A cluster concept is one that is defined by a weighted list of criteria, such that no one of these criteria is either necessary or sufficient for membership. Wittgenstein alleged that game was such a concept; some have claimed that species concepts are cluster concepts. (Homo sapiens was once defined as featherless biped, but there are human amputees who are not (actually) bipeds; moreover evolution may some day give us feathers (or analogues thereof).
It was at one time pretty commonplace to think that cluster concepts are quite common in science and culture. People say, for instance, that democracy is a cluster concept; Denis Dutton has recently argued that art is a cluster concept. I disagree. In my view, clusters are needed for what I call "diagnostic definitions": definitions that tell you how to recognize a concept-instance. But I would argue that despite the surface variability, there is an underlying unity in the sorts of concept that are robustly used in everyday and scientific discourse.
The classic articulation of this notion was of course Wittgenstein's in Philosophical Investigations I 66: "Consider for example the proceedings that we call 'games'. I mean board-games, card-games, ball-games, Olympic games, and so on. What is common to them all? -- Don't say: "There must be something common, or they would not be called 'games'" -- but look and see whether there is anything in common to all. -- For if you lok at them you will not see something that is common to all, but similarities, relationships, and a whole series of them at that." Some games involve cards, others boards, yet others balls; some involve winning and losing, others (like ring-a-ring-a-roses, or a child bouncing a ball off a wall) do not. "Look at the parts played by skill and luck; and at the difference between skill in chess and skill in tennis." And so on.
The classic refutation of this particular claim was due to Bernard Suits. He defined a game as an activity in which you accept certain rules that limit how you can achieve a certain activity. "To play a game is to engage in activity designed to bring about a specific state of affairs, using only means permitted by specific rules, where the means permitted by the rules are more limited in scope than they would be in the absence of such rules, and where the sole reason for accepting the rules is to make possible such activity." In golf the "specific state of affairs" aimed at is that a ball wind up in a hole: but you limit how you will achieve this end in order to play the game.
The difference between Wittgenstein and Suits is while the former looked at ways for an observer to identify games, Suits articulated an underlying commonality that is unobservable to a casual observer. Wittgenstein's definitions were diagnostic; Suits was after the real essence.
The same is, of course, true of species. The theory of natural selection demands that there be variability among members of a species, both at a time and through evolutionary history. This is an obstacle faced by those who would give a diagnostic definition of species (or what naturalists sometimes call a "key" -- the sort of thing you find in bird-watching books -- if you think a key is a diagnostic definition). But historical definitions of species, i.e., definitions that define species in terms of common origin, or definitions that define them in terms of reproductive isolation give unified definitions in terms of non-observables.
My thesis is that no cluster concept -- no concept primarily defined by a cluster -- plays an explanatory role in science or everyday discourse. (I would include social science under science, and aesthetics under everyday discourse -- so the scope of my thesis is quite wide.) I would spice up my thesis by adding: no non-explanatory noun-phrase plays any important role in scientific or everyday discourse. The "spicing up" is just a way of saying that you hardly ever come across a cluster concept. But this is an add-on.
Do people think that this thesis is obvious, vacuous, false, inflammatory . . . what?
My immediate reaction is that real systems are at least somewhat hierarchical and that a diagnostic (or any other phenomenal concept) can be explanatory for the next level "up" (or, more accurately "out"). Explanations accrete and whether something is a mechanism or a phenomenon depends on the resolution of the observer.
ReplyDeleteI agree that these things (species, art, ...) are not usefully understood as cluster concepts. So your position seems unobjectionable.
ReplyDeleteA question: Species concept pluralism is distinct from the claim that species is a cluster concept; the pluralist says that "species" as it is ordinarily used is ambigious between several different specific concepts. Are your grounds for rejecting cluster concepts compatible with pluralism?
Of course, Suits' example doesn't show that cluster concepts fail in general. Rather, if the example works, it just shows that 'game' isn't a cluster concept. But does Suits' proposed definition work? Two potential problems. It doesn't seem to give a necessary condition for 'game' because I could design a game one purpose of which is (for fun) to broaden rather than limit the scope of what Suits calls the 'means permitted'. Also, it seems that we could have games in which the rules are chosen for reasons additional to the 'sole' purpose of making the game possible. But maybe Suits defends against these worries?
ReplyDeleteI wonder, what is the evidence for the view that
there is an underlying unity in the sorts of concept that are robustly used in everyday and scientific discourse?
