No Miracles Argument

Introduction: The Success That Needs Explaining

The no miracles argument begins from an undeniable intellectual achievement: scientific inquiry has produced theories and instruments capable of predicting phenomena far beyond the circumstances in which they were devised. A theory can organize measurements, anticipate unobserved effects, and guide reliable intervention in processes whose underlying causes cannot be inspected with unaided perception. Scientific realism interprets such success as evidence that at least some central features of successful scientific descriptions answer to an independently existing world. The question concerns the extent of this epistemic entitlement. The scientific realism overview identifies the familiar argument alongside its major historical and probabilistic criticisms.

A prediction and an interpretation of that prediction are different intellectual acts. The experimental result belongs to the evidential history of a discipline. The inference that the theory responsible for it is approximately true is a further philosophical judgment. Realists insist that this judgment supplies the most satisfying explanation of sustained success. Their opponents argue that explanatory satisfaction cannot by itself secure belief in the unobservable entities described by the successful theory. Both sides accept much of science's achievement, so the dispute is about the commitments that achievement warrants.

The present inquiry assesses the argument as a family of reconstructions rather than as one unchanging syllogism. Historical texts, bibliographic records, publisher abstracts, and accessible primary articles are kept distinct from my analytical examples. The aim is to determine what would have to be established for scientific success to provide strong support for approximate truth, and what evidence permits that inference to be made in a specific case.

The Historical Formulation in Putnam

Hilary Putnam's formulation in his 1975 collection presents realism as the position that renders scientific success intelligible. His discussion in Mathematics, Matter and Method, particularly the familiar sentence on page 73, became the standard reference point for the argument. The formulation is rhetorically concentrated. It invites the reader to compare the apparent coincidence involved in repeated successful predictions with the explanatory economy of a theory that accurately describes what makes those predictions come out right.

This historical point requires a bibliographic qualification. Putnam's claim is not a formal probability theorem, and later reconstructions should be presented as developments of the argument rather than as his own mathematical demonstration. The broad maxim could support different versions of realism: confidence in theoretical reference, confidence in approximately true relationships, or a weaker commitment that science identifies stable causal structures. These versions have different implications for how theory replacement counts as evidence against them.

The language of miracles is therefore a comparative judgment about explanatory costs. If precise prediction were systematically generated by theories whose central descriptions were disconnected from the processes producing observations, the fit could appear astonishing. Yet astonishment is a starting point for explanation, not evidence that all alternatives have already been eliminated. A careful reconstruction must identify both the success being explained and the rival accounts against which the realist explanation is being assessed.

Boyd and the Methodological Development of Realism

Richard Boyd expanded the discussion beyond a slogan by treating scientific methods themselves as part of what realism must explain. In The Current Status of Scientific Realism, Boyd examines the strengths of realist arguments and the limitations of conventional empiricist replies. The motivation is methodological: theories help scientists design instruments, decide what to measure, and correct previously accepted beliefs. Successful methods appear to rely on a background account of what the world is like, including claims about things beyond ordinary observation.

This argument changes the unit of analysis. Instead of asking only why a single equation predicts a result, it asks how a historically developing practice manages to improve its measurements, discriminate errors, and produce new knowledge. The realist reply is that a theory's partial contact with the world makes possible the reliability of these theory-guided procedures. That reply has substantial explanatory reach, but it also increases its obligations. It must show how the procedures discriminate truthlike claims from inherited approximations, methodological conventions, and instrument-dependent regularities.

The relationship between successful inquiry and realism cannot be established merely by assuming that the best scientific methods reliably lead toward truth. That assumption is closely related to what the argument is intended to vindicate. Boyd's contribution is most fruitful when it encourages investigation of identifiable error-correcting practices. One can then ask which specific features of a method would fail if its underlying theoretical interpretation were radically mistaken.

Three Commitments Commonly Called Scientific Realism

Scientific realism combines several claims that should be assessed separately. One is metaphysical: the world investigated by science exists independently of our beliefs about it. A second is semantic: scientific assertions about unobservables are capable of truth and falsity. The third is epistemic: at least some successful assertions about those unobservables deserve belief. The Stanford account of realist commitments carefully distinguishes these dimensions and surveys ways of limiting their scope.

The no miracles argument chiefly attempts to establish the third commitment. An antirealist may accept a mind-independent world and interpret talk of electrons literally while declining to believe that a given theory has correctly characterized electrons. It follows that describing antirealism simply as denial of reality, or as a belief that scientific claims lack meaning, obscures the principal controversy. Empirical adequacy and realist truth can be rival standards of acceptance within a shared commitment to rigorous experimental investigation.

The analytical issue is therefore narrower and more demanding than the popular contrast between believing and disbelieving science. A working scientist may rationally rely on a theoretical model for a particular task while withholding commitment to its full underlying interpretation. Equally, a realist may believe certain experimentally supported components of a theory while expecting its general representation to be revised. The no miracles argument needs to explain which of these positions is supported by successful performance and why.

