1. INTRODUCTION

Following the lawsuit brought by James Maurice against Samuel Judd, a curious trial took place in December 1818. The plaintiff, a dock inspector of the New York’s harbour, claimed a commission for conducting the inspection of the defendant's whale oil. Under New York State law, all fish oil had to be inspected. Judd refused to make the payment, claiming that whales are not fish, and therefore their products cannot be included as fish oil (Burnett, 2007).

Beyond historical interest, this case shows us that scientific classifications have consequences that exceed science itself. In fact, it is easy to find examples of the effects of scientific classifications at multiple levels. Perhaps one of the most obvious is the classification of psychiatric disorders, exemplified in the different versions of the Diagnostic and Statistical Manual of Mental Disorders (DSM). Various factors are involved in its classification (Regier et al., 2013; Browne, 2017), although clinical aspects certainly prevail. Classifying in psychiatry has relevant consequences in diagnoses and treatments (Zoellner et al., 2013). In addition, classifying in one way or another may involve the allocation of social and economic resources, and have legal consequences, such as incapacitation and modification of criminal responsibility (Filone et al., 2014; Scott, 2015).

Classifying is an important activity, not only for scientists but also for all those who may be involved in the consequences of using one or another classification system. As taxpayers, patients or users, this means that we are all affected to a greater or lesser extent by scientific classifications. Many of them justify their validity in their objectivity, according to the real structure of nature, regardless of the interests and values of humans. This objectivity and independence is what allows us to appeal to them when establishing rational judgments and decisions. Considering this, it is pertinent to ask: According to what criteria a classification reflects the real structure of nature?

One answer to this question is to consider that scientific classifications are objective insofar as they refer to natural kinds, that is, those kinds that exist in nature independently of humans. Chemical elements, for example, exist regardless of whether humans know about them or not. The existence of a real essence justifies the naturalness of such kinds. With some variations, this is the answer given in some formulations of natural kinds, originally attributed to Putnam (1975) and Kripke (1972).

This view of natural kinds has shown limitations, especially in disciplines like biology and social sciences. Key issues include its inadequacy in addressing relational and historical aspects of certain kinds, varied classification criteria across disciplines, and the problematic status of non-epistemic components. The second section examines alternative approaches by scholars like Boyd (1999), Dupré (2002), and Ereshefsky and Reydon (2015, 2022), which address these shortcomings. The third section will argue that "metaphysically weakened” approaches fail to justify the metaphysical independence of natural kinds, as they, like intrinsic essentialism, maintain a division between the real and the conventional.

It is also important to consider that not only metaphysical commitments, but also other types (epistemic, ethical or political) are taken into consideration in scientific classifications (Dupré, 2002; Magnus, 2012; Brigandt, 2020). However, the role of each of these commitments is controversial. This problem, as it will be shown in the third and fourth sections, is one of the reasons for dispute among different proposals, as well as the origin of the skepticism defended by some authors. This paper aims to show that the problem is not in the prevalence of metaphysical, epistemic or other commitments. Metaphysical, epistemic and even axiological, ethical or political commitments can be distinguished, but in scientific kinds they appear inextricably linked. Scientific kinds are global entities, understood as interdependent sets.

Once the context in which the debate on natural kinds unfolds has been outlined, Section 4 aims to show why it is advisable to abandon the concept of natural kind. The problem is not only how to define natural kinds. Beyond this issue, an undesirable metaphysical commitment underlies natural kinds (Hacking, 2007). The concept of natural kind leads to a problematic separation between the truly natural components (real and objective) and the conventional elements of natural kinds. This idea is the origin of the problem in order to consider scientific classifications.

Finally, based on the conclusions drawn in sections 3 to 5, it is intended to offer an alternative to natural kinds, proposing the concept of scientific kind in its place. In this concept the contextualization of each category in its context would prevail. Categories would not be considered as metaphysically independent, but as contextual ways of understanding reality, based on certain commitments of different types (metaphysical, epistemic, axiological, political or ethical).

2. NATURAL KINDS: THE HISTORY OF A LACK OF CONSENSUS

The first proposals on natural kinds (Kripke, 1972; Putnam, 1970, 1975) were characterized by their strong metaphysical commitment. Although there are differences among them, they share some fundamental features. Thus, let there be an individual x, a natural kind K and an essential property P. It is said that ∀x, x ∈ K if and only if x possesses the essential property P. In case there is more than one essential property, ∀x, x ∈ K if and only if x has $\sum _1^n{P}_i$, where i is each one of the essential properties.