Perhaps cluster concepts help explain if there are true exception-prone generalizations or ceteris paribus laws. If there are such laws, the corresponding kinds whose members the laws quantify over would seem good candidates for being cluster kinds. But I think cluster concept advocates haven't yet argued for this. And not everyone like ceteris paribus laws.
Thanks for these useful comments.
ReplyDeleteMatt, you're right: Suits's analysis of game is only an example -- but a useful one because it illustrates the possibility that classes that are diverse with respect to diagnostic characteristics may nonetheless be unified at a deeper level
Does Suits's definition work? I haven't given it any deep thought, but it has a certain plausibility. (I took the definition from a 1966 PhilSci article, and he did work on it a lot more -- ultimately it came out in his book, The Grasshopper.) I don't think the "broadening" manoeuvre works: Suits's idea is that you take a goal like putting a ball in a hole, or a puck in a net, and then you restrict the means of achieving the goal. Implicitly, you are starting with all of the means available; that's why you can't broaden the scope. And the thought is that you restrict the scope -- you say "only with one of thirteen clubs, or with a stick of this size and shape" -- in order to make a game of it.
Anyway, what's the evidence that there is an underlying unity? I was thinking inductively. What is an example of a cluster concept, I wondered.
P.D. Magnus has a nice problem-case -- species. Not any particular species such as Canis lupus familiaris, but species itself. The question is this: if one accepts species-pluralism, would one then be forced to say that species is a cluster concept?
I guess my response is that it depends on how you formulate species-pluralism. One formulation is that there is no unitary species-concept; rather, there are several kinds of species. This formulation is not cluster-friendly if each kind of species is unitarily defined, and it is denied that there is any single overarching species-concept. Alternatively, one could try to take pluralism on-board by disjoining the kinds of species. I myself favour this approach. This is not a cluster-friendly approach either because it gives a necessary and sufficient condition, and isn't open-ended in the way that Wittgenstein suggests. So yes, P.D., there are ways of accommodating species-pluralism.
Finally, gepr: before I can comment, I would have to know more clearly what you mean by "whether something is a mechanism or a phenomenon depends on the resolution of the observer." On the face of it, I would disagree: if you take "played with cards" as an example, it's a characteristic that helps you identify certain activities as games -- it isn't what makes those activities games.
What about the role of 'measurement' in QM. Of course, most people agree that we would either like to have a version of QM where measurement plays no role, or else give a precise account of what it is to be a measurement. But right now we seem to function, for the most part, with a version of QM where measurement plays a central role, despite the absence of anything like clear N&S conditions for what it is for something to be one.
ReplyDeleteEric, I don't know enough about this, but granted that N&S conditions aren't available, is this really a cluster concept? -- one where there are multiple weighted criteria no one of which is either necessary or sufficient, and such that some sort of threshold judgement is made? It seems to me that the relevant idea here is something like vagueness, rather than clustering. But tell me more.
ReplyDeleteThanks for the reply Mohan. Your comment on species pluralism makes me wonder if by distancing yourself from Wittgenstein, you move closer to "homeostatic property cluster" (HPC) accounts of kinds. I'm not sure if HPC advocates have discussed whether definitive property clusters are open ended. But on first blush I don't see that HP clusters need to be open ended. The HPC view says that the cluster of properties that individuates the kind 'species' consists of properties (e.g., being reproductively isolated, monophyletic, etc.) no single one of which need be necessary for membership, but some sub-sets of which are sufficient for kind membership. It seems that the disjunction of *all* sub-sets that are sufficient as such must be necessary for kind membership. HPCers can take this view on board. Wittgenstein didn't, but I don't see that he has a good argument for denying it in the case of scientific kinds. (His rope metaphor is opaque and no help.) To deny that all sufficient sub-sets are necessary, I think you have to claim that the set of all sufficient sub-sets is infinite. Even then, I'm not sure you can deny the necessity claim. Granted, the HPCers allow (though I'm not sure whether they insist) that no single property, such as reproductive isolation, is neccessary for kind membership. But on your 'disjoining view', you'll have to allow this too. Then the only difference between your view and the HPC view is that you think there are no cases in which it is multiple properties that, together, MAKE some group a species; you think that in each case of a species, a single property MAKES the group a species, and that the disjunction of all such properties individuates the kind species. Whether this is the right view depends on the role that 'species' play in the appropriate theory. This is a controversial issue. But one popular view is that species are distinguised from higher taxa in virtue of being cohesive evolutionary units. If this is right, then 'species' should be defined in terms of properties that make species cohesive as such. And in many cases, multiple properties work in concert to make for the distinguishing cohesion. The HPC view builds on this fact, when it says that it is clusters of properties that make for species; your singleton view, however, papers over this fact. In short, the singleton view raises this question: if reproductive isolation is sufficient in some cases for being a species, and occupying a particular niche is sufficient in others, why say that only one (either one) of these does all the sufficiency work in cases where groups have both properties? The question becomes more pressing when we realize that such properties are often reinforcing, e.g., reproductive isolation promotes unique-niche-habitation, but also vice versa.
