Map of the Realism Debate Scientific success → closeness to truth? → unobservable entities → theory change Realist answer — successful theories describe the structures or entities of the real world with sufficient approximation Empiricist answer — empirical adequacy is sufficient; no additional commitment to unobservable entities is required Intermediate answer — examine structure, causal intervention, perspective and domain limits separately Problem The success of science is extraordinary. From planetary motion to gene editing, from semiconductors to immunology, from telescopes to particle accelerators, scientific theories make possible prediction, control and explanation far beyond ordinary conjecture. The question is whether this success occurs because science captures, more or less correctly, the real structure of the world, or merely because theories are empirically successful instruments. The new debates in scientific realism unfold many versions of this question. Scientific realism is not a single theory. Three claims can usually be distinguished. First, the world described by science exists independently of human belief. Second, successful scientific theories can yield genuine knowledge of unobservable entities such as electrons, genes, black holes, viruses or fields. Third, the best theories may be not merely calculational tools but more or less true, or approximately true, descriptions. These three claims do not have equal strength, and modern debate centres precisely on their differences. The best-known argument on the realist side is the no-miracles argument. If scientific theories did not capture some genuine aspect of the world’s structure or causal relations, their long-term predictive success, capacity to discover new phenomena and success in technological intervention would look almost miraculous. The argument therefore treats scientific success as evidence of closeness to truth. But evidence and proof are not the same thing. The opposing side looks to history. Many theories were highly successful in their own time but their central entities or explanations were later abandoned—for example, various forms of ether, phlogiston, caloric fluid and certain older theories of disease. From this history arises the pessimistic meta-induction: if successful theories of the past were later shown to be wrong, how much confidence should we place in the truth of successful theories today? The realist reply is that earlier theories were not wholly failures; some relations, mathematical structures, causal capacities or truths within a limited domain survived in later theories. A third problem is underdetermination. Available evidence can sometimes fit more than one theory. An experiment does not test a theory in isolation; background assumptions, instruments, statistical models, initial conditions and auxiliary hypotheses are also involved. Thus, when we say that ‘the evidence supports this theory’, we must also ask what role alternative explanations, auxiliary assumptions and measurement procedures are playing. A fourth problem concerns the relation between truth and utility. A theory can be highly useful without all of its ontological claims being true. Conversely, some deep structural feature of a theory may be correct while a simpler model is more useful for practical calculation. Scientific practice itself operates at many levels—fundamental theories, effective theories, idealised models, statistical tools and computational simulations. A single true/false binary cannot capture this layered practice. A fifth problem is the collective character of modern science. Scientific claims are not merely relations between theory and experiment; instrument construction, data cleaning, statistical choices, laboratory standards, peer review, replication, open data, computational code and institutional incentives also shape knowledge production. This social mediation does not abolish an objective world, but it destabilises the simple picture in which science merely ‘reads nature directly’. A sixth problem concerns perspective. Different measurement systems, scales and scientific aims can represent the same object differently. Cell biology, histology, population genetics and ecology capture different aspects of a living system. Such plurality of perspective does not mean that reality is arbitrary; the question is where to draw the boundary between perspective-dependent representation and perspective-relative truth. Core Proposition The central proposition of this chapter is that the strongest form of scientific realism is not the simple view that ‘whatever science says is true’. A more durable realism must be fallibilist, level-sensitive and evidence-based. Successful science captures some aspects of objective resistance, causal structure and real relations; but not every term, model and ontological image in a theory is equally trustworthy. For this reason, ‘realism versus anti-realism’ is not a simple binary. Constructive empiricism holds that the aim of science may be empirical adequacy rather than full belief in unobservable entities. Structural realism argues that relations or structures may remain more stable through theory change. Entity realism places special confidence in unobservable entities when causal intervention involving them succeeds. Perspectival realism accepts the historical and instrumental perspectives of representation while attempting to preserve objective standards. Parallel Philosophy will read this debate through seven criteria: objective resistance, plurality