Essentialist proposals with a strong metaphysical commitment soon found some problems. Few entities meet these strict requirements. In any case, biology and social sciences entities would be excluded. A first reason is that, in most cases, the boundaries of the kinds are not clearly drawn. Species limits, for example, are often hard to define, as is evident in cases such as plant species hybridization. The boundaries of some natural kinds in social sciences also turn out to be vague, as is the case with social classes (Khalidi 2013). Besides, it is important to consider that natural kinds in biology and social sciences are historical. If biological species evolve over time, then attributing fixed essential properties to them makes little sense. Insofar as essences are fixed and immutable, fundamental ideas such as the theory of evolution have no place within essentialist frameworks (Boyd, 1999).

In order to solve the difficulties found by essentialism, other authors propose a double strategy. First, they weaken metaphysical commitments, overcoming difficulties such as the integration of fuzzy cases. Second, they consider that natural kinds are not determined exclusively by nature, but also by the interests and values of researchers. Probably the most influential author within this line is Boyd (1999), with his theory of natural kinds as homeostatic property clusters (HPC).

In contrast to essentialism, these theories consider that natural kinds are not defined by one (or many) necessary and sufficient property (or properties). Thus, the relationship between a kind and its properties is not analytical, but homeostatic. According to this, if a series of properties in F are considered, “the presence of some of the properties in F tends (in the right conditions) to favor the presence of the others, or there are underlying mechanisms or processes that tend to maintain the presence of the properties in F, or both” (Boyd, 1999, 143). On the other hand, the metaphysical commitment in Boyd’s HPC account is grounded in mechanisms. Mechanisms ensure that natural kinds are genuinely natural, rather than mere conventions.

The change introduced is important, since if naturalness is not given by necessary and sufficient properties, then it is possible to account for complex situations in which limits are diffuse. Such is the case of those studied in biology or social sciences. In addition, according to Boyd (1999), scientific research is accommodated (accommodation) to the causal structures of nature, and this accommodation can be instantiated in different ways, depending one very discipline. Accommodation is therefore relative to a particular field of study.

Furthermore, HPC theory is compatible with the historicity of natural kinds. In other words, there are no permanent essences, in any context, both from a synchronic perspective and from a diachronic one. Considering this, it is possible to solve one of the main obstacles that prevented considering natural kinds in biology, namely, their incompatibility with the modern evolutionary synthesis. Furthermore, this issue is relevant not only in biology, but in all sciences where kinds are historical. Some authors develop similar ideas, like Khalidi (2013), with some modifications regarding Boyd's proposal. In particular, while Boyd emphasizes homeostatic mechanisms as central to the stability of natural kinds, Khalidi broadens this view by focusing on causal mechanisms more generally, without requiring homeostasis. On the other hand, Slater (2015) contends that it is possible to consider natural kinds without any underlying mechanism, still maintaining the stability of the kind (a perspective exemplified by subatomic particles).

Dupré (2002), for his part, considers that there are no privileged classifications by virtue of metaphysical considerations. To the extent that explanations are relevant and useful in certain contexts, it is possible to consider different ways of understanding natural kinds. This even leads to the elimination of the metaphysical priority of scientific kinds over the rest. That said, it is interesting noting that Dupré explicitly considers his proposal as realistic, so that the classifications considered by a certain discipline are not mere conventions. There are multiple ways of classifying nature, and these classifications are context dependent. Notwithstanding, this does not mean that all classifications are equally valid, which justifies, in Dupré's opinion, the realism of his proposal.

The latest contributions on natural kinds take into account the presence of non-epistemic elements in scientific classifications (Brigandt, 2020; Reydon and Ereshefsky, 2022; Ludwig, 2023). According to Reydon and Ereshefsky, the proposals on natural kinds have not taken into consideration the relevance of non-epistemic aspects. Even in those cases where they are considered, such as Boyd or Slater, such aspects are external to the theories. The GFA (Grounded Functionality Account) theory, through its double commitment (with the functionality restriction and grounding condition), considers both epistemic and non-epistemic aspects as co-constitutive of natural kinds. Non-epistemic elements, therefore, are elements proper to the kind, not external to it.

As noted, the GFA is justified by a double commitment. On the one hand, the functionality condition considers that natural kinds must satisfy what they are postulated for. On the other hand, the grounding condition refers to “a relation of metaphysical dependence of a kind upon aspects of the world” (Reydon and Ereshefsky, 2022, 12). Both aspects are not independent and allow explaining a variety of situations where metaphysically weakened proposals find difficulties, without renouncing a certain metaphysical commitment. Thus, GFA is a proposal where metaphysical, epistemic and non-epistemic aspects are compatible.