ReplyDeleteOn Suits, I think the set of 'all means available' for achieving a goal is going to involve social or conventional means. These are often the sorts of means we restrict in designing games. For example, before designing the game of golf, the set of all means available for getting a golf-ball-like ball into a golf-hole-like hole included, say, just dropping ball in hole with your hand. But it excluded having another person do the dropping by threatening the person with murder. (It seems these sorts of means and not just means defined by physics are relevant, since it is not physics that rules of games restrict.) My thought was that someone could design a perverse game where these previously ruled-out means are allowed in--indeed, where anything nomologically possible is allowed.
Mohan,
ReplyDeleteFair point. I wasn't so much thinking in terms of cluster concepts specifically, as about the claim that "there is an underlying unity in the sorts of concept that are robustly used in everyday and scientific discourse." I suppose I think there is an underlying unity (with some obvious vagueness) to the concept of measurement at a superficial level, but that at the relevant level of detail, there isn't. Maybe a better way to put it: its the concept of a "measurement device" or a "measurement setup" (which is the concept that I need to know whether or not applies in QM) that I doubt has an underlying unity.
I would have to think a lot more about whether or not its a cluster concept or not.
Thanks for these nice and thoughtful comments. Very useful as I prepare for a commentary on Dutton's conception of art, to which I alluded in my original post.
ReplyDeleteMatt, it's ironic that you should mention the HPC or homeostatic property cluster view. Marc Ereshefsky and I argued here:
https://tspace.library.utoronto.ca/handle/1807
that this is the wrong approach to species. The mistake, we claim, is that HPC views are responses to the problem of defining membership conditions on species. We claim that species are by contrast POPULATIONS with certain properties -- properties like the ones you mention -- and that once you have defined or isolated this higher-order entity, then individual members are identified not by their own intrinsic properties, but by their participation in the higher level entity. We call this the Population Structure Theory. I later applied it to the problem of SPECIATION here:
http://web.mac.com/mohanmatthen/Site/Bibliography_files/Chickens%20and%20eggs%20final%20%282%29.doc
This is, of course, a comment on the species concept specifically, not about cluster concepts in general. However, Ereshefsky and I characterized the HPC account as a "deep" cluster concept account and attacked it as such. That is another facet of my scepticism about cluster concepts.
Eric: I think I understand where you are going with the idea that the concept of a measurement device is a cluster concept. But can you give me an explanatory context in which this broad concept, as opposed to some specific kind of measurement device, is invoked? That would help me evaluate the claim.
In the meanwhile, thanks for these thoughtful challenges.
Martin says...
ReplyDeleteHi Mohan,
What do you think, then, of Hilary Putnam's invocation of a 'law-cluster' term (in "The Analytic and the Synthetic")? Putnam argues that terms like 'energy' are constituted by clusters of laws into which they enter and, collectively, these roles in laws constitute the meaning of such terms. However, abandoning/revising a single law does not thereby alter the meaning of such concepts (that's why he calls it a cluster view).
Now, it looks like the concept of energy plays a substantive role in scientific explanations. So is your thesis that Putnam is simply wrong about the concept of energy, or is the law cluster view not the kind of cluster account of concepts relevant to your thesis? If it isn't, why not?
Hi Martin,
ReplyDeleteIt's been a long time since I read that Putnam article, and it isn't available to me at home on a Sunday afternoon. Here is a question:
There are two sorts of thesis one could hold about terms like 'energy'. The first is that their meaning is cluster-determined by the laws in which they figure. The second is that their reference is somehow cluster-fixed by these laws. As I figure, the first thesis yields a cluster conception, but the second not.
I myself would have guessed that Putnam would be attracted by the second, but not the first. Do you disagree? I'll need to look again when I get to the office tomorrow.
