8 The inferentialist view of natural kinds

In this chapter, I begin to unpack the various elements in my definition of Natural Kinds with a Human Face (NKHF), which I stated in Chapter 7 as follows:

Natural kinds are (i) historically identified and open-ended groupings of modally robust phenomena, (ii) each displaying lawlike dependencies among relevant features, (iii) that enable truth-conducive conditionals-supporting inferences over time.

My attention in this chapter is on (i) historically identified and open-ended groupings of modally robust phenomena. Phenomena do not make up kinds via any part–whole relation. A natural kind is not just any set of phenomena. Only interrelated phenomena in a well-defined historically identified grouping qualify as natural kinds—though they form an open-ended, malleable and revisable grouping, under the view I articulate.

I see our encounter with natural kinds as akin to how evidence is gathered to make inferences about what might have been the case in a detective story. If we knew from the beginning who the criminal was and how the events happened, we would not need any forensic science and there would be no story either. Similarly, if we knew from the beginning the natural joints of nature, we would not need the natural sciences, and all the experimenting, model-building, theorizing that go with them. That is why I see our encounter with natural kinds as navigating our ways in the space of possibilities as a guide to actuality. This is something human beings have learned to do over millennia. It is the historical-cultural achievement of epistemic communities working within and across scientific perspectives on a variety of modally robust phenomena. This chapter lays out the inferentialist story underlying NKHF.

Thinking of natural kinds is first and foremost thinking about what a bunch of things have in common. Consider two examples from botany and chemistry. The eighteenth-century Dutch botanist Jan Frederik Gronovius classified with the name of Linnaea borealis flowers that had a distinctive Y-shaped stem, campanula-like petals, a white or pink colour, and whose geographical distribution we now know ranges from the mountains of Alaska to the Dolomites. Hugo Erdmann classified as ‘noble gases’ those with the least chemical reactivity. And Ramsey’s discovery of argon forced Mendeleev to add a special column for them in his periodic table.¹

Thinking about natural kinds as ‘given’, part of nature’s furniture, has engendered a metaphysical exercise of asking which properties are constitutive for each kind. One can think of flowers as coming with defining morphological properties (say, a Y-shaped stem, particular shapes for their foliage, petal colour), the properties one would find listed in a pocket-size guide to Alpine flowers. Similarly, chemical elements come with the properties systematized in the periodic table, with atomic numbers, electronic arrangements, and associated chemical properties. Particles in physics equally come with defining properties such as their mass, their electric charge, and spin values.

That is how one often learns about natural kinds. School textbooks imply that we live in a world of properties and some of them define the difference between angiosperm and gymnosperm, alkalis and bases, vertebrates and invertebrates, and so on. Such kind-constitutive properties are usually regarded as essential properties of the kind. This tradition begins with Aristotle and his theory of predicates, with how he saw genus–species relations originating from the way adjectives like ‘rational’ vs ‘irrational’ are brought to bear on nouns such as ‘animal’, demarcating the division between, say, man qua ‘rational animal’ and the rest of the animal kingdom, as Aristotle saw it.

In contrast, I see natural kinds as the outcome of humankind’s scientific and cultural history. We come to historically identify them among a bewildering array of empirical regularities in nature. They are the end products of concerted efforts of generations, who have successfully identified relevant groupings of phenomena in nature. But what counts as a relevant grouping?

One can imagine Gronovius going about collecting specimens of Linnaea borealis on his mountain treks and observing some properties (‘one Y-shaped-stem white flower, two Y-shaped-stem white flowers, three Y-shaped-stem white flowers, . . . ’) and on that basis concluding ‘All Y-shaped-stem white flowers belong to the natural kind Linnaea’. But this is a caricature. For the minimal units of identification for natural kinds cannot be properties—too many to count, and too diverse for kind identification. Is a slightly more pinkish petal still passable as a Linnaea borealis? What about other flowers with Y-shaped stems?

Considerations of this nature have traditionally pushed discussions of properties from observable macroscopic properties to unobservable microstructural essential properties—one is here reminded of Putnam’s discussions about jadeite vs nephrite, molybdenum vs aluminium. One would not define water as a colourless transparent liquid, but instead as having a certain microstructural composition as a molecule with two atoms of hydrogen and one of oxygen.² And for Linnaea borealis, looking at the genome might give a more clear-cut definition of kindhood than looking at the colour of the petals or shape of the stems.

However, as indicated by isomers and the story of DENDRAL and chemoinformatics in Chapter 7, there is more to chemical compounds than just atomic composition. Topological considerations about structure and bonds are equally crucial. And AI-led techniques help to differentiate among isomers in a swarm of possible combinations of atoms, situating engineered kinds in a continuum with our most familiar natural examples.

Here, then, is an alternative approach to natural kinds. The minimal units of classification for natural kinds are not properties but phenomena, as I have defined them. To identify Linnaea borealis, one has to be able to distinguish a variety of phenomena first. Some concern leaves and stem-shapes: this is the task of plant morphologists. Others concern the geographical distribution of the plant (is the Linnaea of the Dolomites the same kind of plant as the similar one found in Alaska?)—a task for phytogeographers. The various chemical processes going on in the plant fall under the remit of phytochemists, while the relation between the plant and its biome is a specialty of plant ecologists. Equally important are phenomena concerning how poisonous the plant might be, whether it might be usable for medicinal purposes. Here, often knowledge of local communities proves important (be they the Ladin community of Süd Tirol for Linnaea, the Scottish Gaelic communities of the Hebrides with their knowledge of seaweeds, or the Malagasi community for the rosy periwinkle, or any other similar examples—I shall return to the latter two in Chapters 10 and 11, respectively).