Reconstruction as Inference to the Best Explanation

The classic reconstruction has three steps. First, some mature scientific theories display significant and independently tested success. Second, their approximate truth would explain this success better than available alternatives. Third, we have grounds to accept, provisionally, the approximate truth of at least the theory components responsible for the success. This is a form of inference to the best explanation rather than deductive entailment. Gilbert Harman's 1965 account of explanatory inference provides an early systematic locus for the inferential pattern, while Peter Lipton's Inference to the Best Explanation elaborates its evaluative structure.

The second step carries most of the philosophical weight. An explanation must be compared with alternatives under standards that are themselves defensible. Simplicity, coherence, causal articulation, and the prediction of unexpected results can make one explanation attractive, but attractiveness is not equivalent to demonstrated truth. If a theory was deliberately tailored to observations, its ability to accommodate those observations requires little explanation. The stronger case concerns results whose discovery or measurement was substantially independent of the theory's initial construction.

An inference that stops at the statement that successful theories are often true is also insufficient. The actual conclusion can concern all components of a theory, its central causal claims, its mathematical relations, or merely its predictions in a defined domain. Each conclusion demands a different evidential argument. Inference to the best explanation supplies a useful structure for reasoning, but the success of the inference depends on the content and comparison of the particular explanations.

What Scientific Success Means

Scientific success is not a single measurable quantity. Successful accommodation of already known observations, prediction of previously unknown phenomena, convergence of independent measurements, stable control of an apparatus, and export of results to new contexts each establish different achievements. They also have different vulnerabilities to flexible fitting, background assumptions, and unnoticed selection. John Worrall emphasizes the relevant difference in Structural Realism, where he excludes results inserted into a theory during its construction from the strongest intuitive version of the no miracles consideration.

A model can be useful through a deliberately idealized representation. It may yield a correct value within a limited range while misdescribing conditions outside that range. Another model can have excellent predictive accuracy without offering a credible causal story of why its outputs occur. Conversely, a causal claim may be secure even when a global theoretical framework is incomplete. The degree and kind of performance matter because realist confidence has to attach to what the achievement actually tests.

For this monograph, strong empirical success means independently assessed performance that exceeds what would be expected from direct fitting or routine adjustment, has survived relevant criticism, and can be connected to a specified inferential task. This operational description organizes the inquiry; it is not a newly proposed doctrine. It leaves open whether any such success justifies claims about truth, while preventing the phrase 'scientific success' from silently changing its meaning across examples.

Novel Prediction and Evidential Independence

Novel prediction appears particularly favorable to realism because a theory that anticipates an independently discovered result seems to reveal more than the data that guided its initial construction. Jarrett Leplin's Novel Defense of Scientific Realism develops a careful version of this intuition. Novelty involves both independence from the result's role in the theory's origin and an assessment of whether a competing explanation could yield the same outcome. A chronological surprise is not enough if the supposedly new result already constrained the theory's formulation.

The philosophical value of this approach is its attention to evidential independence. If a researcher uses an observation to choose free parameters and later cites that same observation as a major confirmation, the evidence has effectively been counted twice. A genuinely independent test limits this circular reinforcement. Yet independent prediction remains compatible with multiple theoretical explanations, particularly where they share mathematical structures or rely on common empirical constraints. Novelty can make evidence more discriminating without automatically establishing a single ontology.

The strongest evidence comes from identifying why the prediction succeeded. Which assumptions were indispensable? Which were convenient idealizations? Could the same result follow from several models? These questions shift the discussion from the prestige of a theory to the provenance of a particular accomplishment. They also make the no miracles argument more capable of learning from histories in which a successful prediction survives the abandonment of its original theoretical rationale.

Accommodation, Fitting, and Retrodiction

Accommodation is often rational and scientifically indispensable. A model may be built to reproduce known observations because those observations define the relevant problem. Retrodiction can also be highly informative when independent records make the predicted past event testable. The mistake is to treat every successful fit as equally strong evidence for the full theoretical interpretation. Worrall's analysis of predictive support clarifies why a result engineered into a theory has a different epistemic role from a result whose fit remains independent of its construction.

Consider an invented model containing a free constant calibrated to match the measured output of one apparatus. Its exact agreement at that calibration point is expected. Evidence improves when the same calibration predicts accurate outputs of different apparatuses or new ranges without further adjustment. The growth of explanatory entitlement depends on the breadth and independence of successful tests. The example is an analytical illustration, not an historical experiment or an original empirical finding.

This distinction matters for evaluating scientific prestige. A long record of agreement may incorporate repeated parameter refinement, changing definitions of acceptable error, and selection of favorable datasets. Such activities can be legitimate science. Their existence means that the philosopher must reconstruct the history of evidential dependence before declaring the success inexplicable on nonrealist assumptions. A transparent history of correction is itself an achievement, but it has to be interpreted with appropriate attention to the practices that generated it.

The Logic of Approximate Truth

The phrase 'approximately true' protects realism from the unreasonable expectation that a successful theory's every sentence is exactly correct. It also introduces a serious demand for precision. A theory may give accurate quantitative relations in a bounded domain while its causal ontology, geometry, or underlying mechanism is mistaken. Such cases invite a distinction between numerical approximation, structural preservation, successful reference, and broader truthlikeness. The discussion of approximate truth identifies these as contested issues in assessing scientific realism.

