1. Kinds

Some categories have a rich structure, in that their instances share a great many properties.

For example, all horses are alike, not only in being horses, but in eating grass, growing manes, having uncloven hooves, and sharing a great number of other behavioral, anatomical, and physiological features.

Following John Stuart Mill, we call this sort of category a Kind (and retain his capitalization for this and related terms).Footnote ¹ The defining characteristic of Kinds is that any Kind K will enter into a plurality of informative synthetic generalizations of the form: All K are G.Footnote ²

Note that by no means all properties pick out Kinds. For example, square things, or soft things, or small things, do not form Kinds. There are no informative generalizations of the form: All square/soft/small things are G.

Higher biological taxa, as well as species, also form Kinds. For example, all mammals have fur, sweat glands, milk glands, and other distinctive anatomical features. Note how the features common to some higher taxon will be a subclass of those common to its subordinate taxa. The features common to all mammals are among the features common to all horses.

Biological categories are not the only Kinds. Chemical substances are also Kinds. All samples of sulfuric acid have the same melting and boiling point, the same propensity to combine with other substances in fixed proportions, the same liquid density, the same electrical and thermal conductivity, and so on.

Astronomical objects also fall into Kinds. All main sequence stars (those powered by hydrogen fusion) are spherical, in hydrostatic equilibrium, radiate energy of certain wavelengths, and so on. Other astronomical Kinds include red giant stars, white dwarf stars, and supernovae. Further types of Kinds include meteorological categories and geological formations.

2. Kinds are Useful but not Anthropocentric

The rich structure of Kinds makes them very useful to human beings.

For one thing, we can normally ascertain that something is an instance of some Kind K by observing just one, or some few, of its characteristic Gs. For example, you can reliably identify horses by the characteristic shape of their heads or gold by its characteristic electrical conductivity. And then, if you are appropriately informed, you can infer all the other characteristic Gs from this.

For another thing, there are often shortcuts to becoming appropriately informed about the Gs that are characteristic of some Kind. This is because Kinds fall into Types that specify which properties will be shared by all the instances of Kinds of that Type.

So, for example, vertebrate species are one Type of Kind, in that all the animals in any given vertebrate species will share morphology, anatomy, physiology, diet, reproductive behavior, and various other properties (but not injuries, or learned behaviors, for example). Of course, the determinate morphology, anatomy, and so on, of any one vertebrate species will be different from that of the others; still, the different vertebrate species will all coincide in that each is characterized by determinates of the same set of determinable properties.

Similarly, chemical substances are another Type of Kind. All samples of any given chemical substance in a given state of matter will typically have the same color, taste, odor, density, electrical and thermal conductivity, and so on (but not, for example, the same shape, or size).

If you know the ‘template’ of determinable properties for a certain Type of Kind, as Ruth Millikan has put it (Reference Millikan2000: chap. 1.8), then you are in a position to make one-shot inductions. From a traditional perspective, it can seem puzzling that one experiment on one piece of gold can tell you about the electrical conductivity of all gold, or that one dissection of a hippopotamus can tell you about all hippopotamus bladders. However, once you know the template for chemical substances, or for vertebrates, then these inferences are secure.

All in all, facts about Kinds are very useful to humans who want to make inductive inferences. But it should not be thought that Kinds somehow come into existence in order to serve the inductive interests of humans. Rather, the Kinds first have their structure, and then humans use this structure to help them find their way around the world.Footnote ³ Most Kinds would exist even if no humans or other intelligent beings had ever evolved to base inferences on them. (Indeed it is likely that many of our distinctive intellectual powers evolved specifically to enable us to discern the preexisting structure of Kinds.)

In what follows we observe that entities like works of literature and social groups are also Kinds. Such entities do depend on humans for their existence. But the basic point remains. They are not constituted in order to serve the inferential purposes of humans. It is just that their structure lends itself to these inferential purposes once they do exist.

3. Super-explanatory Properties

When we have a Kind K whose instances share many different properties G, there will typically be some single property E of their instances that causally explains this multiple commonality. Our eventual aim in this article is to show that such super-explanatory properties are also essential in the modal sense of being possessed by Kind instances across all possible worlds. But that is for later. Our first task is to explain the role of super-explanatory properties in the actual world.