In short, there is a wide range of theories, each with varying degrees of metaphysical, epistemic, and non-epistemic commitments. In any case, there are many proposals and little consensus on what natural kinds actually are.

3. PROBLEMS WITH THE CONCEPT OF NATURAL KIND

As it has been seen in section 2, metaphysical essentialism has been followed by a whole series of proposals, which try to respond to those problematic situations in which essentialism does not seem to provide an adequate solution. While it is possible to continue maintaining a hard-line essentialist stance, in that case the price to pay is to extremely limit the entities that can be considered as natural kinds.

The greatest difficulties found by Boyd (1999), and Khalidi (2013), have to do with the justification of the metaphysical independence of the mechanisms (homeostatic or causal, depending on the case). For these authors, mechanisms allow to say that natural kinds are real, beyond the interests and values of researchers. This justification is neither necessary nor sufficient, nor is it ahistorical. But, in any case, mechanisms are what justify realism, beyond our interests and values. Nevertheless, is this way of understanding mechanisms sufficiently justified? Do mechanisms respond to the expectations placed on them? As it will be seen, it is not the case.

First, it should be noted that Boyd's use of the term mechanism is not clearly specified. Beyond justifying HPC theory, Boyd explains little about it. This topic is relevant, insofar as there are different, and sometimes incompatible, ways of understanding mechanisms. Focusing on new mechanism theories, some authors define mechanisms in terms of entities and activities (Craver, 2009), systems (Glennan, 1996), or structures (Bechtel & Abrahamsen, 2005). Understanding mechanisms in terms of activities and entities is compatible with a more open idea of mechanisms than structures, which has consequences when considering natural kinds.

Arguably, the most substantial critique of Boyd’s theory is offered by Craver (2009). According to this author, in natural kinds as HPCs, causal mechanisms are understood as those elements that justify metaphysical commitments. Of course, natural kinds as HPC allow different classifications of the same entities, as seen in section 2. But there is a metaphysical commitment articulated through mechanisms. This commitment is specified through the metaphysical objectivity and reality of mechanisms. And just as importantly, there is a separation between such mechanisms and the interests and values of researchers.

However, according to Craver this position is not sufficiently justified. Mechanisms involve conventional elements, in a way that cannot be eliminated or separated: “I argue that conventional elements are involved partly but ineliminably in deciding which mechanisms define kinds, for deciding when two mechanisms are mechanisms of the same type, and for deciding where one particular mechanism ends and another begins” (Craver, 2009, 575).

Let's look at some examples of what Craver means. The strategies of splitting and lumping allow relating classifications used in a scientific discipline with the mechanisms associated with them. The idea is simple: if the properties of a kind are explained through different mechanisms, then it is convenient to divide the kind into lower order kinds (splitting). In the same way, if different groupings of properties are explained under the same mechanism, then it is the same grouping (lumping).

Correlations, especially in complex cases, can be explained by different types of mechanisms, some involving division and others joining. The point is that on certain occasions, researchers opt for one strategy or another depending on the interests at stake. This, apparently, is not problematic, to the extent that Boyd (1999) accepts that different disciplines use different classification criteria, based on certain interests and values. The problem is that mechanisms are what ensure the metaphysical commitment, that is, the independence of the values and interests of the researchers. Therefore, it is contradictory to use mechanisms to justify the metaphysical independence of the natural kinds when mechanisms themselves are not independent.

A related objection addresses the varying levels of abstraction applied to the same entity. Craver illustrates this with the study of hippocampus, showing that the chosen description strategy —whether detailed (like in surgery) or generalized (as in computational studies)— depends on disciplinary goals. This means that researchers' interests shape the mechanisms used, making purely mechanical explanations insufficient to justify metaphysical independence.

In a similar way, Darden (2002) reinforces this by arguing that researchers' interests inevitably influence how mechanisms are defined and decomposed, as mechanisms function across different levels. Selecting the relevant level is a research-driven choice, not determined by nature. While sociological studies rarely look at molecular details, boundaries between levels can be unclear, especially with multilevel relationships.

Moreover, proposals about metaphysically weakened natural kinds (Boyd, 1999, Dupré, 2002) allow for the incorporation of aspects of different elements (epistemic, axiological, ethical, or political) into the concept of natural kind. In fact, the trend in recent years is to incorporate such aspects (Brigandt, 2020). However, authors such as Reydon and Ereshefsky (2022) criticize that such incorporation considers non-epistemic aspects as external to theories (this would be the case of Boyd or Slater). It has already been pointed out that this is problematic, insofar as such aspects are co-constitutive to natural kinds, not external elements to it. In fact, this problem is in line with what is criticized here, the radical separation between what is real and what is conventional.

