Very Incomplete Notes for _Blackwell's Companion to the Philosophy of Science_ October-December 2004 ***** Explanation The most well-known model of explanation is Hempel's deductive-nomological model (aka, covering law): An explanation of an event (or a law) is a valid deductive argument whose premises are the relevant circumstances and general laws and whose conclusion is the event (or law) to be explained. This model is for ideal explanations; ordinary explanations are explanatory sketches which could be more fully elaborated. The D-N model has too many objections to be taken very seriously. One problem is that the D-N model (without adding further criteria) fails to capture the asymmetric nature of explanation; for example, the height of a flagpole explains the height of its shadow, but not the other way around. Another difficulty involves constructing valid which we do not accept as explanatory, typically because they include irrelevant premises; for example, the stick appears bent because it is submerged in holy water that has been blessed. Laws are often used in explanations, but perhaps (as argued by Scriven) they are not always required. Instead, explanation relies on causation (causal-relevance model of explanation). Unfortunately, the notion of causation is at least as philosophically problematic as that of the D-N model. It seems that at least some explanations can also be understood as models or analogies or as unifying accounts. Pragmatic considerations may also play a role in explanation, or explanation may even be purely pragmatic. Van Fraassen holds that explanations (explanatory answers) are context-relevant. Perhaps there are different kinds of explanation, and there is no single account to capture them all. The lack of any single accepted account of explanation is something of a "scandal" to philosophy of science. (Note: The rest of the section on explanation is taken from articles in Routledge Contemporary Readings, but it seemed best to put it here.) The standard account of explanation is Hempel's (and Oppenheimer's) deductive-nomological (covering law) model. Explanation is a deductive argument with premises of circumstances and laws and a conclusion of the explanandum-event. This account meets many problems, including questions of necessity and sufficiency, relevance, and the challenge of distinguishing laws from other generalizations. Van Fraassen and Achinstein focus on the pragmatic aspects of explanation. Van Fraassen sees explanation as context-dependent (relative); the question "Why X?" is really "Why X as opposed to Y?" Explanation for Friedman and Kitcher involves unification. Friedman's thesis is that scientific understanding is increased with the reduction of the number of assumptions (axioms) needed to explain natural phenomena (regularities). For Kitcher, unification further includes seeing commonalities and connections in what initially appeared to be different situations; explanation involves minimizing the number of patterns of derivations while maximizing the number of conclusions generated. Another approach is to see explanation as rooted in causal structures. Explanation involves identifying underlying causal mechanisms (Scriven, Salmon). Appeal to causality solves such problems as the asymmetry (flagpole/shadow) problem that plagued Hempel's account, but causality is itself problematic (as pointed out by Hume). Salmon offers an account of causal interaction in which a process is causal if it is capable of trasmitting a mark (interacting and making a change). ***** History, Role in the Philosophy of Science During the first half of the 20th century, philosophy of science and history of science were completely separate. Philosophy of science was motivated by logic, rigor, and positivism, not history. Starting with Kuhn (Structure, 1962) history became important. Kuhn as a historian illustrated his philosophical points with instances of history. To respond, critics had to look to the history of science. Lakatos' big thing seems to be the historical research program. All the philosophers write their normative histories of science and a comparison of their histories with the actual history provides a means for evaluating those philosophers' theories. Both philosophy of and history of science have something to learn from each other. Are history and philosophy of science two aspects of a single subject? Kuhn (Essential Tension, 1977) says no; for support he refers to his experiences teaching seminars to students of history and philosophy. Among other differences, philosophers focus on theory while historians focus on individuals. Historians may say philosophers write bad history, historical narratives written with a particular agenda; this should be recognized as a distinct genre. What are the roles of history of science in philosophy? History is a source of philosophical problems. History can be used as a standard against which to test philosophical theories. History provides a richer way to understand theories; understanding science isn't about understanding a particular moment in science, but rather understanding how theories are developed and hold up over time. Historical context limits the range of philosophical questions and answers possible at any given time. History of science provides materials for philosophical narratives. Drawing