For example, the atomic constitution of gold explains why all samples of solid gold have the same density, electrical and thermal conductivity, melting and boiling point, and so on. More generally, the molecular constitution of any given chemical substance will explain why its instances share many corresponding properties.Footnote ⁴

Similarly, the internal physical constitutions characteristic of the various Kinds of astronomical objects explain why the instances of such Kinds (the main-sequence stars, the white dwarfs, the red giants, and so on) share their many other common properties.

Moreover, as we show below, the members of any given biological taxa will also share a single property that explains their multiple commonalities—though in the biological case, we argue, this will not be a shared intrinsic property, as it is with chemical or astronomical Kinds.

We can view such super-explanatory properties as illustrations of the principle of the common cause. In general, when we find that some A and B are correlated (in the sense that they are co-instantiated more often than we would expect given their separate probabilities of occurrence), then it will be the case that either A causes B, or B causes A, or both A and B are joint results of some common cause.

Kinds are in effect rich complexes of correlations. All the many Gs characteristic of a Kind are correlated with each other. As soon as we know that something has one such G, this typically makes all the other Gs much more likely. However, at least in the cases we have considered so far, the different correlated Gs do not cause each other. The melting point of gold does not cause its density. The bladders of hippopotami do not cause their ears. So correlations between properties like these must be due to some single common cause, some super-explanatory property of the instances that is responsible for their all sharing so many other properties.

Some doubt that the principle of the common cause can be defended in any general form (Arntzenius Reference Arntzenius and Zalta2010). We disagree, but need not pursue the issue here. The more important point is that Kinds typically have super-explanatory cores, not whether this fact can be derived from some exceptionless principle.

So for us, Kinds are any categories whose instances display multiple commonalities. A super-explanatory property for such a Kind is then a single property that causes all the other shared properties. Such causal powers will perforce be conditional, depending on laws of nature and possibly normal background circumstances, in addition to the super-explanatory property itself. Still, as long as a single property, plus such conditions, accounts for all the other shared properties of a Kind, we count it as super-explanatory.

4. The Structure of Kinds

Do all Kinds have super-explanatory properties? Functional Kinds are one interesting case. Consider the Kind aerial insectivore. All the swallows, martins, swifts, insectivorous bats, and other flying insect-eaters share a range of properties, including acute sensory systems, ability to swoop, and pointed beaks or mouths. Kinds like this are the result of convergent evolution. The aerial insectivores all belong to species that have been shaped by similar selective pressures.

Some are suspicious of Functional Kinds, on the grounds that they only share a few superficial properties, by comparison with the multitude of properties shared by the members of any biological taxon.Footnote ⁵ While this is certainly true, and indeed a point to which we return below, we are happy to accept Functional Kinds as genuine Kinds, albeit of limited informativeness. In line with this, we would say that the common selective pressures that give rise to Functional Kinds are themselves a super-explanatory common cause for the shared properties of those Kinds.Footnote ⁶

We do not completely want to rule out Kinds that do not in any sense have a super-explanatory core property. Maybe there are Kinds where the shared Gs do cause each other, rather than stemming from some common cause. Perhaps some meteorological categories are like this. (For example, tropical cyclones might be seen as occurring when strong winds, heavy rain, spiraling thunderstorms, and a low-pressure center all reciprocally reinforce each other; see Millikan Reference Millikan2017: ch. 1. Relatedly, psychiatric disorders like depression are arguably sustained by feedback between different symptoms; see Borsboom Reference Borsboom2017.)

As with Functional Kinds, one might feel that there is something rather thin about Kinds like tropical cyclones, or depression, that gain their common properties from cycles of reciprocal causation. We will not push this point, however. None of our arguments depend on the claim that all Kinds have instances that share some super-explanatory property. Rather, we are interested specifically in analyzing those Kinds that do have super-explanatory properties. If there are Kinds that lack such a property, so be it. They are not our focus.

On Richard Boyd's influential account (Reference Boyd1991, Reference Boyd and Wilson1999), natural kinds are ‘homeostatic property clusters’. We certainly agree that Kinds are property clusters, in the sense that their instances share many properties. But we disagree with his emphasis on homeostasis in the sense of self-regulating feedback mechanisms. Perhaps the clustering of properties in a few special kinds—tropical cyclones, depression—is due to some such self-equilibrating homeostatic mechanism. But there is in general nothing homeostatic about the way that microstructures give rise to the shared properties of chemical substances. Nor does our account of biological Kinds developed below in section 6 appeal to homeostasis in any way.

