2.1. The SPC view

The SPC account is a view that conceives natural kinds as corresponding to property clusters. Property cluster views, like the homeostatic property cluster account (the HPC account), hold that the membership in a kind is a matter of having a loose group of properties rather than a fixed set. But what differentiates the SPC from the HPC theory is the type of stability that is possessed by property clusters. Whereas the stability that homeostatic property clusters enjoy is causal stability, meaning that they are stabilized by causal mechanisms, the stability enjoyed by stable property clusters is cliquish stability. Cliquish stability is a type of stability attributed to “cliques” of properties. Properties can be cliquish with one another by frequently being instantiated together, meaning that the possession of some properties by an individual probabilistically indicates possession of the whole cluster (Slater, Reference Slater2015, 397).

But how do the property clusters with cliquish stability remain stable? In other words, how do the SPC kinds retain certain “sameness” so that it is possible for us to make inferences about the members of the kinds? In lieu of a metaphysical entity such as a causal mechanism to ground the stability, the clusters are “epistemically grounded.” What does it mean for the SPC kinds to be epistemically grounded? Simply put, domain experts rely on their scientific convention to set the threshold of likeliness of whether an individual is a member of a kind in a specific domain. This unique feature is coined as an “epistemology-only” feature of the SPC view by Kendig and Grey (Reference Kendig and Grey2021).

The epistemology-only feature is connected to the metaphysical neutrality of the SPC account. It is metaphysically neutral when it comes to the relation of the properties in a cluster. For instance, if a property cluster theory stipulates that there is a causal relation between properties, it would provide a causal explanation to the question of why a particular property cluster “clusters” in a certain way. However, metaphysically neutrality on the relation that is held between properties renders the SPC view an epistemology-only account of natural kinds.

In the following subsections, I will elaborate on two strands of worries; the first worry is what I call the lack of causality worry, and the other, is the undecided metaphysical assumption worry.

2.2. The lack of causality worry

In this section, I bring in two versions of criticism that critique the SPC kinds’ lack of causality between properties. I first mention Khalidi (Reference Khalidi2018). Khalidi starts with a general idea that natural kind terms are projectible to argue that the causality of natural kinds is the ontological base of their projectibility, implying that projectibility is not a freestanding quality. A projectible natural kind term can be generalized to other predicates and used in inferring the predicates. The projectibility of natural kind terms, which is just the robustness of the relation between the natural kind terms and other predicates, explains why natural kind terms are so useful in inductive practices.

But projectibility cannot explain why certain predicates are connected to a natural kind term. According to Khalidi, it would be circular to try to explain the connection with projectibility; therefore, one needs something to ground projectibility that reflects the goal of scientific inquiry, which is to elucidate the causal structure of the world. According to Khalidi, the causality between core properties, which are the causal base for other properties, and derivative properties, which are causally dependent on core properties, is what grounds the projectibility of kinds and explains the causal structure of the world.

The metaphysical model that Khalidi provides reflects his argument that causality is more ontologically fundamental than projectibility. Instead of a property cluster being a unit of relations, individual property types have causal relations with one another, constituting a web of a network that has properties as nodes and causal relations as edges. Still, Khalidi maintains that natural kinds are property clusters; more specifically, they are clusters of nodes (core properties) with many edges that are connected to other nodes (derivative properties).

Lemeire (Reference Lemeire2021) argues that cluster theories only provide a partial explanation of the epistemic fruitfulness of kinds which are understood as projectibility, explanatory power, and taxonomic power. He believes that the usefulness of natural kinds comes from kind-based explanations and categorization of their members. The clustering of properties is insufficient to account for the explanatory and taxonomic powers of natural kinds, and the causal relation between properties is required.

Without an account of a causal relation between properties that make up a kind, according to Lemeire, a theory of natural kinds fails to meet what he calls “the causal homogeneity requirement.” The causal homogeneity requirement demands that “kind membership is explanatorily relevant to the instantiation of one of the properties associated with the kind, only if there is no more than one cause of that property associated with the kind” (Lemeire Reference Lemeire2021, 204). In other words, a natural kind successfully explains an explanandum where there is only one or a combination of properties that causally explain the explanandum. However, the pure cluster theories are unsuccessful in kind-based reasoning because they lack causal homogeneity. They do not provide a causal structure between the properties that make up property clusters.

I believe Lemeire’s argument, like that of Khalidi’s, is valuable in the sense that it points out the importance of the causal relation between properties for the causal explanatory power of a natural kind theory. It is stressed, in both arguments, that the cluster theories must go beyond projectibility to explain epistemic fruitfulness. To summarize, the lesson of the two criticisms against the SPC account is that there is a good reason to identify the causality between properties. It is to elucidate the explanation between a member and its membership to a natural kind, and the lack of an elucidation renders the SPC account an unsuccessful theory of kinds.