As seen in section 2, Reydon and Ereshefsky's proposal incorporates non-epistemic elements as co-constitutive of natural kinds. However, according to Ludwig "there is a clear tension between Ereshefsky and Reydon's case for a naturalist methodology and their ambition to formulate a unified account of natural kind" (Ludwig 2023, 4). In Ludwig´s opinion, Reydon and Ereshefsky's proposal is too restrictive since it prioritizes certain aspects in detriment of others. This is especially evident in social disciplines, such as ethnobiology. For Reydon and Ereshefsky, a folk classification that includes fish and whales would not be a natural kind. This is because the functionality restriction is met, but the same is not true of the grounding condition. This supposes prioritizing a certain metaphysical commitment, which refers to the difficulties indicated in the previous proposals.

Finally, the proposals that dispense with any metaphysical commitment have not been analyzed yet. However, these proposals also face some difficulties. The most important of them has to do with the justification of its realism. Dupré's proposal is explicitly realist but rejects any metaphysical commitment. But according to authors such as Kendig and Gray (2020), Barberousse et al. (2020), or Lemeire (2021), the realism of natural kinds supposes assuming some kind of metaphysical commitment (whatever it is). Thus, to the extent that authors such as Dupré insist on the realism of their proposals, it is justified to ask according to what they are realistic. These questions have not been answered by Dupré. Without justifying this metaphysical commitment, promiscuous proposals differ little from conventionalism (Wilson, 1996). This led Dupré (2018), in light of the criticisms of his original proposal, to elaborate on the metaphysical justification of his position by drawing more explicitly on the concept of process (though whether this justification is ultimately sufficient remains open to debate).

4. SKEPTICISM ABOUT NATURAL KINDS: WHY LEAVING THE CONCEPT?

In the current situation there is a wide variety of proposals, with different levels of metaphysical, epistemic and non-epistemic commitments. Besides, it does not seem that this high number of proposals has reached a minimum consensus. For example, certain key aspects of some proposals often pose significant challenges to their compatibility with others: intrinsic essences, for instance, are difficult to reconcile with relational approaches such as the HPC theory; likewise, the absence of metaphysical commitments tends to be unconvincing for those who defend natural kinds without explicitly adopting a promiscuous stance.

For this reason, it is not surprising that some authors have modified their position over time, from an explicit defense of natural kinds to clearly skeptical positions. The cases of Hacking (2007) and Brigandt (2020) are perhaps the most notable. Thus, in his first works, Hacking (1991) defended a pluralistic and realistic idea of natural kinds. Subsequently, Hacking (1999) continues to defend the validity of natural kinds, as long as he distinguishes between natural kinds and human ones. In line with the metaphysical lack of definition of promiscuous proposals such as Dupré's, the criterion that allows separating both kinds, natural and non-natural, is not clear.

However, later Hacking (2007) leaves the concept of natural kind. The proliferation of incompatible theories and the impossibility of reaching a minimum agreement is reason enough to consider that the concept has self-destructed and is obsolete. Hacking does not deny the greater naturalness of some classifications compared to others, but what it contests is the very concept of natural kind: “Some classifications are more natural than others, but there is no such thing as a natural kind” (2007, p. 203).

For his part, Brigandt (2011) is originally favorable to a pragmatist conception of the concept of natural kinds. To this author, naturalness is a matter of degree, so that it can be said that there are more or less natural kinds. In this case, natural kinds are no longer understood as something precise. An interesting example of this is the one offered by Brigandt (2011) regarding functional kinds. Traditionally, functional kinds have been understood as non-natural kinds. This is the case of many of the kinds typical of ecology, such as predator. There are multiple instances of this, whose common link is that functional property object of study. The important thing is whether such kinds justify generalizations and explanations. To the extent that they can do so, there is no reason to regard such kinds as unnatural.

However, Brigandt (2020) turns towards positions openly critical about natural kinds. In this case, unlike Hacking (2007), the main problem with the concept of natural kind is the neglect of relevant aspects beyond metaphysical and epistemic commitments. Using examples primarily from the social sciences, Brigandt emphasizes the importance of non-epistemic aims in our practices of classification. He discusses gender as a socially significant category shaped by ethical, and political factors, demonstrating that such categories can be valuable without relying on the traditional notion of natural kind.

It is true that criticisms about natural kinds are not new. In fact, Goodman (1978) substitutes the idea of naturalness for that of relevance. What is important about Hacking and Brigandt is not only that they criticize the proposals about natural kinds, but also that this position has been the consequence of an evolution starting from a defense of them.