In what follows, I use the expression ‘local knowledge’ in a specific sense following Suresh Canagarajah (1993, 2002): namely, to denote knowledge that is ‘context bound, community specific, and nonsystematic because it is generated ground up through social practice in everyday life’ (Canagarajah 2002, p. 244).³I shall expand on this point in Section 8.3 and show its relevance to my discussion of NKHF.

To know the natural kind Linnaea borealis involves grouping a variety of phenomena very different in nature. Each community uses different data as evidence for their phenomena. Hence the data-to-phenomena inferences are perspectival as each epistemic community resorts to data, experimental techniques, modelling resources, and methodological-epistemic principles that belong to their own situated practices.

That human practices condition how we carve the world’s joints has become common currency since the work of Hacking (1991, 1999, 2007a) and Dupré (1981, 1993). More recently, human epistemic practices behind each natural kind have been put centre-stage by Kendig (2016a) in what she calls ‘kinding’ processes, and by Reydon (2016), who refers to the ‘co-creation’ of categories by merging empirical properties with human cognition. Ludwig’s practice-dependent kinds have further strengthened this approach by looking at examples in ethnobotany and ethnozoology (see Ludwig 2017, 2018a, 2018b, and Ludwig and Weiskopf 2019). And Bursten’s work on kinds in nanotechnology (2016, 2018, 2020a) has offered further material for reflecting on what natural kinds really are. In what follows, I clarify how I see the relevance of this scholarly tradition to the perspectival realist view and where the differences lie.

Traditionally, appeal to a variety of epistemic communities at work in parsing natural boundaries has often been combined with brands of realism: realism about individuals, or entities, or dispositional properties, combined with pluralism about taxonomic classification. Dupré’s (1993) promiscuous realism combines realism about individuals with promiscuity in the semantic cross-classification. Hacking’s experimental realism (1991, 2007a) combines realism about entities with nominalism about kinds.⁴ Chakravartty’s defence of what he calls ‘sociable properties’ combined with ‘manifestation-based pluralism’ is yet another realist view that in this case takes the properties of scientific interest as dispositional.⁵ For these different realist views, one can be a realist about tigers, lemons, hellebores, or electrons in believing that all these entities or biological individuals or their dispositional properties are ‘out there’ in the world.

Perspectival realism, while sharing the pluralism of other varieties of realism, eschews discussions about properties, causal-dispositional roles, or ways of packaging properties together. But how, then, to make sense of the idea of natural kinds as historically identified groupings of modally robust phenomena?

Consider again chemical elements. Identifying elements as natural kinds involves being able to identify phenomena such as atomic spectra, chemical reactions, melting and boiling points, oxidation, and so forth. Each of them typically occurs in a particular domain: spectra are the fingerprints of atoms; oxidation is a chemical reaction; melting and boiling points mark phase transitions. All these phenomena are modally robust in that epistemic agents infer what could, would, or should happen. What would happen to the atomic spectrum if the sodium atom were placed in a weak magnetic field? How does oxidation occur in iron? In which way can atmospheric pressure affect the boiling point of water? Examples multiply endlessly.

This shift from properties to phenomena-first reflects the central role played by epistemic communities occupying a plurality of situated perspectives in the identification of groupings of phenomena candidates for natural kinds: the variety of scientific practices they engage with, the inferences drawn from them and the phenomena they accordingly model and encounter. We encounter kinds by identifying relevant groupings of phenomena. They are as historical as our historically and culturally situated perspectives are. Can an account of natural kinds that centres on epistemic communities and perspectival vantage points qualify for the label ‘realism’? What will it take to replace the marble-solid metaphysical foundations of natural kinds with something that looks as flimsy as Neurath’s Boat?

That Boat is indeed the inspiration for my inferentialist view of NKHF. As Nancy Cartwright, Jordi Cat et al. have argued, ‘What propelled Neurath was an idea: the idea not simply that our stock of knowledge claims keeps on changing forever, but that a decisive revision of our concept of knowledge is required if reason is to fulfil its Enlightenment promise’ (Cartwright et al. 1996, p. 92, emphasis in original). Neurath’s anti-foundationalist programme in philosophy was a reaction both against the presumption of first foundations (pace Descartes and Kant) and against ‘the unbridled relativism supposedly encouraged by the absence of foundations’ (pace Spengler) (p. 136).

Neurath’s suggestive metaphor was put to varied uses.⁶  Quine (1969), for example, adopted it as an emblem of naturalism in epistemology. But there is more to the metaphor’s aptness for the view of natural kinds canvassed here than mere naturalism. The metaphor points to the importance of communication among epistemic communities as a guard against methodological solipsism (and Spengler’s type of relativism).⁷ Replacing metaphysical foundations with a view of scientific knowledge that gives epistemic agents their due has long been central to a family of philosophical views that style themselves as ‘inferentialist’.