on the history of science is a way to ensure that philosophy of science is descriptive rather than just normative. History creates a critical distance from which to study science. Aside: One possible answer to the demarcation question is that progress is an important part of science; if something is science, it must make sense to ask whether it is progressing. By this criteria, astrology and folk psychology are not sciences. ***** Incommensurability Incommensurability (literally "no common measure") is associated with both Kuhn and Feyerabend (1962). Evaluative incommensurability (Kuhn): Scientific theories/paradigms contain within themselves their own standards for success and so a general assessment of theories outside any theory is impossible. The claim of linguistic incommensurability of scientific theories held by both Kuhn and Feyerabend is that rival scientific theories cannot be phrased in a common set of linguistic terms. And yet both seem to think *some* comparison and translation is possible. Linguistic incommensurability (Kuhn): Kuhn's ideas about incommensurability seem to shift dramatically over time. 1970: A linguistic mode of comparison between theories is impossible. However, some translation is possible by using shared vocabularies to translate. Or, there is no language into which many theories can be translated without loss. 1980: "incommensurability thus equals untranslatability" The problems of translating a scientific text into a foreign language are similar to the problems of translating literature. In both cases translators are faced with sentences which can be rendered in several ways, none of which capture them completely. Examples of translational differences (aside): (1) clusters of interdefined terms (phlogiston). (2) conceptual disparity (how to translate a word which has multiple meanings). Linguistic incommensurability (Kuhn), cont: It is not possible to phrase all the claims of two scientific theories in a single language so they can be put side by side and their exact points of difference pinpointed. Theory choice, then, cannot be based on point-by-point comparison. Scientists who learn a new theory do not translate, but learn a new language from scratch to become a native speaker of the new language. Linguistic incommensurability (Feyerabend): When a transition is made from a restricted theory (eg, impetus theory) to a wider theory (eg, Newton's), there is a replacement of ontology. Contextual theory of meaning: All strong alternative theories (those which differ everywhere) are incommensurable. Still, large parts of our total theory remain constant across some scientific theory changes. Differences between Kuhn and Feyerabend: (1) Feyerabend's variety of incommensurability is more global; fundamental changes of theory lead to changes in the meanings of *all* of the terms in a particular theory. (2) Kuhn's incommensurability stems from specific translational difficulties while Feyerabend's results from extreme holism about the nature of meaning. Agreement of Kuhn and Feyerabend: Incommensurability need not mean incomparability. Both have alternative means of comparison. Both use the analogy of learning new theories and learning new languages as native tongues (rather than through translation). Responses to incommensurability: (1) At least one component of meaning is unaffected by untranslatability, namely, reference (Scheffler, Putnam, Davidson). (2) Non-linguistic comparisons can be made (2a. model-theoretic, Sneed, Stegmuller; 2b. cognitive, Churchland, Giere, Thagard). ***** Inference to the Best Explanation Scientists judge whether observations support, disconfirm, or are irrelevant to given hypotheses. Often these judgments are made on non-deductive grounds, such as the inference to the best explanation (similar to Peirce's abduction). The inference: Scientists infer from the available evidence to the hypothesis which would, if correct, best explain the evidence. This may seem backwards or circular, but it's not. Explanatory virtues: What makes one explanation better than another? Is it likeliest? Loveliest? Plausible candidates for explanatory virtues: scope, precision, mechanism, unification, and simplicity. Humean skepticism: Is there any good reason to believe that our inductive practices are reliable? Putnam's miracle argument for scientific realism: The best explanation for a theory's predictive success is that the theory itself is true, for it would be something of a miracle for the theory to be instrumentally successful but not actually represent the world. (Objections include circularity, that truth is not the best explanation of predictive success, and the underdetermination of theory by data). ***** Judgment, Role in Science Evaluating a hypothesis is not merely a matter of observation and (deductive and inductive) logic; it requires judgment of scientists, but this does not mean we should draw skeptical conclusions about science. Judgment is a cognitive skill; it is a skill but not one which is performed by following an algorithm. Scientists learn to exercise judgment in their specialties as they master their fields. (Note: In virtue terms, this sounds something like: good science results from the practices of virtuous scientists, where virtuous scientists have cultivated good judgment.) Scientists use judgment to