"can you give me an explanatory context in which this broad concept, as opposed to some specific kind of measurement device, is invoked? "
ReplyDeleteWell, the fundamental problem with QM is that it is invoked anywhere and everywhere. Orthodox QM has two laws of evolution: schroedinger evolution and "wavefunction collapse". Every system evolves according to the deterministic schroedinger equation unless it interacts with a "measuring device", in which case an entirely different dynamical law takes over, and the system behaves stochastically, "collapsing" into an eigenstate of the observable being measured.
So, any explanation of the behavior of a quantum system (in orthodox QM) that has undergone measurement will necessarily have to mention the fact that measurement occured.
Now, there are various proposals for how to redo QM that avoid this problem (Bohmian mechanics, etc), but ordinary physics makes use of orthodox QM, and so makes fundamental use of the concept of a measurement device to explain the behavior of systems.
Or so it would seem. I'm just throwing this out for consideration.
Hi Mohan,
ReplyDeleteLooking back at what I wrote, I realize I used 'meaning' at one place when I should have written 'extension.' From what I remember of the article (I don't have it in front of me), Putnam's thesis is that the meaning of the concept is determined by the totality of the laws into which is figures, but that something need not satisfy all the laws in order to be in the extension of the concept. He makes reference to Einstein here, suggesting that physicists before and after Einstein were talking about the same thing, despite the changes in physical theory.
I'll look it up tomorrow and see if I am misremembering/misrepresenting what he said.
Martin
A word about HPC kinds...
ReplyDeleteI agree that the HPC view of species struggles with polymorphism -- indeed, in light of the prevalence of polymorphism it seems doubtful that species are natural kinds at all. Natural kinds are supposed to be the sort of things across which inductions tend to work. But take, for example, the angler fish: you can't make remotely reliable inductions about the male by extrapolating from observations of the female.
So suppose species aren't natural kinds. It would be premature, I think, to conclude that there are no natural kinds in biology. After all, if this were the case, methods of biological inquiry that involve inductive extrapolation from laboratory experiments or field observations would be epistemologically suspect. What good is a "model organism" if there are no biological kinds?
So if there are biological kinds, but they aren't species, what are they? Perhaps distinct morphs within species are the natural kinds: for example, the /female/ angler fish. Phylogenetically conserved developmental outcomes like the mammalian eye and the vertebrate limb look like good candidates too.
And it seems plausible to me that:
1) such biological kinds are HPC kinds;
2) they are identified by a cluster of properties;
3) they play an indispensable role in biological inquiry.
Hi HPC commenters,
ReplyDeleteThanks for engaging the topic. Mohan, I'm familiar with your piece with Ereshefsky. Myself, Rob Wilson and Ingo Brigandt respond to your worries in an article about to appear in /Philosophical Topics/. There, we address the issue of polymorphism (a preview: all angler fish, even females, have the property "if male, then have testes"). And we clarify that definitive cluster properties needn't be intrinsic and that the HPC view predicts that many are relational--properties picking out how organisms relate to other population members. We give reasons for thinking that although deep heterogeneity characterizes species populations (such that no one property is necessary for membership in an HPC species kind), there are interesting generalizations that quantify over their members.
Hi Matt, Jonathan
ReplyDeleteThanks for your comments on HPCs. I'll respond briefly, though I should remark parenthetically that I take HPC theory to be a great advance on everything that came before.
1. In our paper, Marc and I address the prospect of unifying species in terms of properties such as 'if male, then testes'. (We call this the "Mereological Maneuver". Our objection was that one can unify the whole biosphere via properties such as these. So if you want to understand what is special about species you need to understand how the population maintains a particular kind of structure in its environment. Conditional properties that fall out of such active population-structure-maintenance are characteristic of species; other conditional properties are neither here nor there.
2. There certainly are, as both of you point out, generalizations that take in all members of a species. The problem that polymorphism poses is not that there are no species wide uniformities, but rather that species-wide variation exists. HPC theory posits mechanisms that maintain property clusters; PST posits mechanisms that maintain heterogeneity.
The difference between HPC and PST can be summarized this way. While both recognize variation in species-populations, HPC theory seeks to smooth these out by means of conditional properties and cluster properties, and seeks to explain the prevalence of these conditional and clusters as a failure of homeostasis. Population structure theory takes variation to be an essential property of populations, and takes the same homeostatic mechanisms to be active maintainers of said variation.
Hi Mohan,
ReplyDeleteThanks summarizing the difference you see between HPC and PST. It makes me see that HPC theory can be interpreted as what Sober calls a "natural state model". I think this is the wrong interpretation of it, but that is for another time and place. Thanks for the great post.
