Realism and the success of science argument Leplin: 1) Realism is the default position. 2) The arguments for anti-realism are indecisive. In particular, antirealism offers no serious rival to realism in explaining the success of science. 3) We have a compelling argument for realism (from novel prediction). Kukla/Walmsley: 1) Realism is the default position. 2) Anti-realism offers explanations for the success of science that are as good as realism. 3) We have a compelling argument for rejecting the success argument: explanatory power is only a pragmatic virtue, not an epistemic virtue. Our objectives: 1. What are the main arguments on each side? 2. Are they all terrible, or do some of them hold promise? 3. What to say in light of inconclusive arguments?
I. Leplin: A Theory s Predictive Success can Warrant Belief in the Unobservable Entities it Postulates 5.1 Burden of Argument Provisional acceptance of distinction between observable and unobservable entities. Unobservable entity = theoretical posit, inaccessible to experience Scientific realism: evidence can prove provide adequate reason (= justification) to believe that some theoretical (= unobservable) entities exist and have certain properties. Anti-realism: there can never be adequate reason (= justification) for belief that extends beyond the observable. At best, theories have explanatory and predictive utility. Miracle (success of science) argument: but for realism, the success of science is a cosmic accident. Naïve version of the miracle argument: to reject realism is to reject science. Problem: the meaning of success. An anti-realist can interpret success as empirical adequacy, and view the theoretical part of science as a purely formal system. Burden of proof: Even anti-realists need a philosophical view compatible with science. The anti-realist must explain the role of unobservables and why claims that these entities really exist are not justifiable. Strategy: Leplin will undermine two major anti-realist arguments and then offer what he sees as a compelling realist argument.
5.2 Underdetermination The underdetermination argument asserts: for any theory T supported by observation statements O 1,, O n, there is a rival T that is equally supported by O 1,, O n. Conclusion: we are not justified in asserting the theoretical portion of T. The falsity of T is logically compatible with all observational statements O 1,, O n that support T. This is logical underdetermination. Affirmation of T requires ampliative inference (induction). Let standard ampliative inference forms be those used in common-sense reasoning. The key question is whether theory T is underdetermined by evidence O 1,, O n, given accepted forms of ampliative inference. This is ampliative underdetermination. Although it s easy to construct logical rivals to T, it s very hard to construct ampliative rivals. Objection: There is no guarantee of serious (ampliative) rivals. Example: O 1,, O n & ~T. This is a logical rival: it s logically consistent with the evidence but not supported by it! In fact, it s not confirmable at all By hypothesis: O i s support T. They don t support ~T. They don t support O. So: they don t support O 1,, O n & ~T [Note: in Bayesian terms: Pr(O & ~T / O) > Pr(O & ~T).] Other gerrymandered rivals are equally silly. Conclusion: Logical underdetermination guarantees rivals, but none is supported by accepted inductive inference rules. Ampliative undetermination would work, but there is no easy guarantee of rivals. At best, a few cases of real ampliative underdetermination.
5.3 Superseded science Pessimistic induction: past successful theories have been wrong, so currently successful theories may be wrong. Counter-argument: current physical theories have been severely tested and so are unlike past theories. There is no inductive evidence for their failure. Key issue: second-order evidence (evidence about method). We distrust methods that led to belief in theories that have been falsified. We have second-order evidence against method, which leads us to reject first-order evidence. Main objections: Both methods and standards for evaluating methods have changed. Rejected theories may not have stood up to current standards Some past failures are partial successes. Use of past failures pre-supposes truth of current theory [dubious]. Anti-realist rejoinder: INCONSISTENCY is enough to reveal falsehood. Leplin: Inconsistency doesn t tell us exactly what is false. Inconsistency may be merely apparent.. Inconsistencies may pre-suppose truth of some theory. Broader point: The overall history of science is progressive: why not make that the basis of an optimistic induction?
5.4 Selective Confirmation First steps towards a positive argument for realism. Question: If we are realists, to which entities are we committed? Leplin rejects: 1) Anti-realism: commitment to no entities. 2) Holism: equal commitment to reality of all entities. Arguments for holism: - Rejection of confirmation for individual hypotheses/posits - Smallest unit of confirmation: the entire theory An entity is real if it participates in a successful theory Against holism: holism is a disaster for realism, since any change counts as rejection of the entire theory. The realist needs selective confirmation of some entities, while others can come and go. Starting point: The posited entity must be used essentially in a derivation/explanation The success of the theory must not be explicable without the posit Example: Maxwell and the ether. The ether is not essential to Maxwell s equations. Objection: Leplin s evaluation of what counts as essential is afterthe-fact, unhistorical.