determine which hypotheses are worthy of further consideration, what tests to carry out to further consider them, how to interpret the results of the test, and what to do with that interpretation. No body of observations can prove a universal hypothesis true. Many conclusions (green and grue) may be compatible with the observations. Judgment is used to opt for one of several possible hypotheses. Falsification, as well as confirmation, requires judgment. Because no single hypothesis can be tested in isolation, falsification shows that at least one hypothesis is false, but does not point to which one. For this, judgment is used. Judgment is also used to determine which observations are acceptable. (Eg, are observations made with telescopes reliable?) ***** Laws of Nature Laws of nature express regularities and are taken to hold universally and be in some way necessary. Three views of laws of nature: 1. the essences view (Aristotle), 2. the summary of sensory experience view (Hume, Mach), and 3. the tendencies/powers view (Locke, Cartwright). ***** Observation and Theory One major set of questions in the philosophy of science concerns unobservable entities (electrons, forces, etc). How should theories about unobservable entities be construed? Do such entities exist? How do we know if they are unobservable? Etc. The standard realist view is that unobservable entities exist. They exist independently of us and how we describe them. Scientific theories which describe unobservables are to be taken literally. Claims about unobservables have truth values; they are either true or false depending on the relationship between the claim and the entities themselves. There are numerous anti-realist accounts, the most well-known being instrumentalism. Unobservable entities do not have ontological existence. They are useful/convenient/instrumental names. Theories do purport to describe unobservables, but in doing so they are not making assertions which have truth values. An alternate view is that unobservable entities do not exist, but that theories involving unobservables can have truth values (on some other grounds besides the relationship between claims and independently existing entities). A third important view is van Fraassen's constructive empiricism. Van Fraassen considers the aims of science. For the realist, the aim of science is to provide a literally true story of what the world is like; to accept a theory is to believe that it is true. For the constructive empiricist, the aim of science is not truth but empirical adequacy. Unobservables entities may exist, but they are of no concern to the constructive empiricist and his scientific aims. There are a handful of common arguments for and against realism. Arguments for realism include: (1) appeal to common sense (scientists do speak of their theories as true/false, their entities as existing); (2) Putnam's miracle argument (the best explanation for a theory "saving the phenomena" is its truth, otherwise its agreement would be a miracle); (3) the principle of common cause (a common unobservable cause is behind two correlated observable facts/events). Arguments for anti-realism arguments: (1) appeal to empiricism (realism is unsatisfying in postulating a mysterious unobservable realm of unobservable entities, a world of metaphysics not physics); (1) ontological simplicity (having fewer entities in one's ontology is simpler/preferable, so nominalism about unobservables is better); (3) scientific aims/practice (scientists aim not at truth about an independent realm of unobservables, but about the observable consequences of theory). ***** Pragmatic Factors in Theory Acceptance What does it mean to accept a scientific theory? For a scientific realist, it means to believe it to be (approximately and/or essentially) true. But this doesn't seem to be entirely right. Theory acceptance does not always involve (ontological) truth commitments. Theories often are accepted by those who do not believe that the theories are literally true (nor that the entities that they postulate actually exist). Acceptance need not always involve belief in truth; accepting a theory can involve working with it, using it as the basis for improving it, developing other theories, or even to falsify it. It is sometimes even useful to accept theories which are known to be false; Newtonian, rather than relativistic, mechanics may be used by engineers because the math is easier. Is theory choice uniquely determined by empirical evidence alone? No. Theory choice/acceptance involves comparing available observational evidence to theories and their consequences. According to the underdetermination thesis (underdetermination of theory by observation), no amount of evidence can deductively establish one theory over another; for any given theory, there are always indefinitely many different theories which have equal empirical adequacy. Are theories which are empirically equivalent then fully equivalent? Looking at the history of science, it seems not; scientists do prefer some theories over others. They do so by appealing to criteria beyond having the right observation consequences. Scientists often appeal to such factors as simplicity (freedom from ad hoc assumptions and corrections), unity, and coherence. Are these factors pragmatic (factors which reflect features we happen to