Scientific realism — the view that our best empirically successful theories are at least approximately true — faces Kyle Stanford's 'New Induction': past scientific communities repeatedly failed to conceive the serious rival theories that later displaced their own, so our best contemporary theories probably have unconceived rivals too. Dellsén (Synthese, 2017) replied with a frequency-based analysis in which the probability that a theory lacks rivals, given that none has been discovered, depends on PLAUSIBILITY (the base frequency of theories genuinely lacking rivals in a reference class) and on SENSITIVITY (the frequency with which existing rivals are discovered), governed by four social-historical indicators: domain conceptual accessibility, community anti-dogmatism, community size and competence, and time-in-play. Alai (Axiomathes, 2019) replied that theoretical virtues such as explanatory power and simplicity dissolve most historical underdetermination. Dellsén explicitly leaves the empirical question open: he never estimates PLAUSIBILITY or SENSITIVITY, he declines to provide any algorithm for aggregating his four indicators — which serve only as guides to comparative rather than absolute degrees of sensitivity — into an overall estimate, and he therefore leaves how prevalent residual underdetermination actually is across scientific subfields, and whether the governing factors can be measured at all, unquantified. The later literature — a 2019 Synthese special issue on unconceived alternatives, Alai's own 2021 follow-up, and case studies in biology, cognitive neuroscience, gravitational-wave astronomy, and chemistry — treats the question conceptually or locally; no work operationalizes the indicators or measures the distribution. The problem asks for that measurement program: an explicit, reproducible measurement of the frequencies that govern warranted confidence in the absence of rivals, applied to a systematic cross-subfield sample of historical and contemporary theories, yielding an assessment of residual underdetermination risk and its distribution across the sciences.

Background

Scientific realism is the view that our best empirically successful scientific theories are at least approximately true, including their claims about unobservable entities such as electrons or genes. Its most influential recent challenge starts from underdetermination — the thesis that the available evidence fails to single out one theory as uniquely warranted. Philosophers distinguish holist underdetermination (Duhem–Quine holism: any theory can be protected from refutation by adjusting auxiliary assumptions) from contrastive underdetermination (the existence of distinct theories that would be empirically equivalent on all possible evidence). The historical form at issue here is Kyle Stanford's problem of unconceived alternatives, also called recurrent transient underdetermination: his 'New Induction' over the history of science argues that past scientific communities repeatedly failed even to conceive of the serious rival theories that would eventually displace their own — for example, nineteenth-century theories of inheritance before Mendelian genetics — and infers that our own best theories likewise probably have unconceived rivals. 'Transient' means the rivals were eventually conceived, ending the underdetermination; 'recurrent' means the pattern repeated across eras and fields. A standard realist reply appeals to theoretical virtues — non-empirical merits such as explanatory power, simplicity, consilience (cohesion with independently accepted background knowledge), and fecundity in novel predictions — as grounds for preferring one empirically equivalent theory over another. Mario Alai (Axiomathes, 2019) defends this route, arguing that historical cases of apparent empirical equivalence were almost always resolved either by later empirical tests or by virtue-based selection, so that underdetermination is for the most part provisional and tolerable.

Finnur Dellsén (Synthese, 2017) offered a different, probabilistic defense. Adapting the slogan 'absence of evidence is not evidence of absence', he asks when the historical absence of discovered rivals to a theory licenses confidence that no serious rivals exist. Using a frequency interpretation over a reference class of theories (a set of past or present theories judged inductively similar), his analysis derives that the conditional probability that a theory has no rivals, given that none have been found, depends on just two aggregate quantities: PLAUSIBILITY, the frequency of theories in the class that genuinely lack underdetermination rivals, and SENSITIVITY, the frequency with which theories in the class that do have rivals have had those rivals discovered. The conditional probability rises with SENSITIVITY, so a theory's insulation from the New Induction depends on features of its domain and its community's history that make rivals discoverable. Dellsén proposes four ceteris-paribus (other-things-equal) indicators of SENSITIVITY: (1) domain conceptual accessibility — highly abstract, counterintuitive, or mathematically complex domains make rival theories harder to conceive; (2) anti-dogmatism — community norms of organized skepticism that encourage entertaining alternatives; (3) community size and competence — larger, more professionalized, better-trained communities generate and recognize more rivals; (4) time-in-play — the longer a theory has been accepted, the more opportunities there have been to find its rivals. He illustrates with Fresnel's ether theory versus the photon theory of light and nineteenth-century heredity theories versus Mendelian genetics, and he acknowledges exceptions — notably Newtonian mechanics, whose long time-in-play did not prevent it from being later underdetermined.