5.5 Argument from Novelty Main claim: realism is needed to explain novel success of theories. Argument: Novel success = successful prediction by theory of a novel (unusual, unexpected) empirical result Novel = new, unknown, surprising, unanticipated independently of the theory, uninvolved in constructing the theory, and unlike results that support rival theories. Examples: Poisson bright spot; gravitational deflection of starlight Two requirements for novel prediction: 1. Independence. The result is not built into the theory T. The result is not just an instance of one of the laws of T. [Probably not quite right] 2. Uniqueness. No serious rival can yield the result. 1) Successful novel predictions by T require explanation. 2) That T logically implies the predictions is not an explanation. 3) The explanation must appeal to some property of T that accounts for its reliability. 4) The only good candidate is realism: T is (approximately) true. Realism can explain a record of successful novel prediction. No other explanation works.
5.6 Consequences and Clarifications 1. Independence. Historical: the result is not part of how the theory was developed. Note: if the theory has multiple sources, then all of its predictions are independent (since each will be independent with respect to some source). 2. Limitation of argument from novel success. A novel result might come to be explained by a rival theory, T. This undermines the argument for realism about T, if R would be novel for T even without T. But if R would not be novel for T, then the argument for realism about T stands. 3. A prediction. If realism about T is correct, we should not encounter a serious rival. Also, theories with successful novel predictions should continue to make such novel predictions. Objection: What about Newtonian physics? It had such a record, surely. 4. Realism is self-supporting. Realism makes a novel prediction (#3 above) and no rival would lead to that prediction.
5.7 Importance of Novelty Two natural anti-realist objections to Leplin s argument: 1) Novel success requires no explanation. 2) If it does require explanation: one can explain it just as well with empirical adequacy of T as with approximate truth of T. To head the first of these off, Leplin argues that anti-realism has no place for novelty. Novelty is possible only if predictive success must be an object of explanation. For an anti-realist, all observed consequences are on par. [WHY? Why can t an anti-realist have low conditional probabilities?] Question: But can t one explain novel predictive success by empirical adequacy?
5.8 Observation and Theory Observable/unobservable distinction: the previous sections show that we have justification for realism even if we allow this distinction. Main argument in previous sections: justification is not limited to the observable. Second argument: could there be justified observational beliefs, if no theoretical beliefs are justifiable? Observational beliefs can be undermined Beliefs about the reliability of observation depend on theory (e..g, theory of perception) So some background theory has to be justified even for observational beliefs to be justified Anti-realism is foundationalist : a privileged class of observational judgments, independent of theory. Dilemma for anti-realism: Either there is NO epistemically basic level, and then we are either realists or anti-realists who embrace complete scepticism; OR there is a basic level, with at least some theoretical judgments justified. So: the only alternatives to scepticism support realism. Question: Is this section confusing the observable/unobservable distinction with the theoretical/observational distinction? Does van Fraassen need the foundationalist picture to make sense of constructive empiricism? [NO]
II. Kukla and Walmsley, Predictive Success does not warrant belief in unobservable entities Strong realism = our current theories are true Minimal realism = it s logically/nomologically possible to attain a state that warrants belief in a theory. Success of science = our theories let us make many more correct predictions than chance. Miracle argument: (strong) realism is true because it s the only explanation that does not make the success of science a miracle. Weaker version: realism is the best exmplanation for the success of science. Main Claim: success of science won t warrant even minimal realism. Ground rules for debate ( 6.2): 1. Admit that science is successful. 2. Admit that realists do have one possible explanation for the success of science. - The miracle argument is not circular. - Approximate truth is meaningful. - Approximate truth does explain success of science Strategies: 1) Show that there is a comparably good anti-realist explanation. 2) Deny that T is explanatory provides support for T is true.
6.2 Anti-realist explanations for success of science Type 1: Evolutionary epistemology. Our theories are successful because they have been selected for predictive success. Rejected by Kukla/Walmsley: This explains why we possess successful theories but not why an individual theory is successful Type 2: Surrealism. Our theories are successful because they are empirically adequate. Kukla/Walmsley basically think this as good an explanation of success as realism. But there are different versions, depending on how we interpret empirical adequacy. Definitions and notation. T* = empirical consequences of T (= consequences about observables) T and T are empirically equivalent if T* = T * (i.e., same empirical consequences) Three versions of surrealism. [Progressively weaker] Strong surrealism: T is successful because T is empirically equivalent to a true theory, T. Weak surrealism: T is successful because the world behaves as if T were true. [All of T s empirical consequences, T*, are true, even those we have not yet derived.] Fragmentalism: Our current set of empirical beliefs X coincides with the known empirical consequences of T (i.e., the known part of T*). [T provides a compact representation of X]. Kukla/Walmsley claim: Leplin successfully demolishes the first and third version, but the second is viable.