Take Brandom’s inferentialism, where in ‘calling what someone has “knowledge” one is doing three things: attributing a commitment that is capable of serving both as premise and as conclusion of inferences relating it to other commitments, attributing entitlement to that commitment, and undertaking that same commitment oneself. Doing this is adopting a complex, essentially socially articulated stance or position in the game of giving or asking for reasons’ (Brandom 1998, p. 389, emphases in original). Or consider Huw Price’s project of rethinking the notion of representation in language (Price et al. 2013); or Richard Healey appealing to Brandom’s inferentialism in his interpretive reading of quantum mechanics as mentioned in Chapter 6 (Healey 2017).

In philosophy of science, Mauricio Suárez (2004) has spearheaded an influential inferentialist account of scientific representation whereby ‘a source s represents a target t only if (i) the representational force of s points to t and (ii) s allows an informed and competent agent to draw specific inferences regarding t’ (Suárez 2004, p. 773). And Suárez (2015a, 2015b) has offered a sustained defence of this inferentialist approach to scientific representation against alternative approaches that have often emphasised a range of relations holding between the source and the target system (from isomorphism to similarity) to explain scientific representation.

Surprisingly, though, discussions about natural kinds have so far escaped the inferentialist turn. I aim to remedy that, in pursuit of perspectival realism. Here, I begin to unpack the NKHF strategy by going back to its first notion, that natural kinds are (i) historically identified groupings of modally robust phenomena (I will have more to say about their being ‘open-ended’ in Chapter 9). The next section offers examples for rethinking the metaphysics of natural kinds as downstream from the epistemology of science—as aids for navigating the Neurath’s Boat of natural kinds.

If the ‘naturalness’ of natural kinds is not derived from the joints where nature is carved, how should one understand it? If we can have an engineered synthetic DNA with the potential of reproducing DNA’s main properties, what does this teach us? I think it teaches us how to think of naturalism in an anti-foundationalist and historicized way.

The assumption that kinds track natural divisions in nature often precedes debates about realism or nominalism about kinds. Quine advocated a minimal naturalism whereby natural kinds are the scientific, discipline-specific outcomes of what he saw as our ‘innate subjective spacing of qualities’ (Quine 1969, p. 126). His main challenge was to explain how innate standards of similarity ‘have a special purchase on nature and a lien on the future’ (p. 126). Quine went Darwinian in his answer: ‘[S]pacing that has made for the most successful inductions will have tended to predominate through natural selection’ (p. 126). His Darwinian approach was inspired by Neurath:

I see philosophy and science as in the same boat—a boat which, to revert to Neurath’s figure as I so often do, we can rebuild only at sea while staying afloat in it. There is no external vantage point, no first philosophy. . . . For me then the problem of induction is a problem about the world: a problem of how we, as we now are (by our present scientific lights), in a world we never made, should stand better than random or coin-tossing chances of coming out right when we predict by inductions which are based on our innate, scientifically unjustified similarity standard. Darwin’s natural selection is a plausible partial explanation. (Quine 1969, p. 127)

Natural selection seems a plausible but only partial explanation, in my view. Cognitive neuroscience and developmental psychology show how children learn the concepts for object-kinds following specific constraints. For example, developmental psychologist Ellen Markman and collaborators (Markman 1989, Ch. 5; Markman and Hutchinson 1984; Markman and Wachtel 1988) have studied how children learn the meaning of a new word from a single labelling event, guided by two main constraints: the ‘whole object constraint’ and the ‘taxonomic constraint’. Children aged 18 to 24 months have a preference for the word to refer to the whole object (rather than parts of the object or attributes of the object—the rabbit, rather than the rabbit’s ears or tail); by age 3 to 4, children prefer to generalize the meaning of the new word to taxonomically similar objects (e.g. rabbits, mammals, animals) rather than thematically related ones (e.g. rabbits, carrots, and burrows). The developmental advantage of doing so is clearly expressed by Markman (1989, p. 111):

By expecting unforeseen nonperceptual properties to be common to members of a kind, children could go beyond the original basis for grouping objects into a category and discover more about the category members than they knew before. Children might start out assuming that categories will have the structure of natural kinds. With development, they would then refine these expectations, limiting them to properties, domains, and category types that are appropriate.⁸

But Darwinian considerations are only a partial explanation. For a gulf separates the basic similarity standards at work in, say, children’s pre-scientific classificatory abilities from the complex and sophisticated taxonomies of scientific disciplines. Distinguishing, say, brown edible mushrooms from brown poisonous ones is one thing. Plotting scientific taxa is quite another. The University of Oslo’s Nordic mycological herbarium features 14,695 currently accepted names of fungi (taxa and genera).⁹ Accounting for this exceedingly complex system must call on our cultural–scientific history, which Quine’s naturalism did not pay attention to. A distinctive variety of historical naturalism is at work in my Neurathian approach to kinds. Our subjective spacing of qualities has a purchase on nature only insofar as situated epistemic communities historically learn how to classify relevant and open-ended groupings of phenomena into evolving kinds. But how should this historical naturalism be characterized?

Appeal to historical considerations is not entirely new. As already mentioned, the historicity of natural kinds is emphasized by philosophers of biology reacting against ‘eternal natural kinds’. Ruth Millikan (1999) has objected to the Kripke–Putnam view and argued that biological kinds are identified not by properties but by their lineages. These lineages are traceable to a clade having a common ancestor. Other philosophers of biology have appealed to the notion of homologues (see Ereshefsky 2012) to explicate phylogenetic continuity with a common ancestor. Many physical and chemical kinds can also be regarded as historical in some relevant sense. Muhammad Ali Khalidi, for example (2013, pp. 139ff.), has observed how the history of many chemical kinds (say, gold) coincides with the history of our universe and the formation of such elements inside stars.