like or find useful in theories) or do they actual point to truth in a realist sense? Van Fraassen claims these factors are pragmatic and that they play a role in theory acceptance. But theory acceptance does not imply belief in truth. The only belief involved in theory acceptance is belief in the theory's empirical adequacy, belief that a theory saves (all and only) the phenomena. There is, for Van Fraassen, no reason to think that simplicity, unity, etc. increase the likelihood of a theory to be true. What happens when scientists face rival theories contending for acceptance? Often the older theory accounts "naturally" for a certain observed phenomena. Post hoc corrections (think Ptolemaic epicycles) are made to the older theory to account for the phenomena. Scientists seem to make the judgment that even though the old theory can accommodate the phenomena, the newer theory is to be preferred for epistemic reasons: the theory which explains the phenomena more "naturally" is more likely to be true. Similarly, when comparing two theories, one of which has broader explanatory scope (without resorting to post hoc corrections), we have better reasons to believe in the truth of the theory which unifies a greater number of phenomena (and/or theories). Are explanatory virtues pragmatic or epistemic? Van Fraassen claims that we have evidence for theory truth only through evidential support for its empirical adequacy. Therefore, we can never have better evidence for the truth of a theory than we have for a related theory which is equivalent except for requiring no truth commitments. Belief involves empirical adequacy only; belief in truth is supererogatory; it is unnecessary metaphysics. Against Van Fraassen, some realists claim that while at any given (frozen) moment in science, two theories may be equally empirically adequate, that is no guarantee that they will be equally empirically adequate in the future. Theories do not just explain known phenomena, but also make predictions about new types of phenomena. To achieve this kind of empirical success requires a pursuit for true theories, not just empirical adequacy. Boyd among others claim, then not just that scientists have seemed to treat explanatory virtues as epistemic, but further that the progress of science requires some (minimal) metaphysics. ***** Realism and Instrumentalism This article is actually about the status of abduction (inference to the best explanation), at least as much as it is about realism and instrumentalism. Roughly, the abductive form is: (1) The surprising fact, C, is observed. (2) If A were true, then C would not be surprising. (3) Therefore, C gives us reason to suspect that A is true. If abductive inference is accepted, an argument for realism follows: The explanatory success of a science theory warrants belief in the truth of the theory. At the very least, realism maintains that explanatory success could warrant belief in theory truth. Quine's argument against the abductive argument for realism: For any explanatory theory, other alternative empirically equivalent theories can be constructed. This leads to underdetermination -- no body of evidence supports a theory to the exclusion of all rivals. Explanatory success of a theory, then, cannot point to truth since multiple incompatible theories can be successful. Against Quine, Laudan and Leplin argue that two different theories are only empirically equivalent at a particular point in time; future observations can determine between different theories. Van Fraassen's argument against the abductive argument for realism: A theory may be explanatory in one context but not in another. Theories are explanatory only relative to an explanatory purpose; theories are not objectively explanatory. Against van Fraassen, Worrall argues that ordinary explanations are usually given in a particular context so they are phrased in an incomplete, context-sensitive way; a complete causal account can be specified for a theory, and such an account is explanatory in a non-relative sense. Laudan's argument against the abductive argument for realism: Theories which in the past were considered to be well-supported have since failed. Present theories are no different; despite their current success, they are likely to be false. History shows us that explanatory success/virtue cannot justify belief in theory truth. Against Laudan, Leplin argues that while not all explanatory success requires realism, the novel predictive success of theory does. Those theories which really are successful are those which make novel predictions which turn out to be true; such theories must themselves be true to have this predictive success. Fine's argument against the abductive argument for realism: Use of abductive inference in arguing for realism is a vicious circularity. Even if Fine is correct, it follows that an abductive defense of realism cannot be presupposed to be legitimate, but it does not follow that abductive inference is illegitimate. ***** Simplicity It is generally recognized that observation/data alone cannot always single out the best hypothesis. Multiple hypotheses may equally fit the data. Yes in such times, scientists do choose one hypothesis over others, and they often do so by appealing to simplicity. The classic example is that the Copernican world system was chosen over the Ptolemaic because it is conceptually simpler. Simplicity (parsimony, Occam's razor, etc) can be seen in claims of a preference for hypotheses which have fewer assumptions or postulate fewer entities or the preference for (mathematically) simpler models which fit the data. A useful analogy is that of curve fitting. Multiple curves (equations) can fit a set of points. When the sum of squares method cannot decide between two curves, scientists often believe that the simpler curve better fits the data. This raises several questions: 1. How should the simplicity of a curve be measured? 