Dellsén's paper then explicitly leaves an empirical research program undone. He assesses the indicators only qualitatively and comparatively; he supplies no algorithm for combining them into a numerical SENSITIVITY estimate — indeed he states that he does not believe any such agglomeration procedure could reasonably be provided — notes that operationalizing them in specific historical cases requires contested historical and sociological judgments, and leaves PLAUSIBILITY and SENSITIVITY as unestimated frequencies, so the residual risk of underdetermination and its distribution across scientific subfields remain unquantified. Alai, by contrast, defers no empirical question: he resists Stanford's recurrence argument conceptually, asserting without measurement that the number of unconceived but epistemically reasonable theories today is smaller than in the past, concedes only that we lack an absolute measure of the probability that our best theories are false, and defers just the discussion of Stanford's specific version of underdetermination to a subsequent paper (taken up conceptually in Alai 2021); his paper does not cite Dellsén. Later work has sharpened rather than settled this: Stanford himself has argued that professionalization, peer review, and 'Big Science' make contemporary scientific communities more theoretically conservative and hence more vulnerable to unconceived alternatives — a prediction of the opposite sign for the very sociological indicators Dellsén cites. Whether residual underdetermination is in fact rare or widespread across the sciences, and whether the governing social-historical factors can be measured well enough to tell, is therefore the empirical hinge of the whole debate.

Problem Statement

The problem is to carry out the empirical-historical assessment of residual theoretical underdetermination that Dellsén (Synthese, 2017) explicitly leaves undone. Its single overarching objective is to determine, by explicit and reproducible measurement rather than qualitative illustration, how prevalent residual underdetermination is across the sciences and how its risk is distributed across scientific subfields. The measurement program therefore has to (i) operationalize the quantities on which Dellsén's frequency analysis makes warranted confidence in the absence of rivals depend — PLAUSIBILITY (the frequency of theories in a reference class that genuinely lack serious rivals) and SENSITIVITY (the frequency with which rivals that exist have been discovered) — into measures that can be coded or computed for actual historical and contemporary cases; (ii) apply the operationalized measures to a systematic, non-cherry-picked sample of theories spanning scientific subfields, with explicit criteria for what counts as a serious rival and for distinguishing rivals that were genuinely unconceived from those that were considered and rejected; and (iii) aggregate or comparatively analyze the results into an assessment of the residual underdetermination risk facing theories in different subfields. The question 'can it be measured?' is part of the problem: the resolution has to show, on concrete cases, whether these quantities or indicators, or any equivalent, can be operationalized coherently given the surviving historical record, or establish by principled demonstration that they cannot be (for example because unconceived rivals by their nature leave no recoverable trace). The distribution that results would be the scientific-realism debate's first empirical one, telling us which subfields' best-accepted theories face high residual underdetermination risk and which are epistemically insulated.