A common theme of some of these approaches is the identification of historicity with causal history. Biological lineages are causal lineages of mating, breeding, and survival-adaptive evolution to a changing environment. Similarly, the causal history of our universe since the Big Bang—interspersed with supernova explosions, and the formation of stars and galaxies—underpins the historicity of chemical elements formed inside the stars.

But there is another, equally important, non-causal sense in which natural kinds are historical. Their historicity is also the outcome of how real epistemic communities across a plurality of situated scientific perspectives have come to historically identify a group of modally robust phenomena as candidates for natural kindhood. This kind of historicity goes to the heart of human epistemic practices.¹⁰ It is not about locating a natural kind within a causal network of events, a phylogenetic lineage, or similar. But it is about the natural kind being a historically identified grouping of phenomena that different historically and culturally situated epistemic communities have encountered over time. Historical naturalism so understood offers then an answer to the question left open by Quine’s naturalism: why is it that we seem so good at navigating nature and encountering ‘functionally relevant groupings’?

In previous work (Massimi 2014), I sketched an answer through a version of naturalized Kantianism. The idea behind it is that natural kinds

latch onto stable empirical clusters evinced by robust experimental data, i.e., observable records of occurrences that cannot be ascribed to error or background noise. From the pre-scientific ability of children to cluster objects with same empirical properties (pears with pears, apples with apples), to the mineralogist’s ability to cluster minerals, it is our human ability to identify and track recognisable patterns of empirical properties in nature that gave us the upper hand in the evolutionary gamble. Peaks in magnetometers, sparks in scintillation counters, bubble trails in cloud chambers that have proved genuine (i.e., not due to background noise or experimental error) are the sophisticated scientific counterpart of children and laymen’s pre-scientific clustering ability. (p. 427)

What I then called ‘recognisable patterns of empirical properties in nature’ should have been better characterized as the modally robust phenomena evinced through perspectival data-to-phenomena inferences.¹¹ How any specific grouping could become eligible for the title of natural kind is a question ultimately for scientific practitioners. And as I stressed (p. 428), ‘one would need to tell a very detailed, discipline-specific and context-specific story’ for each of those kinds.

In the following two chapters, I look into this in more detail. But in the rest of this chapter, I focus on how historical naturalism explains the need for a variety of scientific perspectives to contribute to natural kind classifications. The perspectival pluralism at work in modelling practices is an aspect of a more general epistemological pluralism in ways of knowing. This plurality of ways of knowing has far-reaching consequences for realism in science. In particular, it calls attention to the key role of situated knowledge in natural kind classifications, under the Neurathian view of NKHF.

The emphasis I have placed so far on technological, experimental, and modelling resources to reliably advance claims of knowledge should not be misunderstood. Scientific perspectives—as I have been using the term—are not akin to Kuhn’s paradigms: the Newtonian perspective, Lavoisier’s chemical perspective, and so on (more on this in Chapter 11). For I have stressed all along the social, collaborative, inferential nature of the epistemic exercise that underwrites perspectival realism. In what follows, I return to this point and highlight a few implications for the view of natural kinds I put forward:

Ethnobotany, and ethnotaxonomy more in general, are a good illustration of historical naturalism at work. Among the varieties of local knowledge, traditional ecological knowledge plays a special role when it comes to natural kind identification for plants and animals. The United Nations defines ‘traditional knowledge’ as ‘the complex bodies and systems of knowledge, know-how, practices and representations maintained and developed by indigenous people around the world, drawing on a wealth of experience and interaction with the natural environment and transmitted orally from one generation to the next’ (UN 2019a, p. 2/13).

The recent reappraisal of traditional knowledge so understood comes after a long history of denigration (see Canagarajah 2002), erasure, and appropriation. (I shall return in more detail to the topic of epistemic injustices in this particular context in Chapter 11.) Institutional efforts led by the UN, among others, to tackle this endemic problem led in 1992 to the establishment of the Convention on Biological Diversity, whose Conference of the Parties,

in its decision XIII/18 adopted the Mo’otz Kuxtal Voluntary Guidelines . . . for the development of mechanisms, legislation or other appropriate initiatives to ensure the ‘prior and informed consent’ . . . of indigenous peoples and local communities for accessing their knowledge, innovations and practices, for fair and equitable sharing of benefits arising from the use of their knowledge, innovations and practices relevant for the conservation and sustainable use of biological diversity, and for reporting and preventing unlawful appropriation of traditional knowledge. (UN 2019a, p. 5/13)

Indeed, in the 1992 Convention on Biological Diversity (CBD), article 8(j) the UN affirmed the need to

respect, preserve and maintain knowledge, innovations and practices of indigenous and local communities embodying traditional lifestyles relevant for the conservation and sustainable use of biological diversity and promote their wider application with the approval and involvement of the holders of such knowledge, innovations and practices and encourage the equitable sharing of the benefits arising from the utilisation of such knowledge, innovations and practices.

The 1992 UN text mentioned ‘indigenous and local communities’, and the acronym IPLC (Indigenous People and Local Communities) has entered the ensuing literature on CBD¹² and wider discussions at the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (IPBES).¹³ Mulalap et al. (2020, p. 2) have clarified the acronym along the following lines:

“Local communities,” unlike Indigenous Peoples, do not necessarily have a history of being invaded or colonized by external entities. However, like Indigenous Peoples, local communities have cultural values, practices, and systems developed through multiple generations and poised to be passed to future generations. . . . We acknowledge, however, that conceptualizations of indigeneity are contested and highly context-specific.