An intelligible claim of approximation should name a target: an observable quantity, a range of operating conditions, a relation between variables, or a mechanism's identified components. Without such a target, any success could be redescribed as evidence of 'some truth' after its failures are known. That maneuver risks turning realism into a retrospective label for whatever was retained. The inference must specify beforehand, or at least by publicly defensible reconstruction, which elements did the explanatory work.

Approximation is also relative to purposes. An ideal-gas equation can be informative for some conditions without assigning literal reality to every assumption in its idealized construction. The realist's task is to identify the stable relationship that supports reliable calculation, then determine whether the explanatory role of that relationship warrants a claim about the world. This contextual reasoning is more demanding than assuming that good predictions certify the entire language of their producing theory.

Reference and the Unobservable

Scientific theories routinely employ terms for things that cannot be directly perceived. Whether a successful theory supports belief that such terms refer depends partly on experimental access, causal integration, and continuity through theoretical change. The issue is central to the realist and antirealist comparison, which distinguishes literal interpretation from epistemic endorsement and describes forms of realism that concentrate on entities or structures.

A theoretical term can contribute to successful calculations under more than one interpretation. The fact that a mathematical symbol belongs to a powerful formalism does not by itself show that the entity initially associated with it exists. Conversely, independent methods that manipulate a posited entity can provide reasons for ontological confidence beyond mere accuracy of a numerical prediction. These considerations expose how different components of a theory can have very different epistemic fortunes.

Reference therefore requires direct attention to the relevant chain of experimental and inferential practice. A causal claim with convergent support may outlast the theory that first introduced its vocabulary. Another theoretical term may lose its original meaning when a subsequent theory reorganizes the phenomenon. The no miracles argument is strongest when it can explain this uneven pattern rather than demand that every term in an empirically successful theory enjoy the same realist status.

Constructive Empiricism as a Substantive Rival

Bas van Fraassen's constructive empiricism offers a particularly exact alternative to scientific realism. In The Scientific Image, he proposes that the aim of science is empirical adequacy and that acceptance of a theory need involve belief only in its adequacy to the observable phenomena. The theory's language may still be understood literally. On this account, refusal to affirm an unobservable ontology is an epistemic restraint within a successful scientific practice, not a denial of the enterprise's value.

The comparison makes the no miracles argument more demanding. It cannot simply observe that antirealists rely on science, because constructive empiricists explicitly expect a successful theory to guide prediction and practice. The realist must show that approximate truth explains a determinate pattern of success that empirical adequacy plus the historical practices of theory construction cannot adequately explain. In Arguments Concerning Scientific Realism, van Fraassen addresses inference to the best explanation directly, placing its epistemic justification at the center of the dispute.

One result of this disagreement is conceptual clarity. A theory's observable consequences can be richly informative without settling what attitude one should take toward every unobservable element in its models. The realist may still argue that scientific practice demands stronger commitment, but that further claim needs independent argumentative support. Constructive empiricism shows why 'science works' is a shared observation rather than an automatic verdict for one philosophy.

The Darwinian Selection Reply

Van Fraassen also offers a selection-oriented response: the scientific theories visible at a particular moment have survived competition partly because they perform well. Successful theories remain in use and unsuccessful ones are discarded or revised. This historical fact changes what must be explained. It is less surprising to find that the surviving theories of an enterprise organized around empirical success are successful, even before any assertion of their approximate truth. The scientific realism debate discusses this Darwinian form of reply to the miracle argument.

The selection explanation should not be exaggerated. Selection by success explains why a community retains successful instruments and theories; it does not alone explain why any particular theory is capable of novel, accurate predictions across independently tested situations. A population of theories could be selected for performance and yet display remarkable theoretical successes whose proximate origins remain philosophically interesting. There are therefore two questions: why the archive is populated by successful survivors, and why certain theories succeed in the first place.

Both questions can be genuine. A persuasive account might use selection to explain the composition of our present theory repertoire and causal or representational accuracy to explain some individual achievements. The confrontation is most productive when the rival explanations are assigned their actual explanatory targets. It becomes misleading when the realist ignores filtering practices or when the antirealist treats survival alone as a complete explanation of every successful prediction.

The Pessimistic Meta-Induction

Larry Laudan's historical attack on convergent realism challenges the claim that empirical success reliably indicates truth or reference. In A Confutation of Convergent Realism, he examines the explanatory roles assigned to reference and approximate truth and mobilizes historical cases in which successful theories were superseded. The objection is formidable because it directly concerns the realism-relevant relation between empirical performance and an accurate description of unobservables.

The pessimistic meta-induction is often summarized as an argument that current theories will also be found false, just as earlier successful theories were. Its most informative role is narrower: the historical record supplies counterexamples to an unrestricted connection between success and the truth of a theory's central ontology. One need not estimate the proportion of false theories in all possible science to ask whether a proposed explanatory link has survived serious historical challenges. The mere possibility of later replacement is weaker evidence than a documented case where the very feature said to explain success was discarded.