2.3. The undecided metaphysical assumptions worry

The second strand of issues criticizes the lack of metaphysical assumptions which is a unique feature of the SPC account. I will briefly describe Kendig and Grey’s (Reference Kendig and Grey2021) criticism of epistemology-only views, such as the SPC account, and Martinez’s (Reference Martinez2017) criticism of the lack of metaphysical ground for potential kinds.

Kendig and Grey argue that epistemology-only views that solely rely on epistemic grounding cannot explain the epistemic value of natural kinds. This is because the epistemic value of natural kinds is based on their metaphysical structure (Reference Kendig and Grey2021, 1). In other words, if a theory of natural kind withholds a metaphysical assumption about the ground on which the explanandum kind is based, then it becomes challenging to explain why the said kind has an epistemic value. The lack of metaphysical assumptions in the SPC account ultimately leads to some cases in which scientists are free to make erroneous metaphysical assumptions about a kind.

Kendig and Grey provide three cases that illustrate this limitation of the SPC account. The case that I’m engaging in this section is the case of Herrnstein and Murray (Reference Herrnstein and Murray1994) on the relation between race and IQ. Kendig and Grey argue that the reason that Herrnstein and Murray contend that IQ and racialized morphological traits are cliquish is because they erroneously presuppose that some genes are responsible for both traits. Genes were wrongfully selected as a metaphysical ground that stabilizes the cliquishness of racial traits and IQ, and this was possible due to the undecided metaphysical assumption in the SPC account. Although I do not endorse Kendig and Grey’s analysis of the SPC account’s blameworthiness in the case, it is possible to see the moral of the case; one cannot guarantee an accurate metaphysical assumption that bases the attribution of kindhood to property clusters if one only relies on epistemic grounds for stability.

The metaphysical neutrality of the SPC account is criticized by Martinez (Reference Martinez2017) as well. He argues that SPC should include an account of a metaphysical ground; more specifically, Slater should accept the grounding claim, which argues that the metaphysical ground that “binds” the properties together determines the epistemic value of natural kinds. Again, as asserted by Kendig and Grey, natural kinds have explanatory power when there is a metaphysical account that explains why some properties cluster together.

A metaphysical ground is crucial in the case that Martinez illustrates where multiple grounds stabilize a new kind. In the case of a classification of lung cancer, there has been a need to categorize a new subclass of non-small-cell lung cancer (NSCLC) that responds to targeted treatment rather than standard chemotherapy. This new subclass of NSCLC is caused by mutations in the epidermal growth factor receptor gene (EFGR). Thus, considering only the property clusters that pick out NSCLC is futile in picking out the EFGR-positive subclass of NSCLC. To respond to such a challenge, a proponent of the SPC account might add properties like responsiveness to EGFR-targeted treatment in the property cluster. However, to evaluate the stability of the property cluster with the addition of new properties, she needs to look for the presence of the EGFR mutation. This means that we need to examine the metaphysical ground to check the stability of the property cluster. Therefore, rejecting the grounding claim is not a good strategy for the SPC advocate because an account of a metaphysical ground determines the stability of potential natural kinds.

The common theme in the two objections described in the preceding text is that a theory of natural kinds requires a metaphysical account of grounds that stabilize the property clusters. The metaphysical neutrality of a natural kind theory contributes to the failure of the theory to explain why the kind possesses an epistemic value. This is clear in the cases that Kendig and Grey and Martinez illustrate. It is insufficient to depend entirely on scientists to stipulate a metaphysical assumption of a kind. Also, to only refer to the original stability condition when categorizing a potential kind is erroneous.

To sum up, from the first strand of issues, it becomes pertinent that there first should be an elucidation of the relation between the properties. From the second stand of issues, it has been emphasized that there should be some kind of metaphysical account of grounds that stabilize property clusters. The positive account that I describe in the next section mitigates the worries that are raised.

3.1. The metaphysical model

The big picture is this: Natural kinds are property clusters. But they are not arbitrary property clusters; natural kinds are property clusters whose natural kind status is metaphysically grounded by local networks that are proper parts of the total network (N) which is the maximal possible network between properties. Local networks like N ₁ , N ₂ , … N _n are subnetworks of the total network N. N consists of properties as nodes and many coinstantiation probability relations, Rs, as edges. A natural kind K is nothing but a certain cluster of properties [ϕ]Footnote ¹ that is associated with a natural kind predicate “ϕ” and metaphysically grounded by a local network N ₁ .

I would like to give further details about the properties in N, and Rs, which are held between the properties.

To recap, the properties I am talking about are those that are connected to natural kind terms. For instance, in the case of the natural kind gold, the properties that I am considering will be those that are connected to predicates that complete a statement like “The members of the kind gold is _____.” Some prominent examples of the properties include gold’s atomic number, its color, luster, malleability, density, conductibility, melting point, function, and size (e.g., all lumps of gold are less than one-mile diameter). Also, I follow Khalidi’s strategy in understanding the properties being at the type level but taking the nodes of his causal network model as representing token-level properties for a more intuitive understanding of the model.

