What Makes a Thing a Thing?

We usually imagine that a thing comes first and its relationships are added later. First there is a cup, complete in itself; then it happens to meet water and hands. But notice how little the word “cup” tells us without those meetings. A perfectly cup-shaped object that dissolves in water, cannot be lifted, and passes through tables would not play the same role. Its outline might survive, while much of what made it a cup would disappear.

This does not mean the ceramic is unreal. It means that material and relation are not competitors. The ceramic matters because of what it lets the object do: resist pressure, hold heat, chip under impact, carry glaze, meet skin in a certain way. Even a property that sounds private—its hardness—is learned through possible interactions. Hard relative to what pressure? Hot compared with what hand? Heavy for which lifter?

Start, then, with a modest claim. Relationships are not mere decoration around an already understood object. They are part of the evidence by which we recognize it, predict it, and decide what counts as the same kind of thing. This is an epistemic claim: a claim about how knowledge reaches us. It is strong enough to change how we look, but it does not yet tell us what reality is made of.

The useful move is to replace one vague question—“What is it really?”—with a family of testable ones. What can enter it? What can leave? What changes it? What does it change? What happens when two interactions are chained? A thing begins to acquire a relational fingerprint: not a list of neighbors, but an organized pattern of possible encounters.

Science often works this way. We did not first see an electron’s private interior and then predict its behavior. We learned a stable pattern across cloud chambers, electric fields, magnetic fields, detectors, and collisions. The pattern became precise enough that different laboratories could coordinate their results and build devices that depended on it.

A single probe is weak. Many unrelated probes are better. But the deepest evidence comes when the probes constrain one another. If pressing changes the tone, and tilting changes the current, a proposed model must account for those cross-connections too. A story invented separately for each experiment is less informative than one structure that survives them all.

Still, finite evidence rarely fixes a unique hidden mechanism. A clever engineer could build two different interiors that agree on every test we have tried. This is underdetermination: the evidence can narrow the possibilities without leaving only one. Relational knowledge can therefore be excellent without becoming omniscient.

The lesson is not “the inside does not matter.” It is that claims about the inside earn their authority through a public pattern of consequences. The better that pattern holds under new probes and combinations, the less our description resembles a guess and the more it resembles a structural identity.

Composition is where a web becomes more than a picture of connected dots. It lets local knowledge travel. If a recipe converts ingredients into dough and an oven converts dough into bread, their composition converts ingredients into bread. If a translation carries English into Spanish and another carries Spanish into Quechua, their composition carries English into Quechua.

Order matters. Walking home-to-café and then café-to-station is possible; reversing the same arrows may not be. Some relations are reversible, some have a different return route, and some are one-way. Category theory does not assume symmetry. It asks only that composable arrows have a composite and that the rules remain coherent.

Every place also has a stay-put route: home to home without going anywhere. It sounds trivial, yet it acts like zero in addition. Composing a route with this identity route changes nothing. Without identities, the algebra of paths has a missing piece and the later idea of representing an object by all arrows into it will not work cleanly.

Finally, composition must be associative. For home → café → station → park, you may combine the first two routes and then the third, or the last two and then the first. The final composite is the same route structure. Parentheses organize the calculation; they do not alter the journey.

The phrase “every test” carries the weight. Matching color is not enough. Matching the locks currently in the room is not enough. Even matching a long list of measurements may miss how measurements transform when another operation happens first. A complete relational signature must cover all permitted probes and preserve their organization.

This is why the formal claim is subtler than “things are their connections.” A social network diagram may give two people the same number of friends while hiding who those friends are and how messages move. Two nodes can look alike under one summary and differ under the network’s full structure.

Mathematics also distinguishes equality from isomorphism. Equality says two expressions designate the very same object under the formal rules. Isomorphism says there are reversible structure-preserving arrows between them. Isomorphic objects can be treated alike for every question the category expresses, even when they remain two entries on the page.

That distinction is liberating. It allows us to ignore irrelevant presentation without pretending there is no difference whatsoever. A triangle rotated on paper, a database copied to another drive, and two vector spaces with chosen coordinates can be structurally equivalent in a specified sense while retaining different locations or histories.

In the category of sets, objects are sets and arrows are functions. In a category of spaces, objects may be topological spaces and arrows continuous maps. In a category built from an ordered list, an arrow can simply record that one item is less than or equal to another. The objects change; the grammar stays.

This abstraction is useful because it forgets on purpose. A category of sets does not ask what ink wrote the elements. A category of continuous spaces does not remember the temperature of the paper on which a diagram was drawn. It retains the distinctions needed for the stated arrows and discards the rest.

That selective forgetting prevents a common philosophical mistake. A mathematical object is not revealed by gathering every relation anyone could name. It is located relative to a chosen category. Before saying “all relations,” we must specify a domain of objects and the structure-preserving maps that belong there.

Once that choice is made, categories let us compare whole relational worlds. A functor sends objects to objects and arrows to arrows while preserving identities and composition. It is a translation that respects the grammar. Natural transformations will then compare such translations in a coordinated, object-by-object way.

Category theory writes this system as Hom(−, X), pronounced “hom into X.” For every test object A, it gives the set of arrows A → X. For every arrow B → A, it gives a way to transform A-probes into B-probes by composition. This entire assignment is a functor from the opposite category into sets.

Yoneda’s lemma says that natural transformations from Hom(−, X) to any set-valued functor F correspond exactly to elements of F(X). A transformation that seems to contain one choice for every possible probe is determined by a single element at X—provided all those choices are natural, meaning they agree with every precomposition.

A famous consequence is the part our story has approached: the Yoneda embedding is fully faithful. If Hom(−, X) and Hom(−, Y) are naturally isomorphic, then X and Y are isomorphic. Their complete incoming relational profiles recover their structural roles inside the category.

The word “naturally” is the hinge. Without it, one could arbitrarily pair route lists of equal size and call the objects the same. Naturality says every pairing must commute with every change of probe. The agreement must survive the structure’s own ways of looking from elsewhere.

The epistemic claim is the safest: we know and distinguish things through patterns of interaction. The structural claim is sharper: within a category, a complete naturally organized map-profile determines an object up to isomorphism. Yoneda proves this second claim. It does not prove that the physical universe is only a category, that every meaningful relation has been captured, or that there are no relation-independent features.

The ontological claim is stronger: perhaps objects do not first exist as self-contained lumps that later enter relations. Perhaps being an object is occupying a stable node-role in a structure, with objects and relations defined together. Ontic structural realists defend versions of this view, partly because modern physics often identifies entities by symmetries, invariants, fields, and patterns of interaction.

Here is the argument’s punch. If removing every relation removes every test, every causal power, every contrast, every persistence condition, and every way the supposed object could matter, what content remains in saying that a bare, relationless something is still there? The burden shifts. The relational view need not prove that a mysterious remainder is impossible; the defender of the remainder must explain what difference it makes.

Yet relations cannot be a fog with no terms. A route needs endpoints; a difference needs positions that differ. The strongest relational picture is therefore not “relations instead of things.” It is that relata and relations are co-defined within a structure. The node is not secretly complete before the web, and the web is not built from nothing. They arrive together.

This conclusion is an argued interpretation, not a corollary of Yoneda. It may be resisted by insisting on intrinsic quiddities, primitive thisness, or physical structure that the chosen category omits. The resistance is intellectually serious. But it must now answer a precise challenge: name the nonrelational remainder, explain how it distinguishes possibilities, and show why the explanatory work cannot be done by structure.