Juha Saatsi's analysis of pessimistic induction also examines inferential pitfalls in the historical objection. A strong realist response investigates whether successful elements of the earlier theory were preserved in a successor. This reply is intellectually significant, but it must not turn every historical counterexample into confirmation by redescribing survival after the fact. The relevant question becomes which components were independently responsible for the success and which were causally idle or dispensable. The history of science thus forces a more discriminating version of the no miracles argument.

The Fresnel–Maxwell Transition

The transition from Fresnel's nineteenth-century wave optics to Maxwellian electromagnetic theory is a classic test of both sides. A successful mathematical account of optical phenomena was embedded within a picture involving a mechanical light-bearing ether. Later physics retained important relations while transforming the ontology and explanation of light. In Structural Realism: The Best of Both Worlds?, Worrall treats this history as a reason to examine structural continuity rather than insist on wholesale continuity of entities and mechanisms.

The case exposes several possible inferences from success. One may infer that the original theory was approximately correct in its mathematical account of certain regularities. One may also infer that its equations tracked stable relations not exhausted by the mechanical model attached to them. A stronger inference—that the original mechanical carrier existed as characterized—requires separate evidence and is not warranted merely by the continued usefulness of some equations. Historical continuity is selective, and the selection has explanatory importance.

The case also warns against an overly easy historical narrative. Different parts of optical theory changed at different times, and the survival of equations does not imply that every theoretical object retained the same interpretation. The philosophical point is confined to what Worrall's reconstruction establishes: significant theoretical performance and substantial ontological revision can coexist. A no miracles argument that survives this example must specify the level of structure or causal dependence to which its realist claim attaches.

Structural Realism and Selective Continuity

Structural realism seeks to reconcile the success of scientific theories with their historical replacement by identifying relational or mathematical structure as an especially defensible object of realist commitment. Worrall's 1989 structural proposal presents the alternative explicitly as a response to the contrary pressures of successful prediction and failed past ontologies. It is a research position with its own interpretive burdens, rather than a simple compromise between optimism and skepticism.

A structural claim is informative when one specifies which relations persist, under which transformations, and why this preservation explains the original success. Mathematical similarity alone may be too permissive. Different physical descriptions can share formal relations, and the significance of a retained equation depends on the empirical context in which it continues to work. A realist must therefore identify the structure being believed and the reasons for treating it as a feature of reality rather than as a useful representational convention.

Selective commitment is a philosophical gain because it places the target of belief within the successful practice itself. Its costs should also be acknowledged. A structural account may provide less information about the nature of unobservables than broader realism promised, and it can be difficult to separate structure from the entities that instantiate it. The proper judgment is neither that theory change defeats realism nor that all retained formalism certifies reality. Historical continuity offers a promising test for disciplined realism where the grounds of continuity can be reconstructed.

Experimental Intervention and Entity Realism

Theoretical predictions are one path to knowledge of unobservables; experimental intervention offers another. Ian Hacking's Representing and Intervening places experimental practices at the center of philosophical inquiry and argues for realist conclusions grounded in manipulation. An apparatus can be used to produce effects, discriminate causal influences, and stabilize measurement procedures, even while the most general theory remains contested or imperfect.

Intervention does not make philosophical reasoning disappear. It requires calibration, interpretation of instruments, and assumptions about causal pathways. Still, when a putative entity can be used in varied experimental contexts, evidence may converge on its reality by routes that are partly independent of any one global theoretical description. An argument for the existence of an entity can thereby be stronger than an argument for the literal accuracy of every model in which that entity figures.

This distinction informs the no miracles debate. Successful manipulation can offer a specific answer to what makes an experimental practice reliable. It may support a realist conclusion about an entity while leaving open a wider range of interpretations of the theory used to predict its behavior. The most useful philosophical comparison therefore involves both representational and intervention-based evidence, asking where they converge and where one exceeds the warrant supplied by the other.

The Brownian-Motion Example and Local Evidence

The history of molecular evidence illustrates why local arguments may be more compelling than broad declarations about the success of all science. Peter Achinstein's experimental argument for scientific realism examines experimental evidence, including Jean Perrin's work, to assess whether realism can be supported through concrete inquiry. The relevant achievement is not simply that some physical theory was famous or useful. It concerns how measurements and predictions related to the hypothesized molecular constitution of matter were experimentally assessed.

An experimental argument gains strength when alternative causal explanations face independent constraints. Evidence from different phenomena can support overlapping commitments, and the historian can examine how changes in instrumentation and theory affected those commitments. The resulting conclusion is local: particular molecules and relations receive support from particular experiments. To generalize from this achievement to all unobservable claims of all mature sciences requires a new and substantial premise.

This example reveals why a verdict about realism can be differentiated. Someone may have good reasons to believe in molecular entities while suspending judgment about a speculative cosmological ontology. That variation need not reflect inconsistency. It reflects differences in evidential pathways and available rivals. The broad no miracles intuition can motivate the inquiry, but the local documentary and experimental case must carry the actual warrant.

Idealization, Model Success, and the Target of Belief

Scientific models often succeed precisely because they isolate selected dependencies and omit complicating conditions. Nancy Cartwright's How the Laws of Physics Lie challenges the assumption that a fundamental explanatory equation directly describes reality in all its applications. Her examination of model construction shows why explanatory and predictive achievement must be interpreted in relation to assumptions, capacities, and particular experimental arrangements.