of evidence, causal intervention, theory change, mediation by measurement, limits of perspective and self-critique. Scientific success here is neither proof of final truth nor merely the product of social agreement. Success is important evidence, but its meaning must be tested against domain, history and alternative explanations. In newer forms of scientific realism, even the expression ‘closeness to truth’ requires caution. A theory’s mathematical relations, causal directions, classifications, measurement structures and descriptions of entities can be right or wrong at different levels. One part of a theory may survive while another is abandoned. Without such partial assessment, both the successes and the failures of scientific history are distorted. Nor is the meaning of ‘real’ uniform in this chapter. Some entities can be controlled through direct intervention; some are tested only through indirect signals; some appear as mathematical structures; some are useful as real patterns at an effective level. Philosophical discipline must clarify these differences rather than end every question with one word. Principal Arguments First argument—scientific success is genuine evidence for realism, but not decisive proof. The more novel predictions, independent measurements and causal interventions converge on the same theory, the harder it becomes to treat success as mere coincidence. Yet alternative theories, selection bias and domain limits must also be examined. Second argument—the no-miracles argument itself depends on inference to the best explanation. The question therefore arises: is closeness to truth alone the best explanation of scientific success, or do methodological selection, the forgetting of failed theories, data fitting and model flexibility also play roles? The argument is strong, but not self-validating. Third argument—the pessimistic meta-induction does not destroy scientific realism, but it makes realism more modest. Defeat of past theories shows that successful vocabularies and entity claims can change. The realist must show what survives—structure, causal relations, limited-domain truth, mathematical form or the entities referred to. Fourth argument—complete rupture in theory change may be less common than selective continuity. Newtonian gravitation is no longer regarded as the final fundamental truth after general relativity, yet its predictions remain reliable at low velocities, in weak fields and in many engineering contexts. This kind of domain-limited continuity weakens the claim that older theories are simply wholly false. गजेन्द्र ठाकु र Fifth argument—the attraction of structural realism lies in its attempt to preserve continuity of relations across theory change. But the word ‘structure’ can itself be obscure. Does it mean mathematical form, a causal network, symmetry, an ordering of relations, or merely successful equations? The strength of structural realism depends on giving a clear answer. Sixth argument—entity realism increases the importance of experimental intervention. When scientists posit an unobservable entity, build instruments around it, control its effects and obtain the same causal results in different experiments, confidence in its reality increases. Yet intervention itself is not theory-neutral; measurement and interpretation of instruments remain necessary. Seventh argument—constructive empiricism is an important reminder of scientific modesty. A theory’s successful fit with observable phenomena may be considered sufficient for the aims of science; belief in unobservable entities is a further philosophical step. The realist must explain why that additional commitment is justified. Eighth argument—underdetermination does not have equal force everywhere. In some contexts several theories can explain the same data; in others, novel predictions, independent experiments and cross-checks progressively eliminate alternatives. Recognising that underdetermination exists is therefore different from assessing how extensive it is. Ninth argument—the Duhem–Quine form of confirmation holism reminds us that when an experiment fails, the fault may lie not only in the main theory but also in auxiliary assumptions or in the interpretation of instruments. But this does not imply that any theory can always be rescued. Long-term revision, independent instruments and multi-method testing constrain auxiliary excuses. Tenth argument—a model is an active mediator between theory and reality. Scientists often use idealised models—frictionless planes, ideal gases, representative agents and point masses. A model can be a deliberately false simplification while still yielding causal or structural understanding. We therefore need criteria to distinguish useful falsehood from misleading falsehood. Eleventh argument—scientific classifications themselves change. Categories of disease, species, genes, planets, mental disorders or particles can alter over time. Such change does not make the objects themselves arbitrary; the aims of classification, measurements and theoretical background change. Realism requires a subtle account of the relation between object and category. Twelfth argument—measurement is not merely passive recording. Measuring temperature, intelligence, poverty, climate risk or biodiversity requires conceptual decisions, instruments and statistical transformations. Mediation by measurement does not destroy objectivity; rather, it increases the need for transparency, calibration and independent retesting. Thirteenth argument—the strength of