Points like these have led a number of authors to argue that Boyd's analysis should be replaced by a ‘simple casual account’ of natural kinds (Craver Reference Craver2009: 579; Khalidi Reference Khalidi2013: 78). On this view, the clustering of properties in any natural kind are due to some causal structure, but this need not involve any homeostatic process. We agree with this turn away from homeostatic mechanisms, but we feel that an undifferentiated appeal to causal structure misses the widespread significance of super-explanatory properties. In our view, the great preponderance of natural kinds owe their clustering of properties, not just to some causal structure or other, but to one single underlying property that serves as the common cause of all the other clustered properties.

For completeness, it is worth observing that some categories technically fit our definition of Kinds in supporting a plurality of generalizations even though there is no unified explanation for the correlations involved, not in terms of any causal structure. For example, jadeite and nephrite are traditionally lumped together as jade on account of their similar color, hardness and density, even though they are two quite different minerals (cf. Bird Reference Bird, Beebee and Sabbarton-Leary2010). We say that a Kind is genuine if its correlations have a unified causal explanation and ersatz when it results from two or more genuine Kinds fortuitously coinciding in some of their properties. Since ersatz Kinds are by their nature rare, depending on a lucky coincidence of properties in different genuine Kinds, we put them to one side in what follows. It might seem odd to count ersatz Kinds as Kinds at all, but nothing hangs on this in what follows, since our focus henceforth is exclusively on Kinds whose instances do share a single super-explanatory property.

To summarize our taxonomy: any category supporting a plurality of generalizations is a Kind; only those for which this clustering has a unified causal explanation are genuine; for most but not necessarily all genuine Kinds, that explanation involves a super-explanatory common cause.

5. Historical and Eternal Kinds

Consider a rather different type of Kind from those discussed so far—all the different copies of Alice in Wonderland. This qualifies as a Kind. The copies share a multitude of properties. Each copy tells the same story, with the same characters and scenes. Indeed they all tell it in the same way, with the same first word, the same second word, and so on, until the end.

The car model Vauxhall Zafira forms another Kind. All cars of this model have the same ingenious system of extra folding seats, the same undercar spare wheel holder, the same type of carburetor, and so on.

The correlations between the properties of these Kinds are not explainable by some common internal physical property, like molecular structure or stellar composition. Different copies of Alice in Wonderland can be physically quite different (paper, magnetic tape, and so on). And even if all Zafiras are physically similar, there is no one aspect of this similarity that provides a common explanation for all the other shared properties of Zafiras.

Rather, the super-explanatory feature shared by the members of these Kinds is their common origin. All the many instances of Alice in Wonderland have been reproduced in one way or another from Lewis Carroll's original manuscript. All the many Zafiras have been manufactured in line with the designer's original blueprint. That is why these instances all share a multitude of properties.

Ruth Millikan, from whom we have derived much of our general framework for thinking about Kinds, has coined the term ‘Historical Kinds’ for those Kinds whose shared properties are due to their instances all being copied from some common source (Reference Millikan1999).She contrasts these with Eternal Kinds, whose shared properties are explained by the instances having some common intrinsic physical property. (The term ‘eternal’ is perhaps not ideal, given that instances of Eternal Kinds can be in flux and short-lived—as with supernovae, say, or decaying elements—but we will stick with it.)

Literary works and car models might seem somewhat frivolous examples of Kinds. But consciously designed items like these are not the only Historical Kinds. Various types of social category can also be seen in this light: the many beliefs and behaviors characteristic of Christians, say, or molecular biologists, or Japanese women, are arguably shared because they are copied from prior models (see Godman Reference Godman2015, Reference Godman2018; Ásta 2018). In addition, as we now argue, biological taxa are also Historical Kinds.

6. Biological Taxa

What sort of Kinds are biological taxa? A first thought might be that they are Functional Kinds, with their shared properties accounted for by common selective pressures. But this fails to do justice to the rich structure of biological taxa. The members of any biological taxon typically share a wealth of features that are not explainable by the selective pressures that shaped that taxon. This is because natural selection does not design ideal organisms, but builds on what went before, is limited to materials thrown up by mutation, and is buffeted around by genetic drift. This ensures that biological taxa standardly share any number of nonfunctional features. We need to look beyond selective pressures to capture the nature of biological taxa.