An idealized model may make accurate predictions in a controlled region while representing the system with entities that cannot exist under all its simplifying assumptions. A literal inference from model success to the truth of every modeling assumption would be self-defeating. More defensible realist reasoning asks which idealizations deliberately suppress irrelevant variation, which relations are preserved under correction, and which factors are independently supported as causal features of the target system.

This point concerns the structure of scientific representation. A successful map can guide action because it tracks relevant relations while omitting the texture of the terrain. The analogy is useful only if it remains an analogy: the epistemic contribution of an actual scientific model must be established by its tests and explanatory use. Approximate truth should identify what the model gets approximately right, instead of functioning as a blanket approval for every simplifying assumption that happened to accompany success.

Underdetermination and Unconceived Alternatives

Scientific communities choose among theories they have actually formulated. The best current explanation may be better than every available rival while still being inferior to a serious alternative nobody has yet devised. In Exceeding Our Grasp, P. Kyle Stanford develops this problem of unconceived alternatives by examining the history of theory replacement. Its force is not restricted to the abstract possibility that the same data might fit multiple theories. It concerns historically documented failures to imagine candidates that later became scientifically compelling.

This problem changes the modal language of explanatory comparison. 'Best among known alternatives' is a statement about an actual comparison set. 'Best possible explanation' is a much stronger claim about a space the investigator may not know. A no miracles argument that moves silently from the former to the latter overstates what scientific inquiry has demonstrated. Strong independent tests can reduce the range of plausible competitors, but historical inquiry cautions against equating the absence of a recognized alternative with evidence that none exists.

The point also helps explain why realism may differ by field or stage of inquiry. Where a proposition is supported by multiple comparatively autonomous experimental routes, the scope for radically different alternatives can be reduced. Where many models fit a thin or indirect evidential base, confidence in the presently preferred ontology should be more restrained. The historical challenge is a demand for appropriate calibration, not a requirement that all scientific judgment be suspended.

A Bayesian Reconstruction

Bayesian reasoning expresses an important logical condition of the no miracles argument. Let R denote a specified realist proposition about a theory, and S a specified success. Bayes' theorem gives posterior odds as prior odds multiplied by the likelihood ratio: P(R|S)/P(¬R|S) = [P(R)/P(¬R)] × [P(S|R)/P(S|¬R)]. Success favors R when it is more likely under R than under its negation. Yet the degree of favoring depends both on the ratio and on the prior odds. Jan Sprenger's probabilistic no miracles analysis develops why disciplinary context and theoretical stability matter to this reasoning.

Consider an illustrative and explicitly hypothetical assignment: P(R)=0.10, P(S|R)=0.90, P(S|¬R)=0.30. The posterior becomes 0.09/(0.09+0.27)=0.25. Success raises confidence in R from ten percent to twenty-five percent, but it does not make R likely on balance. A different prior or alternative likelihood produces a different posterior. These numbers describe no actual theory or empirical frequency; they demonstrate the mathematical insufficiency of inferring high posterior probability from predictive strength alone.

This formal result also directs attention to the meaning of R. If R means that a particular predictive relation is approximately correct, it may have a different prior and likelihood than the claim that a complete unobservable ontology is true. A sound evaluation must state both propositions, identify the background knowledge used in assigning probabilities, and make clear whether S was an independently sampled outcome or a success selected for attention.

The Base-Rate Problem

A common probabilistic reconstruction begins with a strikingly successful theory and asks how likely that success would be if the theory were false. The answer can look favorable to realism because successful predictions seem difficult to obtain by chance. Yet the prior distribution of true and false candidate theories matters, as does the process through which a theory was selected for presentation. Magnus and Callender's base-rate critique of realism makes that background explicit. Even a diagnostic indicator with a favorable likelihood ratio can yield many false positives when the underlying population has a low proportion of true candidates.

Consider another purely illustrative population. Imagine one hundred independent candidate models, ten of which approximate the causal structure sufficiently well and ninety of which do not. If nine of ten approximately adequate models and nine of ninety inadequate models pass a demanding test, eighteen passing models remain, half of them adequate. Test success is informative; it does not warrant an unrestricted estimate that successful theories are mostly true without counting the population from which the successes emerged. These numbers serve solely to clarify inference and represent no historical frequency of scientific truth.

Scientific communities do not normally draw their theories from a random and fully specified population. Theory generation and refinement are conditioned on background knowledge, methodological constraints, previous models, mathematical tractability, and institutional resources. For that reason, estimating an empirical base rate is extraordinarily difficult. The lesson is nevertheless precise: the no miracles argument needs assumptions about the reliability of theory generation and selection, and those assumptions themselves require defense. The mere presence of impressive surviving theories cannot settle their probable truth by itself.

Selection Effects and the Survivor's Perspective

Scientific success is observed after research institutions have invested in improving, comparing, discarding, and replacing candidates. The theories emphasized in textbooks are often the survivors of such processes. Constructive empiricism draws attention to this selection history. Van Fraassen's The Scientific Image interprets successful theories as those that have survived competitive testing, drawing an analogy with biological selection. A history dominated by successful survivors can arise even when survival is explained principally by empirical adequacy rather than approximate truth about all posited unobservables.