perspectival realism is that it accepts human instruments and historical standpoints while preserving the resistance of reality. Two perspectives can capture different aspects of the same object, but not every claim is equally correct. Inconsistency, prediction, intervention and mutual translatability become tests. Fourteenth argument—scientific pluralism is not necessarily anti-realist. Complex systems may require theories at different levels simultaneously—molecular, cellular, organismic, population and ecological. If each level captures real regularities, a multilevel realism may be more appropriate. Fifteenth argument—causal explanation must be distinguished from mere correlation. A model can predict with great accuracy while misrepresenting the causal structure. AI-based prediction makes this issue especially sharp: high predictive success does not automatically amount to a realist causal explanation. Sixteenth argument—scientific realism does not by itself yield moral or political conclusions. Between ‘science shows this’ and ‘society ought to do this’ lie normative values, distributions of risk, justice and policy judgement. Turning factual realism into ethical scientism would be a serious error. Seventeenth argument—the replication crisis is not the end of scientific realism but a test of science’s institutional capacity for self-correction. Negative results, preregistration, data sharing, multi-laboratory replication and statistical reform can strengthen or weaken claims. The discovery of error is not evidence against science; identifying and correcting error can be one of science’s strengths. Eighteenth argument—computer simulation raises new evidential problems. A simulation is not a direct copy of nature; it depends on equations, initial conditions, numerical methods and computational limits. Yet when different models, independent data and real observations converge, its evidential value increases. Nineteenth argument—scientific hypotheses generated by artificial intelligence make the question of realism more difficult. If a system discovers a model that predicts successfully but remains opaque to human understanding, should it be called ‘close to the truth’ or merely predictively successful? Provenance, retesting, causal intervention and explanatory transparency become new criteria. Twentieth argument—the Parallel conclusion rejects both simple formulae, ‘success = truth’ and ‘theory = instrument’. Scientific claims should be tested at least five levels: empirical adequacy, causal intervention, structural continuity, resilience through theory change and independent retesting. This multilevel standard makes realism more fallibilist. Pūrvapakṣa First Pūrvapakṣa: Science produces technology, therefore its theories must be true. Reply—technological success is important evidence, but approximately correct models can also work effectively within limited domains. Success gives a reason in favour of truth; it is not proof of complete truth. Second Pūrvapakṣa: Successful theories of the past turned out to be wrong, so trusting present science is foolish. Reply—history teaches caution, not universal scepticism. We must examine carefully how much of an older theory’s successful content, structure, measurement relations and limited-domain usefulness survived. Third Pūrvapakṣa: If an unobservable entity cannot be seen, belief in its existence is mere faith. Reply—direct visibility is not the sole scientific criterion. Causal effects, independent instruments, different experiments, predictions and interventions can jointly strengthen the case for an unobservable entity. Fourth Pūrvapakṣa: If the same data can be explained by several theories, truth is unattainable. Reply—underdetermination is a real problem, but not a permanent verdict. New data, new measurements, causal intervention and consequences in other domains can distinguish alternatives over time. Fifth Pūrvapakṣa: Science is a social institution, therefore scientific truth is produced by power. Reply—institutions, funding, prestige and classification influence knowledge production; yet the resistance of nature, failed predictions, replication and independent experiments are not controlled merely by social desire. Social critique is not the enemy of objectivity; it can improve its conditions. Sixth Pūrvapakṣa: Structural realism solves every problem. Reply—it is an attractive intermediate position, but the meaning of ‘structure’, the identification of structure, structural continuity through theory change and the ontology of structure without objects are themselves contested. Seventh Pūrvapakṣa: Once perspectival truth is admitted, relativism is inevitable. Reply—accepting perspective-dependent measurement is not the same as saying ‘all perspectives are equal’. Mutual checking, resistance from reality, prediction, measurement coherence and domain limits discipline perspectives. Eighth Pūrvapakṣa: Classical Indian realism already provides a ready-made answer to modern scientific realism. Reply—that would be anachronistic. Classical traditions offer useful arguments about objects, pramāṇa, inference and error, but modern laboratories, mathematical physics, statistics and theory change have different historical structures. गजेन्द्र ठाकु रक समानान्तर दर्शन — खण्ड २ Uttarapakṣa First reply—construct realism claim by claim. Instead of believing an entire theory all at once, examine its entities, relations, mathematical structures, causal directions and domain-specific uses separately. This division reveals more nuanced