Michael Devitt (Reference Devitt2008, Reference Devitt2010) has recently argued that we need to recognize intrinsicFootnote ⁷ biological ‘essences’, in the form of shared genomic material and associated developmental mechanisms, precisely in order to account for the many characteristic phenotypic commonalities displayed by the members of any given biological genus. (We have put ‘essences’ in scare quotes to underscore the point that at this stage we are still not concerned with any modal issues, but simply with the explanatory role of genomic mechanisms in the actual world.)

Devitt's position might suggest that we should assimilate biological taxa to Eternal Kinds, taking their shared genomes and developmental mechanisms to play the role of super-explanatory intrinsic properties. However, this is not the only possible way of fitting biological taxa into our analysis. An alternative option is to view biological reproduction and ontogeny as a copying mechanism that produces new members of biological taxa from old ones. This would then render biological taxa as Historical Kinds.

Does this have to be an either/or issue? The examples offered so far might have suggested that Eternal and Historical Kinds will always be mutually exclusive. We have considered Eternal Kinds, like chemical substances and stars, whose instances have had no causal connection to each other, but have simply arisen whenever the conditions were right. By contrast, the instances of Historical Kinds are necessarily causally connected to each other, related by descent through copying.

Still, the definition of an Eternal Kind as having a common intrinsic core does not itself preclude its instances from being causally connected to each other. So it remains open that some Kinds could be both Eternal and Historical, in the sense that their instances share some common intrinsic property and reproduce by copying.

Biological taxa could be a case in point. We could view them both ways. We could focus on the genomes and developmental mechanisms, and view reproduction as simply a part of the process that creates items with that intrinsic nature. Or we could focus instead on the reproduction, and view the genomes and development as simply a part of the process by which new instances are copied from old ones. The two perspectives are obviously compatible. This suggests that biological taxa are akin to both Eternal and Historical Kinds.

We nevertheless argue that biological taxa should be counted as Historical and not Eternal Kinds. In another article, Godman and Papineau (Reference Godman, Papineau and Bianchiforthcoming) point out that (a) even in sexually reproducing taxa, a significant number of taxon-typical biological characteristics are not inherited through the sexual bottleneck via intrinsic genetic properties of the zygote, but via non-genetic influences like behavioral imitation; and (b) in single-celled non-sexual taxa like bacteria, genomic material does not play any super-explanatory role, but is just one of the features copied in mitosis. On this basis, Godman and Papineau argue that a uniform account of biological taxa should treat them as Historical rather than Eternal Kinds.

It now strikes us, however, that there is a more fundamental objection to viewing biological taxa as Eternal Kinds. It is not just that this position fails to deal with non-genetic inheritance and non-sexual reproduction. Rather, even if we stick solely to genetically inherited traits in sexual species, it never offers the right kind of super-explanation at all.Footnote ⁸

Remember that we are looking for some common-cause explanation of the multiple correlations between the many characteristic phenotypic features of biological genera. However, the genomic material common to the members of a taxon will typically not be suited to play this role. This is because it will normally be a conjunction of different genetic properties, each one of which explains a different phenotypic feature. So we do not have one intrinsic property acting as a common cause for many phenotypic properties, but simply a list of different intrinsic genetic properties explaining different phenotypic properties. And this thus leaves us once more with an unexplained multiple correlation—why are all those different intrinsic genetic properties found together?

By way of comparison, note that it would not be much of an explanation of why all gold, say, has a distinctive color, density, electrical and heat conductivity, and so on, to be told that gold contains an intrinsic property I₁ that determines its color, a different I₂ that fixes its density, another I₃ for its electrical conductivity, and another I₄ for its heat conductivity, and so forth. This would give us no account of why all these intrinsic I-properties occur together in any sample of gold.

By the same coin, it is no explanation of the many features common to all horses, say, to specify that horses all have some genetic material that determines manes, and some other genetic material that determines uncloven hooves, and some other genetic material that determines their head shape, and so on. We would not yet have explained why all these different genetic features are found together in horses.Footnote ⁹

In the case of gold and other chemical substances, of course, it is molecular structure that provides the requisite common cause. The different features of gold all stem from one common source, its atomic structure, and not from a number of different intrinsic properties. But in the biological case, the different genetic elements in some taxon's genome are not tied together in this way.Footnote ¹⁰

To find a property that explains why all the different phenotypic features of a biological taxon are instantiated together, we need to turn away from ontogenetic development and view the taxon as a Historical Kind. The reason the phenotypic features of a taxon are found together is that its members are all descended from common ancestors who had those features. Biological reproduction is a copying process—offspring share the heritable features of their parents.Footnote ¹¹ So a set of properties that is conjoined in an ancestral population will be found together in the descendants, too.

