Inference

Lore & Background

Inference is the step in logical reasoning that moves from premises to logical consequences. The distinction between deduction and induction dates at least to Aristotle in the 300s BC. Deduction derives logical conclusions from premises known or assumed to be true, with valid inference studied in logic. Induction infers a universal conclusion from particular evidence. A third type, abduction, seeks a 'best explanation' based on likelihood, proposed by Charles Sanders Peirce. Human inference is studied in logic, argumentation studies, and cognitive psychology. Artificial intelligence researchers develop automated inference systems to emulate human inference. Statistical inference uses mathematics to draw conclusions under uncertainty, generalizing deterministic reasoning, and uses quantitative or qualitative data subject to random variations. The definition of inference is disputed. One definition is 'a conclusion reached on the basis of evidence and reasoning'; another is 'the process of reaching such a conclusion.' Inductive reasoning is defined as inferring a general conclusion from multiple observations, but this definition is disputable due to lack of clarity, and applies only when the conclusion is general.

Reader's Guide

Inference is central to logic, reasoning, and artificial intelligence. The article presents classic syllogisms, such as 'All humans are mortal; all Greeks are humans; therefore all Greeks are mortal,' to illustrate valid deductive forms. Validity depends on form, not truth: a valid argument with true premises always yields a true conclusion, but a valid form can have false premises and a false or even true conclusion. Invalid forms, like 'All A are B; all C are B; therefore all C are A,' can lead from true premises to false conclusions. Incorrect inferences are called fallacies. Cognitive psychologists document biases such as confirmation bias (seeking confirming information) and the conjunction fallacy (judging a conjunction more probable than a single conjunct), demonstrated by the Wason selection task and the Linda problem. These biases are explained by heuristics. In AI, inference engines extend a knowledge base automatically using valid inferences. Prolog, a programming language based on predicate calculus, uses backward chaining to check whether a proposition can be inferred. For example, from 'mortal(X) if man(X)' and 'man(socrates),' Prolog answers 'Yes' to 'mortal(socrates).' The term 'inference' also applies to generating predictions from trained neural networks, used in image recognition and natural language processing.

Did You Know?

• Inference is traditionally divided into deduction and induction, a distinction dating at least to Aristotle in the 300s BC.
• A third type of inference, abduction, was proposed by Charles Sanders Peirce and seeks a 'best explanation' based on likelihood.
• An incorrect inference is known as a fallacy; cognitive psychologists have documented biases such as confirmation bias and the conjunction fallacy.
• Prolog, a programming language based on predicate calculus, uses backward chaining to infer whether a proposition follows from a knowledge base.

The Architecture of Valid Inference

Logic, at its core, concerns itself with the study of valid inference—the structured movement from premises to a conclusion. An inference earns the label of validity when it is logically impossible for its premises to hold true while its conclusion falls false. In other words, the truth of the starting points guarantees the truth of what follows. This structural guarantee is typically codified through rules of inference: an argument is valid when its form, the particular way premises and conclusion are arranged, conforms to an accepted rule. Classical logic has long served as the dominant framework for articulating these rules, codifying principles that many find intuitively compelling, such as the law of excluded middle, the elimination of double negation, and the bivalence of truth into simply true or false. Yet the landscape of inference is far richer than this single tradition. Different logical systems propose different accounts of what makes an inference valid, and philosophical logic exists precisely to explore, compare, and evaluate these alternative frameworks. The question of how to classify the many non-classical systems that have emerged, many of them only in recent decades, remains one of the field's central organizing challenges.

Extending the Classical Framework

One major strand of philosophical logic holds that classical logic, while sound, is simply too narrow to capture the full range of philosophically significant reasoning. The response is to build extended logics: systems that retain classical inference as their foundation but introduce new symbols together with the rules governing them. Alethic modal logic, for instance, adds operators for possibility and necessity, allowing one to distinguish what is true simpliciter from what is true in some possible world or in every possible world—a move often paired with possible-worlds semantics. Deontic logic brings the apparatus of obligation and permission into formal treatment, giving ethics a precise inferential structure. Temporal logic captures relations such as whether a proposition holds at a particular moment, at all times, in the past, or in the future. Epistemic logic extends the framework to mental states, expressing not merely what is the case but what an agent believes or knows, and articulating what follows from those states. Higher-order logics take a different route: rather than applying classical logic to a new domain, they generalize it by permitting quantification over predicates themselves, not just over individuals. Each extension preserves the classical core while opening a new philosophical territory to rigorous analysis.

Rival Logics and the Challenge to Classical Principles

A second, more radical strand treats classical logic not as a foundation to be extended but as a system containing genuine flaws. Deviant logics reject one or more of its core principles and position themselves as rivals. Intuitionistic logic, for example, grounds truth in the availability of a proof; because not every classically valid rule is compatible with this verification-based conception, it discards principles such as the law of excluded middle. Free logic adjusts the treatment of singular terms—names and definite descriptions that may refer to nothing—so that using such terms no longer smuggles in an existential presupposition. Many-valued logics expand the truth-value space beyond the binary true/false, thereby rejecting bivalence outright. Paraconsistent logics confront contradictions head-on by blocking the principle of explosion, which in classical logic lets any conclusion follow from a contradiction; relevance logic is a prominent example, insisting that for a conditional to hold, its antecedent must be genuinely relevant to its consequent rather than merely truth-functionally linked. Together, these systems demonstrate that the space of coherent inference is far wider than the classical tradition alone suggests.

Philosophical Logic as a Discipline

The term philosophical logic carries at least two distinct readings, and the distinction shapes how the field is organized. In its narrow sense, philosophical logic is the branch of philosophy that applies logical methods to philosophical problems, typically by constructing or analyzing non-classical logical systems. Under this reading, concepts like possibility, necessity, obligation, permission, time, and knowledge are given a logically precise treatment: their inferential roles relative to one another are made explicit through formal rules. In this narrower conception, philosophical logic sits as one area within the broader discipline of the philosophy of logic. Some theorists, however, use the term in a wider sense to cover the entire study of logic's scope and nature—questions about how logic itself should be defined, what its fundamental concepts are, and what problems logic raises for philosophy. On that view, philosophical logic and the philosophy of logic are essentially the same enterprise. The narrow reading, which treats philosophical logic as a subfield, is the one most commonly adopted in specialized literature, and it is the perspective from which the rich taxonomy of extended and deviant logics is best appreciated.

Gallery