Inference to best explanation

When a detective surveys a crime scene and concludes that the butler committed the murder, or when a physician observes a cluster of symptoms and diagnoses pneumonia, or when a palaeontologist discovers a fossilised bone and infers the existence of a previously unknown species, each is employing a pattern of reasoning that philosophers call inference to the best explanation (IBE). Also known as abduction — a term introduced by the American philosopher Charles Sanders Peirce in the late nineteenth century — IBE is the reasoning process in which one considers a range of competing hypotheses, evaluates which would best explain the observed evidence, and infers that the best explanation is probably true.^{2, 3} This pattern of reasoning is pervasive in science, everyday cognition, and philosophical argument, including debates about the existence of God.¹

The structure of IBE

The canonical formulation of inference to the best explanation was articulated by Gilbert Harman in 1965. Harman argued that much of our reasoning proceeds by identifying the hypothesis that would, if true, provide the best explanation of the evidence, and then inferring that this hypothesis is true (or probably true). The pattern can be schematized as follows: (1) some phenomenon E is observed; (2) hypothesis H, if true, would explain E; (3) no available alternative hypothesis explains E as well as H does; therefore (4) H is probably true.² This schema is distinguished from deduction (in which the conclusion follows necessarily from the premises) and from simple enumerative induction (in which one generalises from observed instances to unobserved ones). IBE is ampliative — the conclusion goes beyond what is logically entailed by the premises — and fallible, since the best available explanation may nevertheless be false.^{1, 2}

Peirce, whose work on abduction predated Harman's by several decades, conceived of abduction as the creative process of generating explanatory hypotheses rather than the evaluative process of selecting among them. In Peirce's framework, abduction generates a hypothesis, deduction derives testable predictions from it, and induction tests those predictions against further evidence.³ Harman's IBE collapsed the generation and evaluation stages into a single inferential process, and it is Harman's formulation that has dominated subsequent philosophical discussion, particularly through the influential work of Peter Lipton.^{1, 2}

Lipton's framework: loveliness and likeliness

Peter Lipton's Inference to the Best Explanation, first published in 1991 and substantially revised in 2004, is the most comprehensive philosophical treatment of IBE. Lipton drew a crucial distinction between what he called likeliness and loveliness. A likely explanation is one that is probable given the evidence; a lovely explanation is one that, if true, would provide the deepest, most illuminating understanding of the phenomena. The central question for IBE, Lipton argued, is whether we infer to the likeliest explanation (the one most probably true) or to the loveliest (the one that would provide the most understanding). He defended the latter: we use explanatory loveliness as a guide to likeliness, inferring that the hypothesis providing the deepest understanding is probably true.¹

Lipton identified several criteria that make an explanation lovely: explanatory power (how much of the evidence does it explain?), unification (does it bring together previously disparate phenomena under a single framework?), simplicity (does it posit fewer ad hoc mechanisms or entities?), fertility (does it suggest new predictions or avenues of inquiry?), and fit with background knowledge (is it consistent with what is already well established?).^{1, 7} No single criterion is decisive; the overall loveliness of an explanation is a holistic judgement that weighs multiple considerations. This plurality of criteria explains why IBE can feel both powerful and elusive: powerful because it captures how scientists and ordinary reasoners actually evaluate hypotheses, elusive because the criteria resist precise formalisation.¹

IBE and Bayesian confirmation

A significant question in the philosophy of science is how IBE relates to Bayesian confirmation theory, which evaluates hypotheses by computing their posterior probability using Bayes' theorem. Some philosophers, including van Fraassen, have argued that IBE and Bayesianism are rivals and that IBE is the less reliable of the two, since inferring truth from explanatory virtue involves an unjustified assumption that the world is structured to reward human aesthetic preferences.^{5, 10}

Others, including Lipton himself, have argued that IBE and Bayesianism are complementary rather than competing. On this view, explanatory considerations serve as guides to the assignment of prior probabilities and likelihoods in the Bayesian framework: hypotheses that are simpler, more unifying, and more explanatorily powerful receive higher prior probabilities because past experience has shown that such hypotheses tend to be true. IBE thus provides the heuristic that makes Bayesianism psychologically and practically implementable, since raw Bayesianism gives no guidance on where prior probabilities come from.^{1, 16} Lipton argued that the two frameworks are "friendly" rather than hostile: IBE identifies which explanations are lovely, and Bayesianism provides the formal apparatus for updating beliefs in light of evidence. The best explanation is typically also the one with the highest posterior probability, not because loveliness is a primitive probability-raiser but because the same features that make a hypothesis explanatorily attractive — simplicity, coherence, predictive success — also tend to make it probable.^{1, 7}