Note that nothing in our analysis requires biological taxa or other Historical Kinds to be immutable. Reproduction does not have to mean exact replication; and natural selection operating on so-introduced and preexisting variation can lead to changes in the set of common characteristics displayed by a Kind. When this happens to a significant degree, then this will amount to emergence of a new Historical Kind, as most familiarly illustrated by the emergence of new species and other biological taxa.

One last point before we turn to modal matters. Ernst Mayr (Reference Mayr1961) introduced a well-known distinction between proximal and ultimate biological explanations. A proximal explanation tells us how an individual organism ontogenetically develops the phenotypic features that are characteristic of its species. An ultimate explanation tells us why that species phylogenetically came to have those characteristics in the first place. Proximal explanations will refer to the developmental program that enables the zygote to develop into a mature organism. Ultimate explanations will appeal to the historical circumstances that shaped the species, including selective pressures.

Devitt (Reference Devitt2010) explicitly relates his analysis to Mayr's distinction. He observes that his intrinsic essences answer Mayr's proximal question, in that they explain individual development in terms of genetic material and associated developmental mechanisms. He also responds to Mayr's ultimate question by adding a further historical element to biological essences. So Devitt's overall view is that taxon essences are partly intrinsic and partly historic: the intrinsic component explains proximal issues of individual development, while the historic component explains ultimate issues of taxon origins.

Our analysis of biological taxa is motivated by a question that is quite distinct from both of Mayr's questions. Where he is interested in ontogeny and phylogeny, we are interested specifically in explaining why so many different properties are tightly correlated in the members of any given biological taxon, not in how those properties develop as those organisms mature from zygotes, nor in why the taxon came to have those properties in the first place.

It is clear enough that we are not addressing Mayr's first proximal question. When we explain the clustering of a taxon's properties in terms of the common ancestry of its members, we simply take developmental mechanisms for granted. We assume that offspring will resemble their parents, and use this to infer that all taxon members will share the properties that were originally displayed by the founding population. From our perspective, the development of mature organisms from fertilized zygotes is not explained, but rather presupposed as part of the copying mechanism that ensures the resemblance between parents and offspring (a resemblance, remember, that is differently accomplished in non-genetic inheritance and in non-sexual taxa).

Nor are we answering Mayr's ultimate question. Even though our biological super-explanatory properties are historical, they do nothing at all to address Mayr's ultimate issue. By way of analogy, note that when we explain why all the copies of Alice in Wonderland, or all the Vauxhall Zafiras, share so many correlated properties, in terms of their all being copied from a common source, this does not even start to explain why Alice and the Zafira were designed as they were in the first place. While these are certainly questions worth asking, they are not our present focus, and we do nothing to answer them. We simply take the features of the originals as given and explain the commonalities in the copies by that alone. Similarly, when we explain, for example, why horses share so many common features, we start from the fact that some few original horses had those features and explain the correlations by that alone, without wondering where those original features came from.

Our question is thus different from both of the questions traditionally addressed by biological theorists. They are interested in ontogeny and phylogeny, where we are interested in explaining correlations. Moreover, our approach simply assumes answers to the traditional questions, taking as given both the machinery that allows offspring to resemble their parents, and the original traits of founding populations.

Still, even if our question is not one that has traditionally concerned biological theorists, this does not mean it is not philosophically important in its own right.

In this connection, note first that our correlational question is the one we need to focus on if we want to uncover some notion of core property that applies to Kinds in general, and not only to biological taxa. All Kinds involve correlations that call for explanation in terms of super-explanatory properties. But they do not all have ontogenies or phylogenies about which to ask Mayr's proximate and ultimate questions. Chemical samples, for example, do not develop ontogenetically from seeds, nor do their shared properties result from evolutionary processes.

Moreover, as we are about to argue, it is specifically correlation-explaining super-explanatory properties that hold the key to metaphysical modality. If we want a species of essence that is invariant across metaphysically possible worlds, we need to focus on super-explanatory properties, rather than developmental machinery or evolutionary origins.