Leplin s critique of surrealism (NOT in our article) A. Surrealism presupposes realism. If T s empirical adequacy can explain T s success, then T s truth can explain its success. So realism about T explains T s success. Comment: This proves only that realism is a potential explanation, not that it is the only or the best explanation. B. Surrealism collapses to realism unless we assume EE. And EE is dubious. EE = every theory has an empirically equivalent rival. Suppose T has no EE rival (and hence is empirically equivalent only to T itself). Suppose we accept strong surrealism: T s empirical adequacy explains its success. Then T is empirically equivalent to a true theory; but by assumption, that can only be T. Hence T is a true theory and we have realism. Comment: This defeats only strong surrealism. The argument fails for the other two versions, which don t imply the existence of any true theory. C. Weaker versions of surrealism provide no explanation. Empirical adequacy of T merely restates the explanandum. Comment: This defeats only fragmentalism. The explanans and explanandum both amount to from T we can derive a lot of empirical phenomena. Kukla/Walmsley claim: weak surrealism is not a restatement of the explanandum. So this antirealist explanation for success survives; realism is not the only explanation. Further: there is no case that realism is a better explanation of T s success. What feature p, relevant to beliefworthiness, does theoretical truth possess more of than empirical adequacy?
6.3 Two Realist Arguments from Success of Science Argument 1 (SS). [Putnam s argument] Science is more successful than can be accounted for by chance. Realism is the only (the best) explanation. Objection: Weak surrealism is an alternative explanation, and there s no reason to think it s not as good an explanation as realism. Modification: substitute approximate truth into the definition of realism. Kukla/Walmsley: 1. Approximate truth is not an explanation for success of science. 2. If approximate truth is an explanation, realism does not follow. Argument 2 (SS ). Replace actual success of science with possible success to give a stronger argument for realism. Some theories would be successful if used to make predictions; the best explanation would be their (approximate) truth; hence, there are circumstances in which we d be justified to be theoretical realists. Question: What is the point of SS? Kukla and Walmsley want to show that realism could never be supported by evidence. They need to address the realist objection that there have been no successful theories, but that successful theories are still possible.
6.4 Truth and Truthlikeness as Explanations Claim 1: truth and truthlikeness can t explain the success of science. [Question: Doesn t this go against the initial concession that realism can provide an explanation?] Argument: Some past successful theories are known not to be true. So there must either be no explanation for success, or some other explanation. Realist response: past theories are approximately true. Comment. If approximate truth is too liberal, anything counts as approximately true and then approximate truth is no warrant for realism. If approximate truth is too restrictive, then no failed scientific theory has ever been successful. The argument SS is still available. But the assumption that no failed theory has been a success is too high a price to pay. So: suppose there have been successful theories, some false (?). That refutes Putnam s claim something other than realism must provide the explanation. Question: Why not dodge this worry by arguing only that, for successful past or current theories not known to be false, truth is the best explanation? The anti-realist and weak surrealism have the same problems with approximate empirical adequacy. So why not restrict to those not known to be false? [Compare: explanation of optical illusion vs. explanation of veridical seeing.]
Claim 2: Pessimistic meta-induction. An argument that our current theories are false. Comment. A weak argument. It s hard to suppose our current theories are unsuccessful. One can argue that false past theories were unsuccessful. That blocks the induction to successful current theories. Claim 3: Some successful theories are not even approximately true. Blocks inference that approximate truth explains success. Comment. It s hard to secure the claim that there are successful theories that fail to be approximately true. Claim 4: No reason to believe that approximate truth entails success. From approximate truth, it does not follow that the theory will be successful. There is no clear account of approximate truth. So approximate truth can t give us a decisive argument for realism.
6.5 Circularity of Argument from Success of Science Main claim. Even if truth is best explanation of success, realism does not follow. This is the best anti-realist argument. Short version. SS (see above) is circular. Long version. Suppose T is successful and the truth of T is the only (or best) explanation for this success. We don t get realism about T unless we assume that explanatory power is a reason for believing in truth of T. Van Fraassen: epistemic vs. pragmatic virtues. Explanatory virtues count as pragmatic and don t go towards accepting as true. [It s enough that this view is possibly right: realists can t assume explanatory virtues contribute to truth.] Objection (Boyd): Scientist do use IBE, and count explanatory virtues towards truth. Response: No argument justifies this view. No reason to think anti-realists must be committed to applying IBE in philosophical arguments.