2. How can use of simplicity be justified (as getting at truth rather than as being a pragmatic consideration)? 3. How should simplicity and goodness of fit be traded off? Much of the article considers the first question. Jefferys proposes that curves which are less mathematically complex (lower degree polynomial, etc) are simpler and are assigned higher Bayesian prior probabilities. For Popper, curves which are more falsifiable (need fewer additional observations to be falsified) are preferable. Akaike considers predictive accuracy and parameterization; curves which have fewer adjustable parameters are simpler/preferable. ***** Supervenience and Determinism The goal of unity of science is to fit all scientific theories into a hierarchy of theories reducible to other theories ... reducible to physics. Is there an alternative to reductionism which can support unity of science? One answer is supervenience. A domain x supervenes on a domain y if and only if any change in x requires a change in y. How does supervenience differ from reduction? Reduction is a relation between theories, whereas supervenience is a relation of correspondence between domains/ontologies. (Reduction requires supervenience, but not vice versa). Supervenience, unlike reductionism, can be held even when the links between different domains are not understand. Supervenience allows for emergence (much as Conway's life allows for complexity at higher levels). A scientific discipline which supervenes on another can maintain autonomy. ***** Teleological Explanation Teleological explanations are those that answer "for what purpose?" Some philosophers think such explanations have a place in science, others don't. Teleological explanations include explanations of: 1. goal-directed behavior, 2. functional tools, 3. organs of living organisms, and 4. social functions within societies. Teleological explanations appear frequently in biology, as in "the organ x is present because it has/had function f." ***** Underdetermination of Theory by Data There are many different theses that go by the name of underdetermination. Weak underdetermination of theories is the claim that two theories can both fit equally well to the observational data available at a particular point in time. This is relatively uncontroversial, and the status of the Ptolemaic, Tychonic, and Copernican world views as of 1600 is a standard example. Even when observational data cannot determine between two theories, there can still be pragmatic reasons to prefer/accept one theory over the over (as arguably can be seen with the Tychonic and Copernican theories in 1633). The thesis of strong underdetermination is far more controversial: Perhaps theories are underdetermined not just by the observations available at a given point in time, but by all possible observations. Perhaps, even were all possible observational data to be available, every theory would have at least one alternative (inconsistent) theory which is observationally (predictively) equivalent. Realists, of course, must deny strong underdetermination, or else we have no grounds for thinking that we are adopting truer and truer theories. Instrumentalists often appeal to strong underdetermination to argue against realism. Strong underdetermination has had a number of proponents, including Duhem, van Fraassen and Quine (for a time). Duhem's (also Quine's) underdetermination thesis is based on holism. When experimental observation does not conform with the prediction given by a theory, there is no way to determine which of the experiment's assumptions should be rejected. Adjustments could be made to the theory and auxiliary assumptions in any number of places. There is no way to test assumptions individually, but only as a tribunal whole. Problems with strong underdetermination: (1) Strong underdetermination presupposes a dichotomy between the observable and the theoretical which realists find unacceptable. (2) Apparent cases of underdetermination may be cases of equivocation, not genuine incompatibilities. (3) There is no threat of underdeterminism when one takes a less narrow view of empirical evidence and allows pragmatic factors (e.g., simplicity) to play a role in theory choice which is consistent with realism. (4) There are no actual cases in history of strong underdeterminism, so there is no reason to think that it is true. (5) Strong underdetermination relies on an incorrect analogy with curve-fitting; all points would uniquely determine a curve. Conclusion: Weak underdetermination is likely true, and unproblematically so. The thesis of strong underdetermination raises an interesting question for consideration, but there is no reason to believe that the thesis is true, rather than just a speculative conjecture. *****