The realist has a significant reply. Selection by observed fit does not automatically explain how a theory predicts unprecedented phenomena, extends reliably to new regimes, or sustains quantitatively precise performance across independent instruments. If a theory's continued success was produced by repeated modification to prior evidence, the selection story has substantial force. If a theory predicts results that were neither used to construct nor tune it, the selection story requires further detail about why the selected candidate could do so. The dispute turns on the independence and information content of success, not on the metaphor of selection alone.

Researchers can improve the evaluation by distinguishing the historical training set of evidence from subsequent testing, recording failed rivals where possible, and stating what level of predictive achievement a selection mechanism would be expected to produce. Selection bias becomes philosophically manageable when the provenance of success is exposed. It becomes misleading when an investigator treats the final surviving theory as if it emerged from the outset in the form that later proved successful.

The Alleged Circularity of Abduction

The no miracles argument employs inference to the best explanation to support the reliability of theories, and scientific realists often regard inference to the best explanation as a reliable scientific method. Critics consequently ask whether the argument presupposes exactly what it seeks to establish: that explanatory success provides a guide to truth. Harman's classic abductive account identifies the inferential pattern, while Lipton's explanatory reasoning analysis examines the normative demands made upon it.

The strongest version of the circularity objection distinguishes two questions. One is whether explanatory reasoning can legitimately alter degrees of belief in everyday and scientific settings. The second is whether that reasoning justifies the global claim that mature successful sciences approach truth about unobservable structures. The first may be defended through many comparisons of explanations, including familiar causal inferences. The second requires an additional bridge from local reliability to the reliability of a historically evolving scientific enterprise. Using the word 'explanation' in both contexts does not supply the bridge.

The realist can answer circularity without pretending to derive realism from neutral logical axioms. Methodological judgments can receive mutual support from documented practices, independent testing, instrument calibration, causal interventions, and successful prediction. Such justification is defeasible and partly holistic. The important issue is whether the mutual support contains genuinely independent pathways or merely restates one conclusion in differently worded premises. A good no miracles defense identifies those pathways and shows why competing accounts explain the same facts less satisfactorily.

Contrastive Explanation and Serious Alternatives

An inference to the best explanation is contrastive: a candidate explains an outcome better than alternatives relative to specified background knowledge. A no miracles argument must therefore specify the competing accounts. Possible rivals include empirical adequacy without ontology, structural preservation through theory change, model-based instrumental reliability, selective realism about tested components, and historically local causal explanations. Different rivals can agree that predictions are impressive while differing over what precisely explains their reliability.

Suppose a theory describes the causal role of a particular entity with enough accuracy to generate a novel prediction. A selective realist can credit the portion of its structure implicated in that prediction; a global realist may attribute success to the approximate truth of the entire theoretical system. The same successful prediction may favor both against a crude chance hypothesis while offering little discrimination between them. To identify a preference, investigators must ask whether the larger realist commitment predicts or explains something the selective account leaves unaccounted for. Psillos's study of scientific realism provides a sophisticated context for these selective commitments.

Conversely, if a rival explanation was constructed expressly to accommodate the successes of the target theory and carries additional untested assumptions, it may have a weaker explanatory profile. The comparison must be symmetrical: auxiliary assumptions supporting realism and antirealism should each be counted when relevant. Explanatory strength arises from a documented comparison of scope, independent support, simplicity where justified, and empirical discriminability. The superiority of a candidate is an argument to establish, never an honorific assigned in advance.

Disciplinary Context and Theoretical Stability

The no miracles argument is often presented as if scientific success had a constant epistemic meaning across disciplines and eras. It does not. Theoretical maturity, observational access, opportunity for independent tests, rates of model replacement, and the availability of causal interventions vary markedly. Jan Sprenger's probabilistic reconstruction links realist support to the relevant probabilistic setting and asks how scientific context shapes the significance of success.

In a stable field, predictions produced by an established framework may be tested in new applications that were not central to its original construction. If rival frameworks repeatedly fail or require substantial special adjustment, additional success can increase confidence in particular theoretical commitments. In a rapidly changing field with extensive model tuning and sparse external testing, a similar numerical degree of fit may offer weaker grounds for ontological commitment. The difference arises from evidential independence and competition, rather than a verdict that one discipline counts as genuine science and another does not.

The word 'stability' itself must be interpreted cautiously. A theory can remain institutionally dominant because competing approaches lack funding, because instruments are optimized for its conceptual vocabulary, or because its predictive domain is unusually narrow. Historical persistence is informative only when joined to documented opportunities for demanding criticism. A strong application of the no miracles argument specifies which theory survived, what the surviving components were, how often relevant alternatives were explored, and which achievements remained surprising relative to prior knowledge.

Independent Error Checks

Successful prediction can coexist with shared systematic error. Two instruments built on the same calibration assumption can agree while both misleading the investigator. Two apparently independent theoretical derivations can rely upon the same approximations. The epistemic standing of scientific success therefore depends on identifying the actual causal and methodological relationships among tests. Independent errors are difficult to manufacture by declaration; they require materially distinct pathways of observation, instrumentation, and inference.

