The role of prediction in science

The ability to make successful predictions is widely regarded as one of the hallmarks of a productive scientific theory. A theory that tells us what we will find before we find it — that specifies in advance what observations, experiments, or discoveries should yield — demonstrates a form of explanatory power that goes beyond merely accommodating known facts. The distinction between prediction and post-hoc accommodation is central to the philosophy of science, and it bears directly on debates over the scientific status of evolution, creationism, and intelligent design.^{2, 1}

Prediction versus accommodation

Philosophers of science distinguish between two ways a theory can relate to evidence. A theory predicts a fact when it specifies that fact in advance, before the fact is known. A theory accommodates a fact when it is adjusted or interpreted after the fact to be consistent with it. Both prediction and accommodation provide some support for a theory, but prediction is generally regarded as providing much stronger support, because a successful prediction rules out the possibility that the theory was simply tailored to fit data that was already known.^{2, 15}

Karl Popper argued that the most impressive feature of a scientific theory is its ability to make “risky” predictions — predictions that are specific enough to be falsified, that go against background expectations, and that would be surprising if the theory were false. When such predictions are confirmed, they provide powerful evidence for the theory, precisely because the confirmation would be improbable under alternative hypotheses. A theory that never makes risky predictions, or that can accommodate any outcome through post-hoc adjustment, is not genuinely testable.²

Imre Lakatos incorporated this insight into his methodology of scientific research programmes. A research programme is “progressive” when it generates novel predictions that are subsequently confirmed — when its theoretical modifications lead to the discovery of new facts rather than merely explaining away old anomalies. A programme is “degenerating” when its modifications are exclusively defensive, designed to accommodate known data without ever anticipating new discoveries. The distinction between progressive and degenerating programmes provides a criterion for evaluating the health and scientific productivity of competing theoretical frameworks.¹

Evolution’s predictive track record

Evolutionary biology has generated a substantial record of successful novel predictions across multiple domains. These predictions were not vague or easily accommodated by alternative theories; they were specific, testable, and in many cases would have been highly improbable if evolution were false.

Tiktaalik and transitional fossils. In 2004, paleontologists Neil Shubin, Edward Daeschler, and Farish Jenkins set out to find a transitional form between lobe-finned fishes and early tetrapods. Evolutionary theory predicted that such a form should exist, that it should date to the late Devonian period (approximately 375 million years ago), and that it should be found in rocks of the appropriate age and depositional environment. Shubin’s team identified late-Devonian freshwater sediments on Ellesmere Island in the Canadian Arctic as a promising site. In 2004, they discovered Tiktaalik roseae, a fish with features intermediate between fish and tetrapods: a flat skull, a mobile neck, ribs capable of supporting the body, and fin bones arranged in a pattern homologous to the limb bones of later tetrapods. The discovery was published in Nature in 2006.³ The prediction was specific (what kind of organism, in what geological period, in what type of rock), and the discovery matched the prediction in every respect. No alternative framework — neither young-earth creationism nor intelligent design — predicted or could have predicted this finding.⁴

Human chromosome 2 fusion. All great apes have 24 pairs of chromosomes; humans have 23 pairs. If humans and other great apes share a common ancestor, evolutionary theory predicts that two ancestral chromosomes must have fused in the human lineage. This fusion should leave a specific signature: a chromosome with telomeric sequences (normally found only at chromosome ends) in its interior, and with two centromeric regions rather than one. In 1991, IJdo and colleagues examined human chromosome 2 and found exactly this signature: telomeric repeats at the fusion point in band 2q13, and a vestigial second centromere. The prediction was precise, the evidence was unambiguous, and the result is inexplicable under any framework that denies common ancestry between humans and other apes.^{5, 6}

Endogenous retroviruses. Retroviruses insert their DNA into the host genome. When such an insertion occurs in a germ-line cell, it becomes heritable and can be passed to all descendants. Common descent predicts that endogenous retroviruses (ERVs) inserted in a common ancestor should be found at the same genomic location in all descendant species, and that the pattern of shared ERVs should match the independently established phylogenetic tree. This prediction has been confirmed across hundreds of ERV insertions in primate genomes. Humans and chimpanzees share numerous ERVs at identical genomic locations, with the pattern of sharing precisely tracking the phylogenetic relationships established by morphological and other molecular data.⁷