References

  1. Dellsén, Finnur (2017). 'Realism and the absence of rivals.' Synthese 194(7): 2427–2446. doi:10.1007/s11229-016-1059-3. https://doi.org/10.1007/s11229-016-1059-3
  2. Alai, Mario (2019). 'The Underdetermination of Theories and Scientific Realism.' Axiomathes 29(6): 621–637. doi:10.1007/s10516-018-9384-4. https://doi.org/10.1007/s10516-018-9384-4
  3. Stanford, P. Kyle (2006). Exceeding Our Grasp: Science, History, and the Problem of Unconceived Alternatives. New York: Oxford University Press. doi:10.1093/0195174089.001.0001. https://doi.org/10.1093/0195174089.001.0001
  4. Stanford, P. Kyle (2019). 'Unconceived alternatives and conservatism in science: the impact of professionalization, peer-review, and Big Science.' Synthese 196(10): 3915–3932 (first published online 2015). doi:10.1007/s11229-015-0856-4. https://doi.org/10.1007/s11229-015-0856-4
  5. Bhakthavatsalam, Sindhuja and Kidd, Ian James (2019). 'Science, realism, and unconceived alternatives: introduction to the special issue on unconceived alternatives.' Synthese 196(10): 3911–3913. doi:10.1007/s11229-019-02353-z. https://doi.org/10.1007/s11229-019-02353-z
  6. Alai, Mario (2021). 'Scientific Realism and Further Underdetermination Challenges.' Axiomathes 31(6): 779–789. doi:10.1007/s10516-021-09542-x. https://doi.org/10.1007/s10516-021-09542-x
  7. Dellsén, Finnur (2019). 'Should scientific realists embrace theoretical conservatism?' Studies in History and Philosophy of Science Part A 76: 30–38. doi:10.1016/j.shpsa.2018.09.005. https://doi.org/10.1016/j.shpsa.2018.09.005
  8. Devitt, Michael (2011). 'Are Unconceived Alternatives a Problem for Scientific Realism?' Journal for General Philosophy of Science 42(2): 285–293. doi:10.1007/s10838-011-9166-9. https://doi.org/10.1007/s10838-011-9166-9
  9. Godfrey-Smith, Peter (2008). 'Recurrent transient underdetermination and the glass half full.' Philosophical Studies 137(1): 141–148. doi:10.1007/s11098-007-9172-2. https://doi.org/10.1007/s11098-007-9172-2
  10. Chakravartty, Anjan (2008). 'What you don't know can't hurt you: realism and the unconceived.' Philosophical Studies 137(1): 149–158. doi:10.1007/s11098-007-9173-1. https://doi.org/10.1007/s11098-007-9173-1
  11. Ruhmkorff, Samuel (2019). 'Unconceived alternatives and the cathedral problem.' Synthese 196(10): 3933–3945 (first published online 2015). doi:10.1007/s11229-015-0947-2. https://doi.org/10.1007/s11229-015-0947-2
  12. Rowbottom, Darrell P. (2019). 'Extending the argument from unconceived alternatives: observations, models, predictions, explanations, methods, instruments, experiments, and values.' Synthese 196(10): 3947–3959 (first published online 2016). doi:10.1007/s11229-016-1132-y. https://doi.org/10.1007/s11229-016-1132-y
  13. Curtis-Trudel, André, Rowbottom, Darrell P., and Li, Tjonnie G. F. (2025). 'Computational science and unconceived alternatives: lessons from, and for, gravitational-wave astronomy.' Synthese 206(6), article 297 (published online 2025). doi:10.1007/s11229-025-05362-3. https://doi.org/10.1007/s11229-025-05362-3
  14. Grünbaum, Thor (2021). 'The two visual systems hypothesis and contrastive underdetermination.' Synthese 198(S17): 4045–4068 (first published online 2018). doi:10.1007/s11229-018-01984-y. https://doi.org/10.1007/s11229-018-01984-y
  15. Kovaka, Karen and Novick, Rose (2024). 'Relative Frequency Controversies and the Growth of Biological Knowledge.' Philosophy, Theory, and Practice in Biology 16(1). doi:10.3998/ptpbio.5562. https://doi.org/10.3998/ptpbio.5562
  16. Spelda, Petr and Stritecky, Vit (2021). 'What Can Artificial Intelligence Do for Scientific Realism?' Axiomathes 31(1): 85–104 (first published online 2020). doi:10.1007/s10516-020-09480-0. https://doi.org/10.1007/s10516-020-09480-0
  17. Belkheiri, Nadji (2026). 'Unconceived alternatives and mechanistic confirmation in chemistry: toward a polysystemic meta-empirical framework.' Foundations of Chemistry (published online 2026). doi:10.1007/s10698-026-09581-5. https://doi.org/10.1007/s10698-026-09581-5
  18. Jellema, Hylke (2024). 'Reasonable Doubt from Unconceived Alternatives.' Erkenntnis 89(3): 971–996 (first published online 2022). doi:10.1007/s10670-022-00565-3. https://doi.org/10.1007/s10670-022-00565-3
  19. Turnbull, Margaret Greta (2018). 'Underdetermination in science: What it is and why we should care.' Philosophy Compass 13(2): e12475. doi:10.1111/phc3.12475. https://doi.org/10.1111/phc3.12475
  20. Stanford, Kyle (2023). 'Underdetermination of Scientific Theory.' Stanford Encyclopedia of Philosophy (first published 2009; substantive revision April 4, 2023; no DOI — online reference work). https://plato.stanford.edu/entries/scientific-underdetermination/