Why every green emerald also proves emeralds are "grue"
Imagine the predicate "grue": an object is grue if observed before a future date and green, or unobserved and blue. Every green emerald ever unearthed perfectly confirms that emeralds are grue. Yet no one expects an unmined emerald to suddenly turn out blue. Philosopher Nelson Goodman introduced this riddle in 1955 to demonstrate that pure logic alone cannot tell us which patterns are valid to project into the future.
Hume and the Old Riddle of Induction
In the eighteenth century, David Hume identified a fundamental puzzle at the heart of human knowledge: the problem of induction. Whenever we infer that the future will resemble the past—such as expecting the sun to rise tomorrow or assuming that bread will nourish us because it always has—we rely on an inductive leap. Hume observed that this leap cannot be justified deductively, because there is no logical contradiction in imagining that nature's laws might abruptly change. Nor can it be justified inductively without arguing in a circle, since appealing to past inductive successes already assumes that the future follows past patterns.
For generations, philosophers treated Hume's challenge as a question of justification: how can we prove that inductive reasoning is rational? By the mid-twentieth century, logical positivists and philosophers of science like Carl Hempel and Rudolf Carnap shifted focus from justifying induction to formalizing it. They attempted to construct precise logical rules for confirmation, aiming to show syntactically how an observation report—such as noting a single black raven—provides objective, degree-measurable support for a universal hypothesis like 'all ravens are black.'
Nelson Goodman Introduces Grue
In his 1955 book 'Fact, Fiction, and Forecast', American philosopher Nelson Goodman demonstrated that the project of purely formal confirmation theory was broken from the start. He argued that the real difficulty with induction was not justifying our belief in the uniformity of nature, but defining which patterns count as regularities in the first place. To illustrate this, Goodman introduced what he termed the 'New Riddle of Induction' using a deliberately constructed predicate: 'grue'.
Goodman defined the predicate 'grue' relative to an arbitrary future time or date, often denoted as 't'. An object is grue if and only if it is observed before time t and is green, or is not observed before time t and is blue. To make the symmetry complete, he introduced a companion predicate, 'bleen', which applies to objects observed before time t if they are blue, and to objects not observed before time t if they are green.
Now suppose we examine a vast number of emeralds before time t, and find every single one of them to be green. Under standard inductive logic, these observations confirm the general hypothesis: 'All emeralds are green.' However, because every examined emerald was observed before time t and was green, every examined emerald also satisfies the definition of grue. Therefore, the exact same empirical evidence confirms with equal logical force the competing hypothesis: 'All emeralds are grue.'
The Challenge to Syntactic Logic
The conclusion of this inductive symmetry is bizarre. If the hypothesis 'All emeralds are grue' is confirmed, it leads to the expectation that any emerald examined for the first time after time t will be blue, not green. Yet no reasonable observer expects an unmined emerald to turn out blue tomorrow simply because the clock crossed an arbitrary threshold. The paradox is that standard formal logic cannot explain why one hypothesis is confirmed while the other is not.
The most immediate response to Goodman's riddle is to object that 'grue' is an artificial, qualitative impostor. Critics often argue that 'grue' is inherently disjunctive and temporal—meaning its definition relies on an 'or' condition and a reference to a specific date—whereas 'green' is a pure, timeless, qualitative color term. It seems intuitive that scientific laws should be formulated only with simple, non-temporal predicates.
Goodman showed that this defense fails because simplicity and disjunction are relative to the language one starts with. If a linguistic community took 'grue' and 'bleen' as their basic, unanalyzed color primitives, they would define 'green' as: 'grue if examined before time t, and bleen otherwise.' To speakers of that language, 'green' would appear hopelessly disjunctive, complex, and time-dependent, while 'grue' would be perceived as simple and fundamental. Pure syntax cannot privilege one vocabulary over the other without arbitrarily begging the question.
Lawlike Generalizations vs Accidental Truths
The deeper consequence of the grue paradox is that it exposes a sharp dividing line between 'lawlike' statements and purely accidental generalizations. In classical confirmation theory, such as the criterion formulated by Jean Nicod, an observation of an instance of an A that is B automatically confirms the general proposition 'All A are B.' Goodman proved that this principle is false as a universal rule.
Consider the statement 'All men in this room are third sons.' Even if every man currently in the room happens to be a third son, examining ninety-nine of them does not increase the likelihood that the hundredth man in the room is also a third son. The statement is accidental rather than lawlike, so its observed instances do not project onto unobserved cases. In contrast, testing samples of copper for electrical conductivity does confirm that unexamined copper conducts electricity.
Goodman showed that 'All emeralds are green' behaves like a lawlike hypothesis whose instances confirm it, whereas 'All emeralds are grue' behaves like an accidental generalization. The core problem is that whether a statement is lawlike cannot be determined simply by looking at its grammatical form or logical structure. Both hypotheses share the exact same logical form: 'All x of type E are G.'
Goodman's Solution: Entrenchment
To resolve the riddle, Goodman proposed his theory of 'entrenchment.' Rather than searching for an elusive metaphysical or formal difference between green and grue, Goodman looked to the historical track record of linguistic practice. A predicate becomes entrenched in a language through a long history of actual, successful past projections.
The predicate 'green' is highly entrenched because English speakers have projected it into the future countless times across centuries with consistent predictive utility. The predicate 'grue', by contrast, has virtually no history of past projection; it was introduced artificially. When two hypotheses are equally well supported by the current empirical data but make conflicting predictions, Goodman argued that the hypothesis formulated with the better-entrenched predicate rightfully overrides the unentrenched competitor.
Goodman acknowledged that entrenchment is an pragmatic and historical standard rather than an absolute, timeless logical guarantee. It ties the validity of our inductive inferences directly to the linguistic and scientific habits of our community. While some philosophers found this historical grounding unsatisfyingly conventional, Goodman maintained that inductive validity is fundamentally a matter of bringing our general inferential rules into harmony with our actual accepted inferential practices.
Wider Impact on Epistemology and Science
The New Riddle of Induction reshaped twentieth-century philosophy of science and epistemology. It forced theorists to abandon the hope of constructing a purely formal, content-free logic of induction comparable to deductive logic. Deductive validity depends solely on syntax: if all premises of the form 'All A are B' and 'x is A' are true, 'x is B' follows necessarily, regardless of what words replace A and B. Goodman proved that induction is inherently content-sensitive.
Goodman's riddle also stimulated extensive research into the nature of 'natural kinds' in metaphysics and science. Philosophers such as W.V.O. Quine investigated why certain categories carve nature at its joints while others do not, linking projectability to evolutionary history, similarity spaces, and causal structures.
In modern computational learning theory and statistics, the problem reemerges under the guise of model selection, curve fitting, and inductive bias. For any finite collection of data points, an infinite number of mutually incompatible curves can fit the data perfectly. Deciding which curve represents genuine signal rather than artificial 'grue-like' noise requires background assumptions that formal data alone can never provide.
Key takeaways
•Nelson Goodman's 'New Riddle of Induction' showed that identical empirical evidence can equally support contradictory hypotheses, such as 'All emeralds are green' and 'All emeralds are grue.'
•Purely syntactic and formal logical rules cannot distinguish between valid, projectable hypotheses and artificial regularities without relying on external linguistic or background assumptions.
•Goodman proposed that we favor predicates like 'green' over 'grue' because they possess 'entrenchment'—a history of successful past usage and projection in our linguistic community.
•The grue paradox demonstrated that induction is fundamentally content-dependent, influencing modern theories of natural kinds, confirmation, and machine learning bias.