As mentioned before, R is a coinstantiation relation between properties (at the type level). Here, I describe the features of R as I conceive it to ground the epistemic fruitfulness of natural kinds. First, R is a scientifically relevant relation that connects most of the scientifically relevant properties. I use the term “scientifically relevant” to mean that something is scientifically relevant when it is scientifically observable and is a part of scientific endeavors. Second, when two properties are connected by R, it means that they coinstantiate with varying degrees. For instance, the property of breastfeeding the baby has a higher degree of coinstantiation with having wombs than having an orange coat with black stripes. Third, R is a multiple-place relation and this is not so surprising when considering that many properties coinstantiate with many other properties at the same time. Fourth, R ranges over possible worlds. Fifth, R is an internal relation that is dictated by the nature of the properties it connects.

The fifth feature, that R is an internal relation is related to the notion of “blueprint” coined by Williams (Reference Williams2019). A blueprint is a coding system that dictates with which properties the properties interact. In other words, the internal nature of the properties determines with which other properties they will coinstantiate. This answers the question of why a property p has a certain R with q and not with s. The answer is that what the properties intrinsically have is metaphysical grounding for the clustering of the properties. I argue that it is a fundamental fact of the world that p has R with q. Imagine a plastic ball model of atoms and pretend that they are models of properties. Some plastic balls have little holes in them in different intricate shapes that only allow certain types of sticks to fit. And there will be a type of ball that has the same shape of holes. Though this explanation may not be satisfactory, I believe it provides at least some metaphysical explanation for how properties relate to one another.

At this point, I’d like to specifically explicate what my original contribution is compared to that of Slater’s and Khalidi’s. Firstly, one might inquire how my theory is different from the SPC account. The first answer is that R holds between property types, whereas in the SPC account the probabilistic relation holds between property clusters. Second, not only does R hold between the properties and the SPC account’s probabilistic relation holds between clusters, but R also has a broader reach than the latter in the sense that R’s domain is possible worlds, which means that R represents modal relations between properties. If the degree of R is 0, then there is no possible world where p and q coinstantiate together, meaning that they are metaphysically impossible. If the degree of R is 1, then it means that the properties are connected by metaphysical necessity. R could also be a mere possible relation, for example there is nothing metaphysical that bars t and s from coinstantiating with one another.

One will notice how my theory overlaps with Khalidi’s network theory that argues that natural kinds are nodes in the causal network of properties. Similarly, as I described in the preceding text, in my theory, there is a network of properties that have a coinstantiation relation R to one another. How does my theory differ? The relation R that I stipulate can accommodate coexistence relations that are not explained by causal relations. This is a significant contrast to Khalidi’s relation which only accepts causation between actual properties that have a causal hierarchy. Khalidi concedes that some fundamental properties like the spin of particles coexist with other fundamental properties. My contribution to stipulating R is more appealing because it includes exception cases to the causal relation.

3.2. The epistemic model

One might worry that my theory does not inherit certain upsides of the SPC account like flexibility, due to the fact that in my model, kinds are determined by an objective R between properties. In the SPC account, domain-relative goals and interests motivate domain experts to set specific probabilistic thresholds through counterfactual perturbations. Counterfactual perturbations are the epistemic processes in which domain experts test hypotheses under many different counterfactual scenarios. Slater describes the robustness condition as a set of truths having maximal invariance under “counterfactual testing”; stable sets have the same members throughout the test, so to speak. Likewise, the (cliquish) stability of a property cluster represents the cluster’s invariance under different counterfactual scenarios, meaning the members of the property cluster do not change throughout the test. The degree of stability that domain experts observe through counterfactual perturbations will determine which degree of threshold they will stipulate for the probabilistic relations between subclusters and the whole cluster. This is how the goals and interests of a domain epistemically ground the stability of natural kinds and thus procure flexibility and domain relativity in SPC kinds.

My theory assuages the worry that the kinds that I stipulate are not flexible enough. To explain that rigidness is not the case and to conceptualize the epistemic practice of natural kind-seeking, I introduce the term “constellation” as coined by Williams (Reference Williams2019). Constellation captures the effect that the goals and interests of a domain have on determining what kinds of “patterns” we see in a huge network of properties. We can “see” constellations in the sense that epistemic factors like goals and interests determine the threshold for the probability of R between p and q. In other words, a researcher can spot a certain natural kind that is a property cluster that consists of properties that have a certain degree of R to one another. She might hold a low threshold for the “strength” of the relation between p and q; let’s say any R that is over .7Footnote ² is accepted by her. Then she sees a certain constellation of properties that make up a natural kind. This is logically consistent with another researcher in a different domain spotting a different constellation of roughly the same properties, having a threshold of .8 for R. According to the real pattern theory by Dennett (Reference Dennett1991), two different people can look at the same placement of black and white squares and see two different patterns; one sees a complex pattern with less noise, and the other one sees a simpler pattern with more noise. The difference relies on the epistemic goal of the observers. Analogously, depending on the various kinds of goals of a domain, researchers can delineate natural kinds with different properties in the network.