7. The Necessity of Super-Explanatoriness

We now turn to modal issues. A central thesis of this article is that super-explanatory properties are metaphysically necessary.Footnote ¹²

This proposal fits well with a range of common intuitions about which properties are metaphysically necessary to Kinds.

Take chemical substances. As we have seen, molecular structure plays the super-explanatory role for chemical substances. In line with this, there is a widespread intuition that chemical substances have their molecular structure necessarily. The intuitive consensus is that there are no metaphysically possible worlds where water is not H₂O, and moreover that, in any possible world, anything that is H₂O is water.

It works similarly with biological taxa and other Historical Kinds. Here the super-explanatory property is shared origin. Intuitions about necessary features are more disputed for biological than they are for chemical substances, but even so it seems natural enough to take origins to be essential to taxa: you cannot possibly be a tiger unless you are part of the lineage that starts with the original tigers, and necessarily anything that is part of that lineage (without too many modifications) is a tiger. Corresponding intuitions apply to Historical Kinds in general. For example, you cannot possibly be a copy of Alice in Wonderland if your provenance does not trace back to Lewis Carroll's original manuscript, and necessarily any reasonably faithful copy with such a provenance is a copy of Alice in Wonderland.

As we say above, Functional Kinds tend to be less rich than other Kinds. But, once more, it is not unnatural to regard the selective pressures that super-explain their shared properties as modally necessary and sufficient for Kind membership. You cannot possibly be an aerial insectivore unless natural selection has shaped you to catch insects, and necessarily anything so shaped is an aerial insectivore.Footnote ¹³

We do not want to rest our case on intuitions, however. While we take our position to fit reasonably well with the intuitive judgments of both philosophers of modality and non-specialists, we attach more importance to the fact that our proposal also suggests a principled explanation for these intuitions. In our view, the necessity of super-explanatory properties is not just some arbitrary conceptual quirk, but a natural consequence of the role that metaphysical modality plays in our cognitive economy.

One important aspect of metaphysical modality is the way it structures counterfactual thinking. Many everyday concerns call for us to consider what would have happened under some counterfactual supposition. When we counterfactually suppose that some item lacks some property, we naturally hold most of its other actual properties fixed. (Suppose Nixon had lost the election . . .) But when we counterfactually suppose that some Kind lacks a super-explanatory property, we are prevented from holding most of its other properties fixed, given that counterfactually supposing away a cause typically requires us to suppose away its effects, too. So when we suppose away the super-explanatory core of some Kind, all bets are off. We are hypothesizing away all the many correlated properties that distinguish that Kind from others. The natural reaction is that we have thereby supposed away the Kind altogether.

We thus see how super-explanatory properties guide and constrain counterfactual thinking. Without such constraints, counterfactual suppositions could be developed in virtually any way, and so become theoretically and practically irrelevant. Timothy Williamson (Reference Williamson2007) similarly appeals to our ‘sense of how nature works’ (143) in reliable counterfactual thinking. While he does not speak in terms of essential properties, he stresses the role of background knowledge of ‘constitutive facts’ (143) for correctly developing the supposition in a counterfactual conditional. For Williamson, we can generally trust our capacity for assessing counterfactual conditionals to deliver the correct answers because it is informed and constrained by such a knowledge of ‘constitutive’ facts. Our account here agrees with Williamson's and importantly further clarifies what those constitutive facts are—namely facts about essential super-explanatory properties—as well as why they have such a special status in counterfactual thinking (see Mallozzi Reference Mallozziforthcoming; Vaidya and Wallner Reference Vaidya and Wallner2018).

Note that nothing in our account rules out possible worlds where a liquid with the superficial properties of water (odorless, colorless, tasteless, and so on) is not H₂O, and ones where a species with the superficial properties of tigers (striped, solitary, carnivorous, and so on) are not descendants of the original tigers. But that liquid would not be water, nor that species tigers, precisely because their correlated properties do not stem from the same super-explanatory source as in the actual world. We do not count them as the same Kinds, despite their superficial resemblance, because they have been put together in a different way. They do not have the crucial property that pulls together the categories that exist in the actual world.