IBE in philosophy of religion

Inference to the best explanation has been deployed extensively in philosophy of religion, both by theists arguing that God provides the best explanation of certain features of the world and by atheists arguing that naturalistic explanations are superior. The fine-tuning argument, the cosmological argument, the argument from consciousness, and the moral argument have all been framed, at least in part, as inferences to the best explanation.^{4, 12}

Richard Swinburne's cumulative case for God's existence is perhaps the most systematic application of IBE in natural theology. Swinburne argues that the existence of God provides the simplest and most explanatorily powerful hypothesis to account for a range of phenomena: the existence of the universe, the law-governed order of nature, the fine-tuning of physical constants for life, the existence of conscious beings, moral awareness, the occurrence of religious experience, and the historical evidence for miracles. He contends that theism unifies these diverse phenomena under a single hypothesis — that a personal God of infinite power, knowledge, and goodness created and sustains the world — in a way that naturalism, which must treat each phenomenon as a separate brute fact, does not.^{4, 13}

Swinburne frames his argument in explicitly Bayesian terms, but the underlying logic is that of IBE: God is posited as the best explanation of the total evidence, and the inference is that this explanation is probably true. He argues that theism is a simple hypothesis (because it posits a single entity with a small number of properties) that has great explanatory scope (because it accounts for a wide range of otherwise puzzling features of the world), making it lovely in Lipton's sense and therefore likely to be true.⁴

Criticisms of theistic IBE

Critics have raised several objections to the use of IBE in support of theism. J. L. Mackie argued that even if theism has explanatory power, it is not a simple hypothesis in any meaningful sense: positing a being with infinite power, knowledge, and goodness raises more questions than it answers, including why such a being would create a world containing evil and suffering.⁸ Elliott Sober has contended that the design argument, even when formulated as an IBE, is weaker than it appears because we have no way of assessing the likelihood that a designer would produce the observed world — a probability that depends on the designer's goals and methods, which are unknown.¹¹

Bas van Fraassen's constructive empiricism poses a more fundamental challenge. Van Fraassen argued that IBE is unreliable as a guide to truth because there is no reason to suppose that the true explanation of any phenomenon is among the hypotheses we have considered; the best of a bad lot may still be false. He also questioned whether explanatory virtues like simplicity and elegance are truth-tracking rather than merely pragmatically useful features of theories.^{5, 10} If this criticism is sound, it undermines not only theistic applications of IBE but all applications, including scientific ones — a consequence that most philosophers of science find too revisionary to accept.^{1, 9}

From the naturalist side, Daniel Dennett and others have argued that the track record of supernatural explanations is a relevant consideration in IBE. Throughout the history of science, naturalistic explanations have consistently displaced supernatural ones — lightning is explained by electrical discharge rather than divine wrath, disease by pathogens rather than demonic possession, biodiversity by evolution by natural selection rather than special creation. This inductive track record, Dennett argues, should lower the prior probability assigned to supernatural hypotheses and raise that of naturalistic ones in any application of IBE.¹⁵ Alvin Plantinga has countered that if human cognitive faculties were designed by God to function properly, then reliable inference — including IBE — is better grounded in a theistic framework than in a naturalistic one, an argument that itself invokes IBE at the meta-level.¹⁴

Ongoing significance

Inference to the best explanation remains one of the central tools of reasoning in both science and philosophy of religion. Its application to religious questions highlights a fundamental tension: the criteria for explanatory goodness — simplicity, scope, coherence, testability, and fertility — are widely accepted in scientific contexts, but their application to the existence of God is contested because theological hypotheses resist the kind of empirical testing that distinguishes scientific IBE from philosophical speculation.^{1, 6, 12} Whether the God hypothesis counts as a genuine explanation or merely as a label for what is unexplained — and whether IBE can operate in domains where crucial hypotheses cannot be tested — remain open and vigorously debated questions at the intersection of epistemology, philosophy of science, and the philosophy of religion.^{1, 4, 17}