With this important caveat in mind, the acronym IPLC has entered the legal and scientific literature on biodiversity to refer to holders of traditional knowledge broadly construed. It is against this institutional and legal backdrop that in what follows I urge for the need to expand the classical remit and boundaries of the literature on scientific perspectivism and realism too. My aim is to clarify the mechanisms through which a plurality of historically and culturally situated perspectives leads to knowledge production that is always inherently local and perspectival, and enables the relevant local epistemic communities to identify modally robust phenomena which are candidates for NKHF.

There is one particular aspect that is instructive about situated knowledge and my story here on historical naturalism. I am going to call it the fine-graining and coarse-graining of descriptions of natural kinds. The historical identification of groupings by real epistemic communities depends on their localized epistemic practices and needs. Coarse-graining or fine-graining means that, without necessarily changing the membership of the open-ended grouping (i.e. without necessarily adding or removing specific phenomena from it), one might nonetheless give a description of it that zooms out or zooms in on specific phenomena. Zooming in on specific phenomena involves varieties of situated knowledge that are often a prerogative of perspectival local practices.

Take the following botanical example. There is a species of melliferous flora in Mexico called Gymnopodium floribundum. It belongs to the family of buckwheat (Polygonaceae). It is one of the most common plants in the Yucatán peninsula and its Mayan name (and the name of the honey produced from it) is Ts’íits’ilche’. In the Kew Royal Botanic Gardens herbarium, specimen records for this plant come from Belize. In other databases, such as the National Centre for Biotechnology Information (NCBI), the same species is classified on the basis of a group of proteins and nucleotides. And if we switch from NCBI to the Biodiversity Heritage Library, we find a list of relevant bibliographical references where the role of the plant in ethnobiology is included.¹⁴

It would be thus restrictive to think that the historical classification of this species comes down to the identification of a specimen (along the lines of Putnam’s ‘archetype’) like the one preserved at Kew Gardens. Several concurrent perspectival phenomena have historically entered into the identification of the natural kind known in Western botany as Gymnopodium floribundum. Some are morphological and phenotypic: these concern the comparative analysis of the anatomy of the plant, its reproductive organs, and development, for which herbaria specimens are a useful source of knowledge.

Others concern nucleotides and protein groups inferred from the perspective of cytogenetics. Yet others are karyological phenomena: phenomena concerning the structure of cells and chromosomes. Biodiversity and ecosystem phenomena are also important: these are phenomena concerning the inter-relations among living organisms in a certain environment.

For example, particularly interesting are the plant–pollinator interactions that pollination ecology studies. It has long been known that such plant–pollinator interactions are highly context-dependent and that ‘the degree of specialization within a study system can depend not only on the perspective of interest (plant vs. pollinator) but also on the community context’ (Rafferty 2013/16). Let us then take a quick look at the role of the community context when it comes to identifying the relevant pollination phenomenon for this particular plant of the Yucatán.

Its flowering season peaks between February and April and the plant has traditionally played a key role in the api-botanical cycle. Rain and high temperature make the plant blossom in the spring months and its nectar proves a vital resource for the local bee species (see Quezada-Euán 2018, pp. 195–196). Beekeeping and honey production have historically been an important element in the Maya culture. Archaeological evidence of beekeeping goes back to the late Pre-Classical Maya period (see Crane 1999, p. 295). Local species of stingless bees, including Melipona beecheii, made hive keeping a popular practice among the Maya. Honey production in specific locations, known as meliponarios (see Bratman 2020), continue to be an integral part of the local economy of the communities of Xmabén, Hopelchen, and Campeche in Mexico (see Coh-Martínez et al. 2019).

It is the situated knowledge of these local communities about the api-botanical cycle rotating around the nectar of the plant that has the epistemic upper hand when it comes to identifying particular phenomena such as pollination peak, for example. In my philosophical idiolect from Chapter 6, the pollination of Gymnopodium floribundum is a modally robust phenomenon which involves (a) a stable event (i.e. the transfer of pollen from the anther to the stigma); (b) data that provide evidence for it (i.e. the buzzing of the bees around the scented flowers, the fruity clusters); and (c) the historically and culturally situated epistemic communities teasing out a network of perspectival inferences from the data to the stable event in question. The situated knowledge of local beekeepers is an unrivalled source of information for identifying the particular phenomenon of pollination peak in a way that plant morphologists, or cytogeneticists cannot offer. Let us see why.

The pollen being deposited on to a stigma is a stable event in that it follows lawlike dependencies that are independent of there being or not being any epistemic community or perspective. An example of lawlike dependencies at play in the pollination of flowers is pollinator performance, defined as the product of flower coverage (FC)¹⁵ and pollen deposited (PD).¹⁶ Recall from Chapters 5 and 6 that lawlikeness plays the realist tether in perspectival realism. It grounds a first-tier modality at play in, for example, whether a flower would be pollinated if a pollinator were to visit it (depending on the pollen being deposited).