Before proceeding, it will be helpful to add some precision to our arguments. We have on occasion drifted into talking of super-explanatory properties of Kinds, and of Kinds possessing those properties in all possible worlds. But is not strictly the Kinds themselves that have the super-explanatory properties, but their instances; and correspondingly what is strictly necessary is not that the Kind has the essential property, but that any instance of the Kind does. Thus:

There is also the converse implication:

Principle (1) says that any super-explanatory E is required for Kind membership across possible worlds. E is an essential property, in that you cannot belong to the K if you lack it.

Principle (2) says that any super-explanatory E is sufficient for Kind membership across possible worlds. E constitutes an essence, in that you must belong to the K if you have it.Footnote ¹⁴

Note that it does not follow from (1) and (2) that a property that is essential to, or an essence of, some Kind must be essential to, or an essence of, the individuals that belong to that Kind. Having professional accreditation is essential to and the essence of being a medical doctor—necessarily you have the accreditation if and only if you are a doctor. But that accreditation is not essential to (and still less is it the essence of) individual doctors being who they are. Any doctor could have failed to be a doctor; indeed no doctor was always a doctor, and many doctors cease to be doctors.

8. Metaphysical and Nomological Modality

It has been put to usFootnote ¹⁵ that super-explanatory properties do not really involve metaphysical modality, but only highlight certain aspects of nomological modality. After all, super-explanatoriness depends on nothing more than nomological relations between properties in the actual world, and so it does not have to be seen as taking us beyond actuality.

We agree that our account of Kind essences gives priority to the actual world. Super-explanatoriness is fully determined by actual-world nomological relations. But we do not think that this disqualifies it as an account of metaphysically modally essential properties. After all, super-explanatoriness involves a quite specific kind of nomological structure. It is not just a matter of one property's being nomologically related to another, as gas pressure, say, is related to temperature. Rather, it involves a given property's being the common cause of a many other properties, and thereby explaining why those other properties are so tightly correlated together.

Our proposal is thus not to eliminate modal Kind essences in favor of nomological necessity, but rather to reduce them to a specific kind of nomological structure. This structure is distinguished from causal-nomological structure more generally by featuring single common causes that give rise to many other properties in what we can call ‘one-to-many’ causal networks.

We propose that, whenever we find this one-to-many nomological structure characteristic of a super-explanatory property, then that property will be possessed in all metaphysically possible worlds. We take this to explicate the familiar notion of the nature of things as the source of metaphysical necessity. In our view, we gain a better grip on this notion by equating it with the underlying cause of the many properties and behaviors that consistently co-occur in the instances of a Kind—namely, the essence of the Kind as we understand it.

To drive the point home, note that it is a consequence of our view that modal Kind essences would stay fixed even in possible worlds where the relevant laws of nature were different. Even if variation in laws of nature meant that H₂O were no longer odorless, colorless, tasteless, and so on, it would still be water. We might be explaining modal Kind essences in terms of actual-world super-explanatoriness. But even so our account clearly implies that these modal essences transcend nomological connections, by staying fixed even when the latter vary.Footnote ¹⁶

In this sense, then, super-explanatory properties provide a kind of bridge between the actual world and the modal realm.Footnote ¹⁷ Super-explanatoriness depends on nothing beyond a specific kind of casual role in the actual world. But the specific structure of this causal role shows how metaphysical necessity is tied to the nature of things and so explains why super-explanatory properties remain invariant across all possible worlds.

9. Identity and Super-Explanatoriness

It has been put to usFootnote ¹⁸ that the modal consequences of (1) and (2) follow directly from the necessity of identity, and therefore that super-explanatoriness is not needed to explain them.

The thought is that such identities as water = H₂O, and tigerhood = being descended from the original tigers Footnote ¹⁹, and so on, together with the necessity of identity, are themselves enough to ensure the necessary coextensiveness specified in the consequents of (1) and (2), without any help from the super-explanatoriness referred to in the antecedents.

We agree that the modal consequences of (1) and (2) follow from Kind identities like water = H₂O, and tigerhood = being descended from the original tigers, together with the necessity of identity. But we think that such Kind identities are themselves a consequence of (1) and (2). We identify water with H₂O specifically because this molecular structure is super-explanatory of the many shared properties of water. Because of this super-explanatoriness, we regard water and H₂O as necessarily coextensive, and this then leads us to identify them.