In this context, it is worth asking if, as critical authors affirm, it is appropriate to leave the concept of natural kind. In this sense, the explanations given by Hacking (2007) and Brigandt (2020) are insufficient. It is true that both authors show difficulties with the concept of natural kind, but neither objection is by itself sufficient to invalidate natural kinds. In the case of Hacking (2007), it could be argued that the proliferation of proposals does not necessarily invalidate the concept of natural kind (MacLeod and Reydon, 2013).

In the same way, to solve the problems pointed out by Brigandt (2020), relevant non-epistemic aspects could be incorporated, although this would force the concept of natural kind. On the other hand, even accepting his criticisms, it could be pointed out that most of the objections seem more consistent in the context of the social sciences than in other cases. These objections would show that natural kinds are not appropriate in social sciences, but not necessarily in other disciplines.

Up to this point, it has been shown that the various interpretations of natural kinds have been the target of several criticisms. There is, therefore, no minimal consensus on how the concept of natural kind should be understood. This paper aims to show that one of the most significant reasons for this lack of agreement is something that is rarely made explicit, namely: all natural kind proposals assume a conceptual framework in which natural kinds are justified because they are capable of explain reality independently of the interests and researchers' values, at least partially. This separation between the real and the conventional is considered in different ways. But in any case, this is the framework on which all proposals are based. Assuming this separation is, from the point of view defended in this paper, the main problem faced by natural kinds and for which there seems to be no solution. Let's see what this implies in each of the following cases: essentialism, homeostatic property clusters, and promiscuous proposals.

According to essentialism, there is a single, objective reality that can be described independently of the interests and values of researchers. The primary goal of science, under this view, is to uncover that reality. Essences are what justify the objectivity and independence of scientific classifications. For its part, the HPC account is based on a different metaphysical foundation than intrinsic essentialism. The metaphysical commitment of natural kinds in the HPC model is grounded not in essences, but in mechanisms. Despite this important difference, it shares a common conceptual framework with essentialism, namely, the idea that there is a clear distinction between natural reality and human interests and values. In sum, essences or mechanisms justify metaphysical independence of natural kinds.

However, are such assumptions valid? In previous paragraphs we have already seen that there are many difficulties in this regard, not only for the case of intrinsic essentialism, but also for metaphysically weakened proposals. The conventional and the real appear as inseparably joined aspects, as has been shown in the case of mechanisms. The choice of splitting or lumping strategies, or the specification of abstraction levels, are just two examples of conventional aspects inescapably mixed with mechanisms. On the other hand, although the metaphysically weakened proposals incorporate epistemic elements, the non-epistemic ones are either left aside or are considered as something external. This issue, as Ludwig (2023) points out, is a source of problems, to the extent that the presence of non-epistemic elements is co-constitutive of scientific classifications.

Finally, promiscuous proposals must be considered. In this case, by explicitly rejecting metaphysical commitments, it seems that a different conceptual framework is being assumed. That is to say, it could be thought that the rigid separation between the real and the conventional is not so strict in this case. However, this question seems to contradict itself if we consider the explicit realism of authors such as Dupré (2002). If these proposals are realistic, according to what criteria are they so? Are epistemic and non-epistemic criteria sufficient to justify the realism of these proposals? This does not seem to be the case, at least if some of the criticisms already mentioned are taken into account (Kendig and Grey, 2020). Promiscuous proposals solve the problem without presenting a metaphysical justification. Without such a justification, what distinguishes promiscuous proposals from anti-realism and conventionalism? In short, promiscuous proposals are incomplete, because they are not sufficiently justified, or they are indistinguishable from anti-realist proposals.

Thus, the tension between metaphysical and conventional aspects is maintained in all proposals. And this is so because the concept of natural kind implicitly carries an undesirable metaphysical burden (Hacking, 2007; Brigandt, 2011). The problem is not the existence of metaphysical commitments but considering these commitments as something separated from the conventional aspects. Considering this, it is possible to ask a question, namely: is there any alternative that could replace the concept of natural kind, without falling into conventionalism?

5. AN ALTERNATIVE TO NATURAL KINDS. SCIENTIFIC KINDS

In section 4 it has already been pointed out that according to Hacking (2007) there are more natural ways of classifying than others. On the other hand, there are well-founded reasons to leave the concept of natural kind. However, in line with what was pointed out by Ludwig (2023), leaving natural kinds can be problematic, if philosophical reflection on scientific classifications is left out. Therefore, it is not enough to point out that natural kind is a problematic concept. It is also necessary to offer an alternative.