In turn, a phenomenon is a stable (qua lawlike) event whose occurrence can be inferred in many different possible ways. In this example, the phenomenon pollination is modally robust in embedding the very many ways in which this stable event of pollen transfer might occur. For example, honeybees perform this act by carrying pollen on their legs; wasps may carry it in their mandibles; flies in their abdomen. And of course teasing out all the ways in which this is done by various pollinator species and the impact it has on pollination peaks and flowering seasons for individual plant species is something studied by pollination ecology, entomology, among other disciplines. Modal robustness expresses the many ways in which epistemic communities infer the relevant phenomenon by connecting often diverse datasets to the occurrence of the stable event in question.

Different pollinators perform differently, more or less efficiently, and the presence of insecticides in nearby crops significantly affects the number of local honey bees in each particular area and the associated process of pollination. For example, melissopalynology studies the variety of pollen and pollen sources present in particular samples of honey. In so doing, it provides insights into the percentages and varieties of flower nectars visited by honey bees. Melissopalynological studies in the region of the Yucatán have found that Ts’íits’ilche’ is under-represented at ca. 3% among the single-flower honeys of the region, despite the plant being common there. This finding has in turn suggested that Gymnopodium pollen production must be lower than that of other varieties of melliferous flora in the region (see Alfaro Bates et al. 2010, p. 60). An explanation for this under-representation might be sought in the reproductive biology of the plant.

Local communities and their situated knowledge play an integral role in trying to understand and explain this finding. Beekeepers know best how to protect their apiaries across seasons; how to control insecticides that have devastating effects on bees; and when the nectar peak for the local plants is so as to sustain honey production throughout the year. It is by virtue of their being historically, geographically, and culturally situated that local epistemic communities know best about the phenomenon pollination peak: they know how to identify this modally robust phenomenon among a swarm of stable events. This example of local situated knowledge about a phenomenon (call it P_k) enables in turn other epistemic communities (e.g. plant morphologists) to investigate related phenomena P_j (e.g. about the reproductive biology of the plant and the possible causes for the pollen under-representation in honey).

In my philosophical lingo, the Gymnopodium taxon is all these phenomena. The situated knowledge of different epistemic communities—melissopalynologists, beekeepers, plant morphologists, pollination ecologists, etc.—makes it possible to fine-grain or coarse-grain the description of the taxon by focusing on one phenomenon rather than another. For example, plant morphologists can describe the reproductive organs of the plant, but to gain insight into its reproductive performance, one needs to fine-grain the description at the level of the pollination peak. And it is here that the local knowledge of beekeepers and honey producers has the epistemic upper hand in better understanding what might be causing the under-representation of Ts’íits’ilche’ pollen in the honey of the region as spotted by melissopalynologists.

Under historical naturalism, the identification of relevant groupings of phenomena qua candidates for NKHF is humankind’s collective historical and epistemic achievement, something that geneticists at NCBI, botanists at Kew Gardens, and beekeepers in the communities of Xmabén, Hopelchen, and Campeche equally share and can reclaim as their own. This is what historical naturalism is ultimately about: a celebration of the social and cooperative nature of scientific knowledge where a variety of perspectival situated practices by specific epistemic communities at particular historical, geographical, and cultural locations get intertwined in delivering knowledge of natural kinds.

Each epistemic community contributes one or more phenomena to the grouping. And the open-ended groupings that the natural kinds get identified with can always be fine-grained or coarse-grained by focussing attention on one or more particular phenomena and associated descriptions. But it would be a mistake to conclude on this basis that therefore the natural kind Gymnopodium floribundum should be primarily or even exclusively identified with the herbarium specimen or the nucleotide sequence as if these gave us some privileged handle. It would equally be hasty to reach the opposite conclusion and defend some form of conventionalism about this natural kind. The different epistemic practices are not an invitation to pick and choose a phenomenon P₁ at perspective sp₁ instead of a phenomenon P₂ at perspective sp₂, depending on specific needs.

Historical naturalism does not make natural kinds social constructs. The phenomena that communities have learned to identify over time and across perspectives are as real as the effect of the dry season on the melliferous flora of the Yucatán; as tangible as the specimen at the Kew Gardens herbarium; as reliably inferred from data as nucleotide sequences held at the NCBI database.

There is nothing ‘constructed’ about these modally robust phenomena. Human construction is of course involved in creating instruments, making machines, and devising methods through which modally robust phenomena are inferred from a variety of data and get eventually historically identified as belonging together. But this does not in turn license the metaphysical conclusion that phenomena and their groupings are themselves a human construction.¹⁷

Going back to engineered kinds, their effective naturalness is neither surprising nor mysterious on this account. The epistemic ability to historically identify relevant groupings of phenomena explains why our subjective spacing of qualities has a purchase on nature, no matter how engineered hachimoji DNA or DENDRAL-aided molecules might be. Thus, recalling Lesson no. 1 from Section 7.3 in Chapter 7, the naturalness of kinds is not just the product of our ‘subjective spacing of qualities’ (as Quine maintained) but it is also the result of our perspectival scientific history, where a genuine plurality of scientific perspectives—broadly understood—have historically defined the boundaries of what we think of as a ‘natural kind’.

Yet the discussion so far is incomplete. What still remains to be done is to offer a philosophically more detailed account of how an identified group of phenomena with relevant features and lawlike dependencies become eventually a natural kind.

All natural kinds are born as in-the-making kinds. This might seem a bold assertion. Surely, either something is a natural kind or it is not. The way in which our beliefs, representations, or conceptions change should not affect basic metaphysical facts.