forms of continuity across historical change. Second reply—distinguish types of success. Mere data fitting, prediction, novel prediction, causal intervention and independent replication do not carry equal evidential weight. Novel and independent successes generally provide stronger support for realist confidence. Third reply—do not hide failure. Scientific realism must include not only histories of success but failed theories, dead ends, publication bias and replication failure. Without self-critique, realism becomes propaganda. Fourth reply—measurement methodology is itself a philosophical subject. Abandon the claim that ‘the data speak for themselves’. The process by which data are produced, instrument calibration, exclusion rules, statistical assumptions and uncertainty should be made public. Objectivity becomes stronger through transparent procedures. Fifth reply—it is useful to accept a multilevel reality. Even if fundamental physics is real, patterns in biology, psychology or the social sciences are not therefore ‘mere illusions’. If higher-level patterns support independent explanation, intervention and prediction, their domain-specific reality may deserve philosophical recognition. Sixth reply—treat perspective as a boundary, not a prison. Every scientific perspective is tied to instruments, scales and aims; yet different perspectives can be translated, compared and jointly tested. Objectivity can develop here as coherence across multiple perspectives. Seventh reply—keep scientific realism distinct from scientism. Science may be our best instrument for many facts about natural and social worlds; but the claim that every question of moral value, political legitimacy, beauty, meaning or religious experience can be fully resolved by scientific method is a separate philosophical thesis. Eighth reply—make fallibility the final criterion. The credibility of scientific realism lies not in treating current theories as final, but in the ability to revise commitments in the face of new evidence, new theories and new instruments. ‘Reality exists’ and ‘we have completely known its nature’ are not the same claim. Indian Dialogue Many traditions of Indian philosophy contain deep disputes about external objects, the validity of knowledge, inference, error and perception. It would be improper to call this material a precursor of modern scientific realism, but dialogue at the level of problems is fruitful. Nyāya asks about the existence of external objects, the reliability of pramāṇas and the correction of error; different currents of Buddhist epistemology question the limits of perception, conceptualisation, apoha and mental representation; Vaiśeṣika classification offers an ontological structure of substance, quality, action and relation. A special contribution of the Nyāya tradition is that the validity of cognition is not settled merely by ‘it seems so to me’. Perception, inference, comparison and testimony have different conditions, defeaters, errors and modes of correction. The instruments of modern science are different, but the discipline of claim testing—cause, vyāpti, counterexample and independent confirmation—is comparable at the level of problems. The Buddhist counterposition does not offer a single view of external reality. The Dignāga–Dharmakīrti distinction between perception and conceptual construction, the analysis of apoha and the momentary character of cognition are not directly identical with modern theory construction. Yet the question remains useful: do concepts capture objects exactly as they are, or does the structure of cognition itself produce classifications? Modern debates over scientific models and classifications can engage this problem-axis. It would be mistaken to declare Jain anekāntavāda a direct precursor of modern perspectival realism. Their historical aims, ontologies and logical methods differ. Yet the many-sidedness of an object, the conditional character of assertions and criticism of wholly one-sided claims offer a cautious comparative resource for the question of plurality of scientific perspectives. Questions of testimony in Mīmāṃsā and Vedānta have only limited use in this chapter. Scientific realism is chiefly a modern problem concerning empirical theories, unobservable entities, experiment and theory change. Comparison is useful only when the original technical contexts remain intact; otherwise classical terms become coverings for modern meanings. The conclusion of the Indian Dialogue is therefore not that answers to modern science were already present in the śāstras. Rather, older philosophical disciplines concerning object, pramāṇa, error, inference, standpoint and the limits of knowledge can provide alternative ways of questioning modern debates—not genealogical identity. Mithila’s Parallel Perspective Mithila’s Parallel contribution to scientific realism is not a claim of priority but a methodological discipline. A mode of questioning inspired by Gaṅgeśa and the Navya-Nyāya tradition would ask: what is the claim about? What is its locus? What is the relation? What is the reason in the inference? How is vyāpti established? What is the defeater? Where is the counterexample? These ordered questions should not be translated mechanically into the language of modern philosophy of science, but the habit of refining claims is useful. The strength of Pūrvapakṣa–Uttarapakṣa in Mithila’s intellectual culture is especially instructive for scientific controversy. Winning by presenting the strongest opponent in a weakened form is not