After all, there are theorists who do not agree that

(We will stick to water. Everything that follows in this section would apply equally to tigerhood.) These theorists reject an externalist account of the concept water, and instead hold that it refers to anything that satisfies such nominal requirements as odorless, colorless, tasteless, freezes at 0^OC, etc. They allow that all the actual water is in fact H₂O. But they see no reason why this property should have a privileged role in relation to the nature of the Kind water, at the expense of the properties by which we recognize the Kind. They thus maintain that

and deny that (3) water = H₂O.

In line with this, they hold that it is by no means necessary that water is H₂O, or indeed that H₂O is water. They think that there are possible worlds in which water—the odorless, colorless, etc. liquid—is not H₂O, and worlds in which H₂O is not water—the odorless, colorless, etc. liquid—due to different types of observers or different laws of nature.

Now, we of course think that these anti-externalists are wrong about water (and we would say the same about similar anti-externalist views about other Kinds). But the interesting question is why they are wrong.

Note that nothing is extensionally amiss with their account. They will classify just the same actual liquids as water as anybody else.

In our view, the reason for identifying water with H₂O, rather than with odorless, colorless, tasteless, freezes at 0^OC, etc., is precisely that H₂O is super-explanatory with respect to water, in a way that the nominal features of water are not, despite their extensional adequacy. The identity (3) water = H₂O is thus a consequence of the earlier principles (1) and (2), not some independent claim from which they follow. Because H₂O is super-explanatory, we take it to be both necessary and sufficient for water across modal space, as in (1) and (2). And this is why we then identify water with H₂O, as in (3), and not with any set of its more superficial properties.

A super-explanatory property of a Kind plays a special role. It is a single property that pulls together all the many other common features of the Kind. None of the Kind's other nominal features have this role. They are simply effects of the common super-explanatory core.

This is why we take super-explanatory properties to be modally required for Kind membership, as in (1) and (2). An instance with a super-explanatory property has the feature that actually causes all the other distinctive properties of the relevant Kind. And so possible (or indeed actualFootnote ²⁰) instances still count as members of that Kind, even if they lack some of the nominal features of the Kind. Conversely, if we posit an instance that has all the nominal features, but not the super-explanatory essence, it does not count as a member of the relevant Kind. As we observed earlier, a liquid that was odorless, colorless, tasteless, etc., but was not H₂O, would not be water, despite its superficial resemblance. The category would have been put together by hand, so to speak, rather than deriving its shared features from the core property that pulls them together in the actual world.

So, yes, once we are given such identities as water = H₂O, and tigerhood = being descended from the original tigers, then it will follow that the properties that flank the identity sign are necessarily co-extensional. But the reason for embracing these identities in the first place is precisely that the properties on the right-hand sides are super-explanatory essences.

10. Further Issues

Many believe, following Fine (Reference Fine1994), that necessary truth across all metaphysically possible worlds is never brute, but always a consequence of essences. While this fits our account of necessities that depend on super-explanatory Kind essences, we are uneasy about trying to force all metaphysical necessities into the single mold of essence.

For a start, there is the necessity of logic itself. This is not the place to explore its basis, but there is no obvious reason to suppose that this will hinge on the essence of anything. Nor do we think that the necessity of identity itself is helpfully viewed as depending on essences, given the way it follows from nothing but the legitimacy of de re modal constructions.

Having said this, we do think that the idea of super-explanatoriness promises to cast light on a wider range of metaphysical necessities than just Kind essences. Ruth Millikan (Reference Millikan2000) has observed that individual people, animals, and other persisting objects are akin to Kinds in a number of respects. If we think of the instances of any such individual as its temporal stages, then these instances will share many distinctive properties, small subsets of which will typically suffice to identify the individual in question. Given this, we can view any persisting individual on the model of a Historical Kind, with the shared features of its instances being super-explained by its origin, and elements of that origin—like the parents of people, or the material constitutions of lecterns—therefore coming out as essential properties.

It is worth noting how this further possible application of super-explanatoriness, while rendering certain metaphysical necessities non-brute, is at variance with Fine's program of basing all metaphysical necessity specifically in essences. Since different persisting individuals sometimes have the same origin—think of identical twins—we can have different individuals with all the same essential properties, which means that these essential properties do not constitute modally sufficient essences for those individuals.

Still, these are issues for another time. For now, we hope that we have at least done enough to show that super-explanatoriness is the right way to understand the metaphysically necessary properties of Kinds.