Yet, given the perspectivalist stance against the ‘view from nowhere’, starting from somewhere, namely local and situated epistemic practices, this is inevitable. Perspectival realism deepens the tradition that has emphasized the role of human practices in natural kind classifications as a way of keeping metaphysics in check. What are natural kinds if not a presumption humankind makes about nature and its possible joints given the robust phenomena one gets perspectivally acquainted with?

Some in-the-making kinds survive the ongoing and never-ceasing inferential work within and across scientific perspectives. These in-the-making kinds become evolving kinds resilient across scientific perspectives. They are effectively the ‘natural kinds’ we know and love. Other in-the-making kinds do not survive: they prove to be empty kinds. Nomological resilience plays an important role in the ability of in-the-making kinds to survive and become evolving kinds.

The natural kinds that we know and love evolve with our scientific history. Hence, a perspectival realist who takes seriously the situated nature of knowledge does not take natural kinds as placeholders for clusters of essential properties (discovered or still to be discovered), causal powers, categorial properties, dispositions, or similar. Natural kind is the name we give to what makes our presumptions of ‘natural kindhood’ for in-the-making kinds a little less presumptive.

In this final section, I return to the role of laws of nature and clarify why they provide the realist tether for natural kinds. Let me, then, go back to the second element in my definition of NKHF: that is, the reference to modally robust phenomena ‘(ii) each displaying lawlike dependencies among relevant features’. In what sense do phenomena manifest such dependencies?

As we saw in Chapter 5, perspectival models are an exercise in physical conceivability. They are an invitation to imagine something about the relevant target system that complies with the state of knowledge and conceptual resources of a community C and is consistent with the laws of nature known by C. As the analogy with inferential blueprints suggested, perspectival models involve by and large an inferentialist exercise of physically conceiving guided by the laws of nature and with an eye to delivering modal knowledge of what might be the case.

The laws endorsed by a particular epistemic community support the open-ended exercise of modelling what might be the case. I have discussed the various ways in which laws of nature enter into physical conceivability: by driving analogical reasoning between different modelling practices; by enabling non-causal explanations; or by fixing the general nomological boundaries within which the physical conceivability exercise takes place.

I also stressed the difference between lawhood and lawlikeness. The former is contingent on whichever series of perspectival Best Systems we happen to work with. The latter is displayed in nature among specific features of phenomena underpinning the stability of the event and not contingent on there being a perspectival Best System in place.

The events that are candidates for phenomena are ‘stable’ precisely because they display lawlike dependencies. The stable event associated with the phenomenon ‘cathode rays bending’ is the expression of the lawlike dependency between the electrical nature of cathode rays and the way they respond to a magnetic field. The stable event (structural loop in the kinase regulatory region) associated with the phenomenon ‘phosphorylation’ is the expression of the lawlike dependency between the ability to carry phosphate molecules and changes induced in proteins. The stable events (melting of glaciers, ocean heat uptake, etc.) associated with the multifactorial phenomenon of ‘global warming’ are all associated with various lawlike dependencies concerning increased GHG and associated retention of incoming radiative energy.

These lawlike dependencies can be causal, as with the electrostatic force that causes the bending of cathode rays; or the addition of phosphates to proteins. Other relevant dependencies are non-causal, as with Pauli’s principle constraining nucleon structures. At yet other times, the lawlike dependencies fix general constraints within which the exercise of perspectival models takes place, as when R-parity conservation and consistent electroweak symmetry breaking are involved in the pMSSM-19 to physically conceive ways in which hypothetical candidate SUSY particles might manifest themselves (see ATLAS Collaboration 2015 and Section 5.5.3 in Chapter 5).

Thus, the nomological resilience of in-the-making kinds is a good indicator that the groupings of historically identified phenomena have (causal or non-causal) lawlike dependencies among relevant features. The relevant features might be empirical properties that manifest themselves in the relevant phenomena. I call them ‘empirical’ to avoid confusion both with the ‘sparse natural properties’ of Lewisian memory and with the metaphysical properties of the dispositional realist or dispositional essentialist. The charge-to-mass ratio is an example of lawlike dependency between two empirical properties at play in the phenomenon of cathode rays bending. The kinase inhibitors are another example of lawlike dependencies (well studied in pharmacology and drug discovery) among empirical properties (e.g. activation loop) of particular molecules at work in the phenomenon ‘phosphorylation’.

At other times, the relevant features are not empirical properties but measurable quantities (e.g. the coefficient of viscosity for a fluid, the potential gradient, soil pH). Or they are physical or chemical constants (the Planck constant h, the elementary charge e, various thermal constants for inorganic, organic, and metallo-organic substances). These are just some illustrative and non-exhaustive examples.

But natural kinds talk is not so much concerned with specific lawlike dependencies among relevant features indexed to a particular domain. One wants to find out what is common to groupings of phenomena across different domains. For example, one might be interested in finding out whether the lawlike dependencies at play in the phenomenon of Moon–Earth alignment and the times of the tides are related to the lawlike dependencies observed among the speeds of different kinds of balls rolling down inclined planes. Or whether the phenomenon of the stability of matter at the level of stars is related to the phenomenon of the stability of matter at the subatomic scale. Or whether the lawlike dependencies observed in water electrolysis have anything to do with those in the bending of cathode rays. Or whether the phenomenon of pollen under-representation in Ts’íits’ilche’ honey is related to other phenomena concerning, for example, the reproductive biology of the plant.