Some authors have proposed changing the term natural kind, avoiding the problems arising from its use. According to Crane (2021) the term natural kind could be maintained in the philosophy of language, while in philosophy of science it should be replaced by scientific kind. Ereshefsky and Reydon (2015) also refer to the term scientific kinds, within the criticism made to homeostatic property clusters. The term scientific kind does not have the implicit metaphysical burden that natural kind does. That is why it is the one that will be used here. In any case, the terminological issue is secondary. It is the content of the proposal is what is relevant. So, what is meant here by a scientific kind?

In a first approximation, it could be said that a scientific kind is a relevant disciplinary cluster, which allows explanations, inductions and/or projections to be made, according to the validity criteria considered in the discipline at stake. It is important to highlight that the validity criteria can include diverse elements: metaphysical, epistemic, axiological, ethical or political. In fact, to a greater or lesser extent these different elements can be found in almost all disciplines. Thus, such clusters are made up of a variety of elements:

Each one of these elements fits to a type of relevant commitment (a_n, for example, could refer to elements of a metaphysical type, b_n, epistemic, and so on). These elements can be studied individually, of course. In fact, scientific practice usually considers the variation of one component, keeping the rest constant. But the elements of the matrix must be considered as a cluster. On the other hand, each component has a value relative to its importance for the definition, in a contextual sense.

The concept of scientific kind, understood as a matrix composed of different elements, is stable as long as the elements that compose the kind are stable (in the sense used by Slater, 2015). However, the value of each of the components is not always the same. Regarding the concept of species in a biological sense, reproduction has an importance that does not have, for example, the occupation of an ecological niche (although this aspect may be relevant to the ecological concept of species). The value assigned to the occupation of an ecological niche could be null, depending on the case. It should be remarked that the importance of the different components is not absolute, that is, independent of the interests and values of the researchers.

In this way, referred to a scientific kind y, different components (or the same components but with different values) can be found, depending on the discipline. It would be the case of partial overlapping of concepts (for example, in the case of species or gene). The matrix, therefore, must refer to a disciplinary context n:

There are distinct situations when considering scientific kinds. For example, given the same scientific kind y, from a different discipline z, where the same components are considered, but with different values for each one of them, we have two different matrixes:

Another case would be one in which different components are taken into account for the same concept, depending on the discipline (for example, a₁ is maintained but instead of a₂, a different relevant component is considered, call it a'₂, or simply dispenses with it).

Once scientific kinds have been established as concepts composed of various components, it must be highlighted that they are not static. Their validity will be given by their assessment (positive or negative). Now, what is meant by assessment? To answer this question, one must return to the definition of scientific kind, that is, a relevant disciplinary cluster, which allows explanations, inductions and/or projections to be made, according to the validity criteria considered in this discipline. Therefore, what is evaluated is the ability to carry out relevant explanations, inductions and/or projections in each context. The adequacy of the components is contextual, although the presence of some form of metaphysical commitment is necessary to ensure that there is some connection to the actual structure of nature.

Thus, if doubts arise about specific aspects when evaluating a concept, then it is possible to reformulate it in order to adapt to the set of individuals subsumed in the scientific kind. It is also possible to keep (partly) different definitions for specific subsets. There may also be other cases, in which the negative evaluation is not critical. For example, different negative evaluations could be given simultaneously, none of them alone being determining, but as a whole. In short, kinds must be considered as contextual, while still maintaining a minimal metaphysical commitment to ensure a real relation to the structure of nature.

Now, this proposal seems to be an extension of the concept of natural kind, including the aspects indicated as relevant by Brigandt (2020). So, is this proposal limited to incorporating aspects not considered from the metaphysical and epistemic perspectives?

In order to answer this question, it must be considered that there is a fundamental difference between this proposal and natural kinds, namely: the presence of different elements (metaphysical, epistemic, axiological, etc.) takes place jointly and in an indivisible way. Although each of the elements considered in the kind can be separated, elements are interdependent. For example, if homeostatic (or causal) mechanisms are considered as a basic element that articulates biological species, it must be considered that it is not possible to separate the metaphysical aspects of mechanisms from the interest and values of researchers. Consequently, although metaphysical commitments are consubstantial to scientific kinds, there is no metaphysical priority over other commitments.

According to the proposal defended in this paper, different disciplines can use different species concepts and all of them could be valid in their context, as proposed by authors such as Boyd (1999), Khalidi (2013) or Dupré (2002). However, their concept of natural kind is different from the one proposed here. For these authors, there are multiple ways to classify reality, but between reality and conventional elements there is a metaphysical gap. It is true, as has been seen in section 4, that this is not always explicitly considered. But in one way or another, the separation between the real and the conventional lies at the root of the concept of natural kind. However, scientific kinds include conventional elements, inextricably linked to metaphysical components.