Hacking's analysis of intervention illustrates why manipulation can contribute a line of support that differs from passive agreement between theoretical predictions and recorded data. Achinstein's experimental realism inquiry likewise focuses philosophical attention on how particular experimental results count as evidence for specific claims. These approaches strengthen a local realist case when rival causal accounts would have to explain multiple convergent forms of inquiry, although they still leave room for measurement error, hidden common causes, and revision of the larger theory.

The practical implication for a no miracles assessment is a demand for an evidence map. A critic should be able to see how each reported success was produced, which background commitments it shares with others, whether surprising outcomes were preregistered or otherwise documented before observation, and what correction mechanisms actually operated. A dozen dependent tests may provide less warrant than a handful of strongly independent ones. The criterion is the structure of evidential dependence, not the number of supportive anecdotes assembled after the fact.

What Successful Artificial Systems Show

Computational science provides a contemporary occasion to sharpen the distinction between prediction and ontological explanation. A machine-learning system may predict a measurable outcome by exploiting statistical regularities that are stable within its training and deployment distributions. Such success can support a claim about predictive performance under specified conditions. It does not automatically establish that every internal feature corresponds to a real causal entity or that the system's representation supplies a scientifically adequate explanation of the target domain.

The same reasoning applies to interpretable mechanistic models. Their usefulness must be assessed against domain-specific tests, and a faithful physical interpretation requires a connection to independently evaluated causal or structural claims. Cartwright's critique of literal laws is relevant because successful idealized representation has never guaranteed literal truth of all modeling assumptions. Stanford's unconceived-alternatives challenge remains relevant because strong contemporary performance can coexist with future reorganization of concepts and ontological commitments.

This application is a philosophical extrapolation of the established argument, not an empirical assessment of a particular artificial system. Predictive achievement, model explanation, and the public justification of scientific conclusions have different evidential requirements. The no miracles argument supplies a reason to investigate why computational success occurs, which causal features survive independent tests, and which internal constructions are dispensable. It cannot replace those investigations, and it supplies no warrant for claims about machine consciousness or personhood.

Historical Case Comparison: A Controlled Reading

Three contrasting episodes show how a common explanatory slogan yields different epistemic conclusions. First, the Fresnel-related successes discussed in Worrall's structural realism essay show that predictive relations can remain powerful across a major change in physical ontology. Second, the empirical evidence surrounding molecular reality discussed in Achinstein's experimental defense of realism illustrates how local claims about entities can be supported through converging evidence. Third, the histories examined by Laudan's critique of convergent realism show that past scientists could obtain genuine achievements within theoretical frameworks that were later rejected in important respects.

These episodes differ in relevant features. In the first, the philosophical issue is the retention of mathematical structure through historical transition. In the second, the issue is the evidential force of a specific experimental pathway and its alternatives. In the third, the issue is the danger of treating historically impressive theory success as sufficient evidence for wholesale approximate truth. Their juxtaposition does not constitute a controlled experiment or representative quantitative sample of scientific history. It is a conceptual comparison of documented cases chosen because they stress different assumptions of a global inference.

The comparison supports a qualified conclusion. Scientific success often provides information about reality, and some cases justify robust claims about entities or relations. It also shows that the content justified by success can be more selective than the full ontology of the theory that generated it. The historical record offers reasons to improve the granularity of realist commitments rather than to accept the universal reliability of successful theories on the strength of their success alone.

A Strong Realist Defense

The most substantial defense of the no miracles argument begins with achievements that are difficult to anticipate under independently specified rivals. A theory generates a quantitatively risky prediction about a new domain, the prediction is borne out through reliable measurements, and related theoretical commitments prove useful in further independent research. The realist then argues that the stability of this achievement is explained by the theory's successful grasp of some underlying causal or structural dependencies. Putnam's original explanatory challenge gains philosophical force when stated in this demanding form rather than as praise for science in general.

The defense can accept several concessions. Historical theories have failed; theoretical vocabulary can change; models depend on idealizations; the epistemic force of success varies; and no posterior probability follows without assumptions about alternatives. These facts do not abolish the possibility that independently confirmed structure or successful causal intervention gives good grounds for belief. Indeed, the ability to articulate a specific referent or relation that survived critical inquiry is an argumentative advantage over treating every theory as uniformly reliable.

What realism must still establish is a discriminating explanation: why does approximate truth of this component give the observed pattern of success a probability or intelligibility that rival accounts lack? Once that question is answered locally, the realist has an evidential achievement with an identifiable target. The global claim that successful science in general tracks truth can then be considered as a generalization requiring separate historical and methodological support.

A Strong Antirealist Defense

The most substantial antirealist reply grants the reality of predictive achievement and the legitimacy of treating success as evidence that a theory works within tested domains. Constructive empiricism interprets scientific acceptance as commitment to empirical adequacy. Van Fraassen's constructive empiricist argument shows how this commitment can be maintained while withholding belief in a theory's claims about unobservable entities. The antirealist also points to the history of theory replacement, the selection of survivors, and the availability of alternative frameworks that the successful theory's authors did not anticipate.