Hi Matt,
ReplyDeleteI enjoyed reading your Philosophical Topics paper. We recently discussed it at the Cambridge HPS PhilBio Reading Group.
I wonder what you make of the following line of argument:
1. Intuitively, natural kinds admit of good inductive generalizations. For instance, if gold is a natural kind, observing gold in the lab should tell me something about gold outside the lab. This is a desideratum any account of natural kinds should aim to capture.
2. I see no reason in principle why polymorphism in some species could not be so dramatic that inductive generalizations based on observations of a single morph would lead only to grossly misleading conclusions about the other morphs. Suppose that such a species exists.
3. From 1, this species is intuitively not a natural kind.
4. If some account of natural kinds implies that, contrary to our intuitions, this species is a natural kind after all, the account fails to meet our desiderata for an account of natural kinds, and should be rejected.
Jonathan, I know your comment is not addressed to me, but may I ask a question? In 2, you seem to suggest that the polymorphic species in question supports NO inductive generalizations. There could be two reasons for this. First, the polymorphisms could be so dramatic that there are no significant similarities between morphs. Alternatively, such polymorphisms as exist throw doubt on inductive generalizations even with regard to features that are actually shared. Here is an example: Let's say males and females are different in many ways, but similar in colour. These other differences lead you not to trust the inference, "This male member of the species is c-coloured, therefore all members of the species is c-coloured." One might argue that wherever there is polymorphism, this kind of undermining of induction would take place. Do you agree?
ReplyDeleteMohan
Hi Mohan... I do agree, yes. Any polymorphism within a species gives us a reason to doubt inductive generalizations across that species.
ReplyDeleteThe question then is this: how dramatic does polymorphism have to be before our inductions are so badly undermined that it becomes implausible to call the species a natural kind?
I think we have to draw this line somewhere, and, wherever we draw it, I see no reason in principle why a species could not cross it.
I agree, Jonathan.
ReplyDeleteSome species are natural kinds, others are not.
But as long as there are possible species that are NOT natural kinds, it follows that being-a-natural-kind is not part of the species-concept. Was this the point you were making? (Would your 4 read equally well if the first occurrence of 'natural kind'were replaced by 'species'?)
"As long as there are possible species that are NOT natural kinds, it follows that being-a-natural-kind is not part of the species-concept."
ReplyDeleteI agree with this too.
It seems to me that Wilson, Barker & Brigandt want to counter this by saying that, while it may /appear/ possible to have a species that is not a kind, the HPC account is permissive enough to count any species as a kind, no matter how dramatic its polymorphism.
But if that is what their account says, then it allows for natural kinds that do not support inductive generalizations. This suggests the account is overpermissive.
Hi M and J,
ReplyDeleteSorry, I haven't checked this thread since I last posted.
The appeal to the possibility of radically polymorphic species doesn't sting the HPC view because polymorphism concerns morphology (as the name suggests) but the HPC view allows a lot more than morphological properties into the definitive clusters. (The HPC view is not just fancy pheneticism.) Additional sorts of properties can include reproductive ones, ecological ones, genealogical ones, and so on. An HPC species can *in principal* be radically polymorphic--so much so that inductions about morphology are strikingly unreliable; and yet, inductions about other sorts of properties will be sufficiently reliable by whatever reasonable standard of reliability we require of kinds. (I expect that as a matter of fact, such a yawning gap between the reliability of morphological inductions and others is exceedingly rare, but perhaps this is an uninteresting empirical hypothesis in this context).
The HPC account explicitly leaves it open to biologists to decide how much and what parts of a definitive cluster need to be instantiated for the corresponding kind to be instantiated. If they largely agree that a radically polymorphic group nonetheless forms a species, then I think they are suggesting that the members typically share some cluster of (non-morph) properties--relational properties, behavioral properties, historical properties--that form the basis for some worthy (non-morph) inductions. If they don't think such a basis is formed, I don't see that they'd recognize the group as a species. Our view is permissive to the extent that biologists are permissive. That is also the right amount, I think.
So in short, if a group doesn't support inductions in a way characteristic of a kind, it won't be a species; if it is a species, it will support inductions in a way characteristic of a kind--though perhaps not the inductions your example appealed to.
Thanks Matt, that's helpful.
ReplyDeleteI still think there's a problem though. We're imagining a species so dramatically polymorphic that inductive generalizations based on observations of a single morph will be grossly misleading.
