Every emerald you've ever seen has been green. Every emerald anyone has ever observed has been green. By the standard logic of induction, we can confidently predict that the next emerald examined will also be green. This seems like a textbook case of good scientific reasoning.

But in 1955, philosopher Nelson Goodman showed something unsettling. Using the same inductive logic, we can equally justify predicting the next emerald will be blue. His argument doesn't rely on trickery or hidden premises. It reveals a genuine puzzle at the heart of how we reason from evidence to prediction, one that still troubles philosophers of science today.

Goodman invented a strange predicate: grue. Something is grue if it is examined before a certain future date (say, 2025) and found to be green, or if it is examined afterward and found to be blue. Every emerald observed so far has been green, and each was observed before 2025. So every observed emerald is not only green, but also grue.

Here's the trouble. Standard induction says that if all observed instances of a kind share a property, we should expect future instances to share it too. All observed emeralds are green, so future emeralds should be green. But all observed emeralds are also grue, so by identical reasoning, future emeralds should be grue, meaning they should look blue after 2025.

The evidence supports both predictions equally. Nothing in the logical form of induction tells us which projection to trust. The rules we thought were purely formal turn out to depend on something extra: which predicates we allow into our reasoning in the first place.

Takeaway

Inductive logic alone cannot tell us which patterns in past evidence will continue into the future. The choice of concepts we use to describe evidence quietly does much of the work we credit to logic.

Goodman's own proposal was linguistic and historical. Some predicates, like green, have a long track record of successful use in prediction. Scientists and ordinary speakers have projected them onto new cases countless times, and those projections have generally worked out. Goodman called this entrenchment.

Predicates like grue are newcomers. Nobody has ever built successful predictions on them. When two competing predicates are equally supported by evidence, Goodman suggested we should prefer the more entrenched one, simply because our practice of using it has proved reliable over time.

This solution is pragmatic rather than metaphysical. It doesn't claim green is somehow more real than grue. It claims only that our inductive habits have crystallized around certain terms, and departing from those habits without reason would be reckless. Critics find this unsatisfying, though. Why should the mere history of our word choices constrain what nature is actually like? Entrenchment describes our practice but may not justify it.

Takeaway

Tradition can be a rational guide when other guides fail. The concepts that have served prediction well earn a kind of trust, even if we cannot fully explain why they work.

A deeper response comes from metaphysics. Perhaps green picks out a genuine feature of the world, while grue is a gerrymandered construction, cobbled together from disconnected pieces. On this view, nature itself has joints, and good scientific concepts carve along those joints rather than across them.

Philosopher David Lewis called these natural properties. Being green corresponds to a unified physical fact about how surfaces reflect light. Being grue corresponds to no such unified fact. It requires reference to a specific date and a disjunction of colors that share nothing physically important.

If nature contains real distinctions between natural and artificial predicates, then induction can be constrained by the world rather than by mere convention. Scientific realism, the view that science aims to describe reality's actual structure, fits comfortably here. Good theories succeed because their key terms latch onto genuine features of the world. This solution rescues induction, but at the cost of taking on serious metaphysical commitments about what exists independently of our minds.

Takeaway

The world may not be equally divisible in every direction. Science works partly because reality itself favors certain descriptions over others, guiding us toward the categories that actually matter.

The grue paradox looks like a linguistic trick, but it exposes something real. Induction, the workhorse of scientific reasoning, needs more than logic to function. It needs assumptions about which concepts track the world's genuine structure.

Whether that structure comes from entrenched practice or from nature's own joints remains contested. Either way, Goodman's puzzle reminds us that even our most confident predictions rest on quiet philosophical foundations, well worth examining.