Are these stable (lawlike) dependencies identified in a number of phenomena, each indexed to a particular domain, indicative of how these phenomena might somehow belong together? Depending on how one answers, the in-the-making kind might turn out to be an empty kind, or prove to be an evolving kind. Identifying which grouping of phenomena speaks to an evolving kind and which one hides an empty kind is something that epistemic communities learn how to do over time through a network of perspectival inferences, which I will have to return in more detail in the next chapters.

The ether exemplifies a former in-the-making kind that turned out to be an empty kind. A number of phenomena with seemingly lawlike dependencies were identified by Newton, Boerhaave, and Kant, among other natural philosophers of the eighteenth century. The putative kind ‘ether’ was introduced to encompass a wide-ranging array of phenomena including bodies repelling each other;¹⁸ calcination in chemistry;¹⁹ and even the formation of planets in the solar system,²⁰ among others. Yet there is no natural kind ‘ether’ in this grouping of phenomena across different domains. Although there were lawlike dependencies underpinning each one, there was no genuine bona fide inferential link connecting them. The phenomena of metallic filings floating on liquids, calcination of metals, and formation of the solar system have nothing in common. It took almost a century and a half to downgrade the ether to an empty kind.

Consider, on the other hand, the grouping of phenomena ranging from water electrolysis to cathode rays bending, to black-body radiation. The lawlike dependencies at play in each of these phenomena were this time indicative of how this particular grouping of phenomena did in fact belong together under what—to the best of our knowledge still today—we have reasons for believing to be one of our evolving kinds: the electron. In Chapter 10, I shall return in detail to the nature of the inferential links among phenomena in this historical episode.

In sum, how epistemic communities come to identify which phenomena group together and which ones do not is not a matter of happenstance or convention. As I argue in Chapter 9, successful groupings of phenomena typically satisfy a sort-relative sameness relation. But for now, the main points to stress are the following:

That there are features and lawlike dependencies in phenomena qua stable events is a fact about nature, and ultimately it is the realist tether of perspectival realism. The perspectival pluralist aspect is that situated epistemic communities are able over time to rely on these and engage with epistemic practices and perspectival models—qua inferential blueprints—that allow them to tease out the relevant conditionals-supporting inferences linking genuine groups of phenomena into evolving kinds.

But how do we possibly come to know the world as is if all that is ever given to us is the world as it appears to epistemic communities over time? The question has a genuine bite. If our scientific knowledge is always situated knowledge of phenomena, no matter how reliable in the ways I describe, it still feels like something is amiss. The world as it appears to us is never going to be the world as is, unless we pass phenomena off as noumena, a critic might reply.

I need to show that our modelling—and more broadly epistemic—practices and their associated groupings of phenomena deliver indeed perspectival₂ representations, which offer a genuine window on reality, despite the situated nature of each representation. Let me give you a specific example which I shall return to in Chapter 10. Take J. J. Thomson’s (1897) perspectival₁ data-to-phenomena inference about cathode rays bending, Max Planck’s (1906/1913) perspectival₁ data-to-phenomena inference about black body radiation, and Theodor Grotthuss’s perspectival₁ data-to-phenomena inference about water electrolysis around 1805–1806. This particular group of identified phenomena, over a span of a century, enabled conditionals-supporting inferences that still underpin our knowledge claims today about the electron as a natural kind, one of our best examples of evolving kinds. Evolving kinds are analogous to the unbounded space reflected by the mirror in the Arnolfini Portrait, a space that extends beyond what is visible on the canvas, beyond the perspectival₁ representation.

How can this work? If there is no God’s-eye view from which one can access natural kinds as given, how can there be a God’s-eye view from which to access relevant features in phenomena, and their lawlike dependencies, as somehow indicative of whether a grouping of phenomena belongs together? Are not any such groupings of phenomena ultimately at the mercy of historical contingencies?

J. J. Thomson, doing experiments on cathode rays and X-ray ionization in the late 1890s, was still couching his findings in the idiom of ‘carriers of negative electricity’, ‘corpuscles’, and even ‘Faraday tubes’, harking back to the nineteenth-century view of a strained state of the ether (see Falconer 1987, p. 260). My Neurath’s Boat of natural kinds begins to look alarmingly leaky. How can NKHF resist the ever-present stresses and strains of historical changes and scientific revolutions?

The worry is genuine, but it conceals once more an invitation to hold epistemology in check, by offering a metaphysically more solid tether of some sort. Yet this fails to appreciate the epistemological pivot behind NKHF: the third condition in my definition—how lawlike dependencies among relevant features (iii) enable truth-conducive conditionals-supporting inferences over time.

I will spell out the full details of this point (iii) in the next two chapters, which explain the nature of the inferential exercise that joins the dots among types of phenomena and allows epistemic communities to navigate safely in the Neurath’s Boat of natural kinds. These inferences supporting indicative conditionals over time lead epistemic communities to the identification of the relevant links among lawlike dependencies in different phenomena across different domains (see Figure 8.1).

Subjunctive conditionals in turn act as signposts in inviting epistemic agents to walk in the inferential garden of forking paths. Ultimately, albeit always provisionally, they tell us which among the very many features and lawlike dependencies present in each type of phenomenon are related to which others across a multitude of perspectives. The next two chapters take a closer look at the remaining steps in this journey navigating the Neurath’s Boat of NKHF.