The argument becomes especially powerful when realists move from a limited body of predictions to a complete ontology, or when a theory's predictive accuracy was obtained through extensive adjustment to known outcomes. A statement that all confirmed predictions would be miraculous without approximate truth overstates what competing accounts must explain. Empirical methods are designed to improve predictive performance, and successful selection is consequently an outcome a methodological account can anticipate. Laudan's historical counterexamples show why success and truth cannot simply be equated.

Antirealism in turn faces the challenge of accounting for particular successes that appear unusually independent of the observations used to construct the theory. Its account should specify why a selected model performs well in unexpected settings and which empirical or methodological resources explain the extension. It is methodologically insufficient to repeat 'selection' in the face of such cases. Productive disagreement requires that each side identify the explanatory work its favored mechanism actually performs.

Scope, Method, and Source Limits

This monograph employs historical reconstruction of philosophical arguments, analysis of logical and probabilistic relations, and comparison of selected case studies. Its primary documentary axis runs through Putnam, Boyd, van Fraassen, Laudan, Harman, Worrall, and later probabilistic treatments. The analysis distinguishes a source's published claims from interpretations advanced here, and it gives numerical examples explicitly hypothetical status. No original experiment, quantitative systematic review, disciplinary success-rate estimate, or new named philosophical theory is presented.

Several cited monographs and chapters are represented by authenticated publisher descriptions or bibliographic pages rather than an unrestricted reading of their full book texts. The Stanford Encyclopedia entry is used as a secondary synthesis rather than as evidence of priority. Directly accessible journal abstracts and selected open scholarly papers permit finer verification of specific arguments. These access tiers are recorded individually in the evidence ledger. The absence of a complete page-level critical edition constrains quotation and the precision of some historical claims; hence this manuscript uses sparing paraphrase and avoids pretending to have established the wording of a source beyond the accessible record.

The comparison of Fresnel-era optics, molecular evidence, and discontinued theories is deliberately philosophical. A fully historical account of each episode would need archival data, laboratory notebooks, original experimental reports, and domain-specific historiography extending beyond the present dossier. The present conclusions concern what those well-attested philosophical case discussions establish about the no miracles argument. They have been framed so that limited evidence yields limited claims, while the logic of the argument can be assessed directly.

Integrated Assessment

The central finding is that scientific success creates a genuine explanatory question with several competing answers. Approximate truth about selected structures and mechanisms can sometimes explain successes that would otherwise be difficult to understand. Selection for empirical adequacy, model adjustment, methodological learning, and dependence among tests can also explain substantial portions of the same record. Therefore, an inference based on success must specify its success measure, contrast class, prior information, and particular target of belief. The philosophical weight of any outcome depends on how successfully these items are reconstructed.

The historical criticism changes the unit of assessment. Laudan's challenge to convergence makes blanket inferences vulnerable to counterexamples. Worrall's structural continuity response illustrates how some commitments can survive where broader ontologies do not. Magnus and Callender's base-rate analysis and Sprenger's Bayesian treatment show why the distribution of candidate theories and the structure of the evidence affect formal assessment. Together these contributions support disciplined, conditional realist reasoning rather than a universal entailment from success to truth.

The resulting methodological recommendation is substantive but remains an interpretation within an established debate. Identify the successful proposition precisely; investigate whether its achievement was novel and evidentially independent; reconstruct plausible rival explanations; assess historically retained and discarded commitments; and articulate any realist conclusion at the degree of specificity the record warrants. This approach preserves the explanatory insight motivating the argument while testing the conditions under which that insight deserves epistemic authority.

Conclusion

The no miracles argument explains why successful science presents a philosophical challenge: stable, accurate, sometimes unexpected predictions invite an account of how theories attain their grip on the world. Its classic realist answer holds that the achievement would be astonishing unless some central claims of successful theories were approximately true. The argument's enduring value lies in asking what accounts for this success and why scientific practice can be reliably informative about features beyond direct perception.

The claim becomes philosophically defensible when it is specified. Scientific success ranges from retrospective fit to genuinely risky novelty; theories contain both well-supported dependencies and provisional modeling assumptions; and historical transitions can preserve mathematical or causal achievements while revising asserted entities and mechanisms. Bayesian reconstruction adds that successful prediction supports a particular realist proposition only in relation to prior odds and comparative likelihoods. The stronger the independent evidence and the clearer the surviving commitment, the more persuasive the local realist inference becomes.

Historical and probabilistic objections set enduring limits on the global argument. Laudan's examples challenge automatic convergence; van Fraassen's selection account explains why surviving theories are expected to perform; and the base-rate and unconceived-alternatives problems caution against assigning a high probability of truth merely because available theories outperform familiar rivals. The realist must meet these objections with particular, contrastive accounts of actual success. The antirealist must likewise explain achievements that are difficult to reduce to accommodation or survival alone.

The final conclusion is that the no miracles argument is a powerful program of explanatory assessment, and its universal formulation exceeds what scientific success alone demonstrates. Its strongest defensible applications are local, evidentially explicit, historically informed, and selective about the commitments they justify. The philosophical significance of success arises from its tested connections to the world and from the clarity with which those connections withstand alternative explanation.