Concordance of independent phylogenies. If all life shares common ancestry, then phylogenetic trees constructed from different types of evidence — morphological comparisons, mitochondrial DNA, nuclear DNA, protein sequences, chromosomal rearrangements — should converge on the same branching pattern. This is a risky prediction: there is no reason, apart from common descent, why trees based on the shape of bones should match trees based on the sequence of cytochrome c or the arrangement of Alu elements in the genome. The prediction has been confirmed thousands of times. Molecular phylogenies consistently match morphological phylogenies at the major branching points, with discrepancies typically occurring at fine-grained levels where incomplete lineage sorting or horizontal gene transfer creates genuinely complex signals.^{8, 14}

Retrodiction and explanatory flexibility

The counterpoint to successful prediction is “retrodiction” — the ability to explain facts after they are already known. Retrodiction is not worthless; a theory that can explain known facts is better than one that cannot. But retrodiction is weaker evidence than prediction because a sufficiently flexible framework can accommodate almost any observation after the fact without being constrained by it in advance. This is the distinction between a theory that tells you what to expect and a theory that tells you why you should not have been surprised.^{10, 15}

The problem with extreme explanatory flexibility is that it approaches unfalsifiability. A framework that can explain any possible outcome — that would be consistent with finding ERVs in shared locations or in different locations, with finding transitional fossils or not finding them, with chromosome numbers matching or not matching — does not make predictions at all. It merely narrates. Lakatos argued that this is precisely the distinction between progressive and degenerating programmes: a progressive programme tells you what to look for; a degenerating programme tells you why what you found was to be expected all along.¹

The predictive failure of creationism and intelligent design

Neither young-earth creationism nor intelligent design has produced a successful novel prediction. This is not a minor omission; it is a fundamental indicator of the frameworks’ scientific status. Young-earth creationism, which posits that all life was created in its present form within the last 6,000–10,000 years, is contradicted by virtually every prediction it implies: it predicts the absence of transitional fossils (they exist in abundance), the absence of shared genetic errors across species (they are ubiquitous), and the absence of observable speciation (it has been documented repeatedly). Rather than generating new predictions, young-earth creationism devotes its theoretical resources to explaining away the evidence that contradicts it.¹²

Intelligent design fares no better in predictive terms. Michael Behe’s “irreducible complexity” does not predict which systems will be found to be irreducibly complex, nor does it predict what features a designed system should have that an evolved system should lack. William Dembski’s “specified complexity” framework has not been applied to generate a single novel empirical prediction. In the Kitzmiller v. Dover trial, this absence of predictive content was identified as one of the reasons intelligent design fails to qualify as science.¹³ The framework’s proponents have responded to apparent refutations — such as the identification of evolutionary precursors to the bacterial flagellum — not by revising their predictions but by redefining their terms, a pattern that Lakatos identified as characteristic of degenerating programmes.^{1, 12}

The contrast between evolution and its rivals on the criterion of prediction is stark. Evolution specifies in advance where to find transitional fossils, what genetic signatures common descent should produce, and how independent lines of evidence should converge. These predictions have been confirmed repeatedly. Creationism and intelligent design specify nothing in advance and accommodate everything after the fact. The “explanatory power” of “God did it” or “a designer did it” is maximal in the sense that it is consistent with any possible observation — and for precisely that reason, it provides no evidential support, because a hypothesis that predicts everything predicts nothing.^{10, 2}

Significance

The role of prediction in science illuminates why the demarcation problem matters and why it is not merely an abstract philosophical exercise. The ability to make successful novel predictions is not just a desirable feature of a theory; it is the primary mechanism by which theories earn our confidence. A theory that has repeatedly told us what we will find, and been right, has demonstrated that it has captured something real about the structure of the world. A theory that has never told us what to look for, and can only explain what we have already found, has demonstrated nothing except its own flexibility.^{1, 11}

This distinction has practical consequences for education and public policy. When the predictive track records of evolutionary biology and its rivals are compared directly, the asymmetry is decisive. Evolution is a progressive research programme that continues to generate novel predictions and new discoveries. Creationism and intelligent design are degenerating programmes that produce no predictions and devote their resources to contesting established conclusions. The predictive criterion does not prove that any scientific theory is certainly true — no criterion can do that — but it provides a reliable means of distinguishing frameworks that advance understanding from those that merely resist it.^{1, 10}