Let's add that the polymorphism is more than skin-deep: it extends to things like behaviour, genes and developmental pathways. This is often the case when there is dramatic polymorphism.
I claimed that such a species would intuitively fail to be a natural kind.
As far as I can tell, you want to bite the bullet here. Your claim is that support for inductive generalizations over intrinsic properties is not the sine qua non for a natural kind. Provided we can make generalizations about, say, historical or relational properties, we may still have an HPC kind.
Is that the idea? If so, I don't think I can go along with it. It seems to collapse the kind/individual distinction.
Consider the parts of my computer. They share a common origin and many relational properties. If I observe that the hard disk is connected to an ASUS motherboard, I can reliable infer that the other parts are connected to an ASUS motherboard. If I observe that the hard disk was purchased in 2009, I can reliably infer that the other parts were purchased in 2009.
So do the parts of my computer count as an HPC kind? If we admit relational and historical properties into the cluster, why not? Yet this doesn't look right.
I'm not familiar with the details of HPC and had a quick question. In an earlier comment (2/2/10), it seemed that Jonathan was suggesting that degree of polymorphism somehow drew the line between natural kind species and non-natural kind species. Is this a correct reading?
ReplyDeleteWhile it makes sense that the degree of variability within a species (or any concept) could affect the kinds of inferences we could make about particular members, I don't see what this has to do with natural kinds. Unless all that is meant by natural kinds is that there is a larger proportion of inferences that can be made as compared to non-natural kinds. If that's the case, can we drop the "natural kind" label altogether?
Hi Jonathan,
ReplyDeleteHere is a longer response than you probably wished for!
I imagine a species whose members tend to vary radically with respect to all interesting intrinsic properties. But some of their relational properties seem widely shared and to explain the sharing of other relational properties. Say that one of these properties is R: "is a member of reproductive community X". Also say that this supports some reliable inductions. E.g., if R, then occupies niche N; if R, then shares a common ancestor C with other members of X; if R, then can mate with opposite-sex descendents of X; etc. In *principle*, I'm happy to define a natural kind term by appeal to an HP cluster consisting wholly in relational properties such as R, and which ground inductive generalizations restricted to other relational properties such as those just mentioned.
But in *practice* I'd define a natural kind term as such only with good empirical motivation for doing so. Whether there is such motivation here depends on what you take species to be. I see two prevailing types of definitions of species concepts. One takes species to be evolutionary units. (Most phylogenetic species concepts fall into this category because most maintain that "species" is a non-arbitrary category, and to get this they maintain that species are lineages that are evolutionary units.) On these views, species are distinguished from other things by a property often called species cohesion. Most of these views imply that species membership conditions are wholly relational. However, the cited relational properties are deemed definitive because of the basic causal role they play in generating species cohesion. And this cohesion consists in part in species members having intrinsic similarities that allow them to respond similarly (relative to non-members) to evolutionary pressures (with the intrinsic similarities being explained by the definitive relational properties). So on these views, I wonder whether the radically intrinsically varied group of organisms you imagine would even count as a species. (I've got a forthcoming paper on species cohesion if you're interested.)
The other prevailing type of species concept does not claim that species are evolutionary units--doesn't appeal to "species cohesion" or any related intrinsic similarities. These are strict genealogical concepts on which "species" is an arbitrary rank in the nested hierarchy of genealogically exclusive groups. Only very few phylogenetic species concepts fall into this category--it is an "I give up on species" category. There may be something to it. In any case, because it does not require, imply, or predict ANY intrinsic OR relational similarities among species members, ASIDE from being more closely related to each other than to any organism outside the group (which is simply entailed by the group being exclusive), it does not result in species that are natural kinds if natural kinds must be associated with non-trivial inductive generalizations. This is the only sort of view of species I can think of that is relevantly analogous to your computer. (I doubt whether the inductions you cite concerning your computer are reliable, but this may be a vestigae of my once having been a computer hardware technichian.)
Perhaps I'm sympathetic to Gennady's comment on "natural kinds". I'm most worried about individuation and induction. The HPC view provides a framework for induction that accommodates individuation by non-necessary properties that may be intrinsic or relational. It is the non-necessity and relationality that I think a theory of species indivdiuation needs to handle. These are the things evolutionary theory brought to our attention. If the HPC kinds view can handles these and help explicate how species feature in explanations and predictions, then I'm not sure that the "individual vs. kind" distinction matters much in the context of species.