It has already been pointed out that Brigandt (2020) refers to kinds in social sciences to highlight the presence of aspects beyond the epistemic and metaphysical ones. However, although not so obviously, this same issue can be said about natural kinds in other disciplines, even in physics and chemistry. An example of what is meant can be considered through the study of chemical elements (see section 6).

This proposal includes the idea of Ereshefsky and Reydon (2015) that non-epistemic elements are an essential part of the kind. Thus, non-epistemic elements are an internal part of the matrix, like other elements. Now, here it is assumed Luwig's criticism regarding the excessively restrictive nature of natural kinds, as understood by Ereshefsky and Reydon. According to these authors, it would be difficult to find contexts in which fish and whales were part of the same group. This seems to prioritize a certain metaphysical commitment that is incompatible with disciplines such as ethnobiology (Ludwig, 2023).

Ludwig is correct in his criticism of Ereshefsky and Reydon. However, Ludwig's proposal is also considered insufficient. The reason for this is that it focuses largely on social sciences and related disciplines (ethnobiology, feminist theory), without going into detail on sciences such as physics and chemistry. The incorporation of non-epistemic elements is especially evident in social sciences, but, as will be seen, it is an inherent part of all scientific disciplines.

At this stage, it might be tempting to interpret the notion of scientific kind proposed here as a subtle form of conventionalism. After all, one could argue that there are multiple ways to classify nature, each valid within its own context. From this perspective, it might even seem that any discipline —scientific or not— could meet the criteria for scientific kinds, as long as its standards of validity are defined internally. However, it's important to recognize that the existence of various classification systems doesn't imply that all are equally valid. Their legitimacy is shaped not only by the interests and values of researchers but also by the underlying structure of nature itself. The possibility of making meaningful inductions and reliable projections depends on how well these classifications align with that structure. For example, it is not possible to make valid (non-trivially true) projections in disciplines like astrology (Tyson, 1984).

Thus, although scientific kinds incorporate epistemic, axiological, and even ethical or political considerations as inherent components, this does not mean that such aspects are the only ones taken into account. For a kind to qualify as a scientific kind, there must be at least a minimal metaphysical commitment. Otherwise, any grouping based on a given criterion would be considered scientific —even if that criterion bore no relation whatsoever to the actual structure of nature. While not all scientific disciplines share the same type of metaphysical commitments, all of them, in one way or another, do have them. As Ross (2023) points out, the way in which different disciplines articulate their metaphysical commitments varies. Therefore, imposing a single type of commitment across all scientific disciplines —such as essential properties or mechanisms— means embracing the difficulties discussed in sections 3 and 4.

6. SCIENTIFIC KINDS IN DIFFERENT CLASSIFICATORY CONTEXTS

At this point it is worth asking: to what extent do scientific kinds respond to the ways of classifying different scientific disciplines? This section will show how scientific kinds, understood in a contextual and matrix sense, consider the ways of classifying very different disciplines.

Chemical elements are paradigmatic examples of natural kinds (Hendry, 2011, 2015), although there are some authors who deny considering chemical elements as such (for example, Bhushan, 2006). In any case, if asked about the real structure of matter, few scientists would doubt that the periodic table offers an answer close to reality. On the other hand, the multiple representations of the periodic table are well known (Pyykkö, 2019; Maeno et al., 2021; Scerri, 2012, 2020). Regarding the elements as they are classified in the different periodic tables, there are two relevant questions that indicate to what extent not only metaphysical commitments, but also other considerations, are taken into account.

First, there is the question about the best way to represent the periodic table, that is, which table most accurately reflects the actual structure of matter? This is a question that is framed within what has been called here as metaphysical questions. Chemists with a conception closer to scientific realism would respond that the classification of the elements in the periodic table is objective and real. However, the high number of proposals shows different ways of classification of elements based on different points of view (Leigh, 2009; Scerri, 2012, 2020).

Although atomic number is the fundamental criterion for ordering elements in the periodic table, their arrangement can vary based on additional factors. Consequently, the position of certain elements may shift depending on the disciplinary context. For instance, in the version of the table used by solid-state physicists, both lanthanum and lutetium are placed beneath yttrium (Bosko & Chevary, 1993), a placement that differs from the one commonly accepted in other scientific fields. These authors highlight similarities in electronic structure within solid-state compounds as the basis for this arrangement. As Cao et al. (2021) point out, such variations are not limited to this example; alternative placements appear in other classifications as well. Because the position of elements changes according to different classification criteria, the resulting classifications of the elements are partially distinct.