śāstrārtha. Scientific realism must confront the strongest forms of constructive empiricism, underdetermination, pessimistic meta-induction and social critique of science. The Parallel method applies directly here the maxim ‘evidence is plural; criteria are not arbitrary’. Experimental data, statistical evidence, historical theory change, causal intervention, model success, instrument reliability and the critical processes of scientific communities are all evidence, but they do not all carry equal weight. Mithila’s local experience is not direct proof of scientific realism, but it can illuminate relations between science and society. In areas such as flood prediction, models of river courses, agrometeorological information, public-health measurement, language surveys or archaeological dating, scientific claims can conflict with, be corrected by or be confirmed by local experience. Local knowledge should be treated neither as a substitute for science nor as automatically inferior; its evidential value should be tested against explicit criteria. From this perspective Mithila can become not merely a ‘regional example’ but a testing ground for evidential method. Local data, archives, oral knowledge, satellite measurements, field verification and historical documents can be set together to determine which claims are supported at which level. Parallel Philosophy thus makes realism more accountable. Five Levels for Testing Scientific Claims Empirical fit → independent retesting → causal intervention → continuity through theory change → domain limits A single successful result is not final truth — convergence among multiple independent criteria increases confidence Contemporary Applications In climate science the question of scientific realism has direct political meaning. Global temperature, ocean heat, ice melt and atmospheric carbon are tested through multiple independent measurement systems. Models contain uncertainty, but uncertainty does not equal ignorance. Convergence across different models, historical data and physical causal relations provides combined evidence. In epidemiology, models depend on changing conditions, behaviour, immunity and data quality. The later revision of early estimates does not show that science is ‘false’; the question is whether the change responded to new evidence or concealed methodological error. Transparency in revision is an important condition of realist confidence. गजेन्द्र ठाकु र In AI-based drug discovery, high predictive success is attractive. But whether a model captures molecular causal relations or only correlations in training data is a different question. Without laboratory validation, biological mechanism, independent replication and clinical trials, computational success is insufficient for a realist claim about a genuinely effective therapeutic entity. In astronomy, claims about black holes, exoplanets or dark matter do not depend on direct images alone. Gravitational effects, spectra, orbital behaviour, independent telescopes and predictions from different theories combine to form evidence. Such examples show the inadequacy of the simple criterion ‘seen = real’. In particle physics, unobservable entities are established through complex instruments, statistical significance and theory-guided signals. A single anomalous signal should not immediately be declared a new entity; independent experiments, analysis of background effects and replication are required. Scientific realism here depends on institutional restraint. Realism is still more complex in the social sciences because classifications can themselves alter behaviour. Measurements of unemployment, caste, poverty, crime, gender or educational quality are tied to social definitions and institutional rules. This does not imply that ‘social objects are unreal’; rather, classification, causality and normative effects must be examined separately. Changes in disease categories in medicine provide an excellent example of scientific realism. Biological mechanisms, symptom clusters, genetic risks and treatment responses may not align perfectly. Domain-sensitive realism allows that a category can capture a useful and real pattern even though future classification may revise its boundaries. Open science can strengthen the institutional basis of realism. When data, code, preregistration, negative results and replications are public, scientific claims move away from private prestige and become objects of shared testing. Objectivity is not only a philosophical position; it is also a transparent institutional practice. Chapter Conclusion The new debate over scientific realism is not the simple question of whether to trust science. The real questions are: how much ontological conclusion may be drawn from scientific success; how far the history of theory change should limit present confidence; what evidence is sufficient for the reality of unobservable entities; and how measurement, models, perspective and institutions mediate objectivity. The Parallel answer favours fallibilist realism. The world offers resistance independently of human wishes; scientific methods can yield genuine knowledge of it; but not every term in every successful theory is final truth. Multiple sources of evidence, the history of theories, causal intervention, independent retesting and the limits of perspective jointly determine degrees of confidence. Chapter 82 maxim: ‘Reality is independent of science; scientific knowledge can approach reality, but it cannot treat each of its present representations as final reality.’