate goal that motivates scientific methodology. One natural response to Laudan s objection is to replace truth with attractive lowerlevel goals that are not transcendent. For example, scientists may aim for theories that are simple relative to their explanatory breadth, and/or make successful novel predictions. These goals are more realistic in that (properly explicated) we can tell when they have been achieved. I shall call these applied goals. Applied goals can guide or constrain the methodological rules scientists follow according to the naturalist rule for justifying method. Replacing ultimate goals with applied goals is a move that won t appeal to everyone. Most philosophers and scientists would agree that goals such as simplicity, explanatory breadth, and successful novel predictions should be preferred to their respective opposites: complexity, explanatory narrowness, and unsuccessful or unsurprising predictions. However, many philosophers would be dissatisfied if there were no reason for this 11 Laudan actually uses the term transcendental. But Kant introduced a distinction between transcendental and transcendent which I will follow here. Goals are transcendent if they are beyond human knowledge, and this is what Laudan wishes to say about truth. 24
preference. For example, it would be nice if we could justify simplicity on the grounds that nature tends to be simple, and hence simpler theories are more likely to be true. Alternatively, it would be nice if we could justify simplicity on pragmatic grounds, by showing that simpler theories are easier to use. If either truth or utility is the ultimate goal, then this would provide a good justification, because truth and utility both seem obviously desirable. But otherwise, why should scientists prefer simplicity? It is not so obvious that simplicity is desirable as an end in itself. It seems somewhat arbitrary as a goal for science, and not one that is likely to unite scientists always and everywhere. Similar comments could be made about other applied goals. Without a very desirable ultimate goal, they seem inadequate to motivate a universal scientific method. We could extend this argument even to the goal of confirmation, or belief worthiness. If a scientist aims for truth, then she has good reason to make sure her claims are adequately justified or strongly supported by empirical evidence. If she does not hope to advance empirical knowledge, then she needn t be concerned with justification or support at all. In fact, she may prefer to make claims that are unjustified or contradicted by empirical evidence, if only to make them more interesting! Without an ultimate goal such as truth, it seems that scientists may have any applied goals whatsoever, and may change goals whenever they wish. Scientific method remains unfixed and appears to lack all epistemic credentials. The universal-goal proposal faces an impasse: the achievement of proposed scientific goals is either not adequately verifiable or not adequately attractive. However, there seem to be (at least) two moderate paths we can take to avoid this difficulty. Firstly, one could argue that empirical accuracy with respect to available data is the most we can ever know we have achieved. If this is true, then current empirical accuracy does seem to be the second-best ultimate goal. It does not offer a deep understanding of the world. But an accurate description of observed phenomena would be all the truth we can be certain we have obtained. Moreover, it should still allow us to successfully predict and manipulate the world. We can reasonably expect that a well-tested theory would be accurate (at least) about the kinds of phenomena we have already observed, even if we have less reason to feel confident about predicting kinds of phenomena we have never observed. Therefore, this kind of empirical accuracy seems appropriate as an ultimate goal. Presumably, empirical accuracy (relative to currently available empirical results) is such that we can recognise it when we see it. In this case, we may empirically test the correlation 25