In addition to the already mentioned examples from solid state physics, it is interesting to see how disciplines such as geochemistry or astrochemistry use modified versions of the periodic table. Considering geochemistry, Railsback (2003, p. 737) explains that the “conventional periodic table has not provided a good framework for understanding the chemistry of Earth and its life”. The criteria by which the chemical table is established are of little use in geochemistry. However, other criteria allow valid inductions and projections to be established in the context of geochemistry.

Among these other criteria Railsback (2003, p. 737) highlights the following: “One fundamental concept in rationalizing these geochemical patterns is the difference in bonding exhibited by hard and soft cations, which favor O^2- and S^2-, respectively.” These criteria, different from those usually considered in chemistry, are not mere conventions of geochemistry. On the contrary, these criteria allow valid projections and inductions to be made, in such a context.

Second, along with researchers studying the most realistic periodic table, there is some concern about bringing the periodic table closer to chemistry students (Leigh, 2009). This could lead to modify the traditional model of the periodic table considering educational purposes. This educational concern does not dispense with metaphysical commitments. In this case, both commitments, metaphysical and educational, are present. And what is more important, it makes no sense to separate metaphysical commitments from educational ones. Authors such as Bart (2021) point out the need to update the periodic table, focusing much of their argument on educational and social issues. Does this mean that it dispenses with metaphysical commitments? It doesn't seem that way; they are simply assumed implicitly.

So, is there an "ideal" periodic table, a unique correct way to classify the elements? The answer seems to be negative, which does not mean that all periodic tables are equally valid. Some can be clearly improved. But, at least partially, the differences among tables respond to distinct kind of commitments (metaphysical, epistemic, didactic…). The assessment of each table will therefore be based on the relevant criteria considered. In a periodic table where, for example, educational aspects are prioritized, explanatory clarity is an especially relevant element. In any case, the different commitments are interdependent, so it makes no sense to say that those tables that emphasize didactic aspects dispense with metaphysical commitments. In the cases where it is necessary to prioritize among different aspects, clarity of exposition prevails over other issues.

Another example where it is easy to see how metaphysical, epistemic, axiological and ethical aspects overlap is in the classification and conservation of species. Authors such as Mace (2004, 717) point out that: “taxonomists and conservationists need to work together to design some explicit rules to delimit the units included as species for the purposes of conservation planning and assessment”.

On the other hand, Coates et al. (2018) point out some difficulties in the interaction between conservation and classification. Although regulations are flexible enough to cover complex situations (species, subspecies, etc.), there are others, such as hybridization and intraspecific genetic variations, that complicate the interaction between taxonomy and conservation. In any case, the role of stakeholders in the taxonomy is controversial, with conflicting views (Garnett and Christidis, 2017; Raposo et al., 2017; Thomson et al., 2018), as well as attempts to try to reconcile the different proposals (Conix et al., 2021). This question has also been considered in detail by Reydon and Ereshefsky (2022), who study the convenience of using one or another definition of species for nature conservation.

As in the case of periodic tables, the species concept is discipline-dependent. Traditional debates about the concept of species (biological, ecological, phylogenetic, etc.) focus on aspects that we could call metaphysical. But the approaches of the different disciplines are given, at least partially, by their different interests and points of view. This is even more evident if we consider aspects such as conservation or biological diversity. Therefore, the concept of species cannot separate metaphysical commitments from epistemic and non-epistemic ones. There is not once and forever definition of species, but rather different contextual definitions that partially overlap.

These examples illustrate how scientific classifications incorporate different elements (epistemic, metaphysical, ethical, or political) to the extent that the different proposals of natural kinds tend to describe different commitments as independent not only in social sciences, but also in disciplines such as biology and chemistry. Of course, this article constitutes only a preliminary exploration of the concept of scientific kind, which must be further supported and refined through detailed analysis of concrete cases of scientific classifications.

7. CONCLUSION

The current situation of natural kinds, with a proliferation of incompatible proposals, poses a challenge to philosophers of science. This situation has led some of them to rethink the very concept of natural kind. Following this, in this paper has been shown that the concept of natural kind is not adequate since it inherently includes the idea of metaphysical independence. This is the reason why it is intended to introduce the concept of scientific kind. This concept would include different commitments (metaphysical, epistemic, axiological, ethical, and political) with different scope depending on the case. However, unlike the concept of natural kind, in this case the different aspects involved are not independent. In short, the different commitments are considered as a whole. That is what makes a difference between natural and scientific kinds.