Normativity, Function, and the Limits of Physicalist Supervenience

Stuart Kennedy  ·  2026  ·  Revised Draft Philosophy of Biology Metaphysics

A submission-ready draft arguing that biological normativity cannot be fully grounded in physical facts — and that the best physicalist responses presuppose rather than explain the normativity they target.

Abstract

Biological functions are normative: to attribute a function is to invoke standards of success and failure, such that a trait can genuinely malfunction. It is widely assumed that this normativity poses no serious metaphysical problem for physicalism. Selected-effects theories of function aim to naturalize biological normativity by grounding proper function in evolutionary history — a physical and causal matter. This paper argues that the attempt fails. Functional attributions involve correctness conditions that are underdetermined by physical and historical facts alone, even when evolutionary history is fully specified. The underdetermination is not merely epistemic: even complete physical knowledge leaves indeterminate which of a trait's many effects constitutes its success condition. Selected-effects theories do not resolve this difficulty; they presuppose the normativity they are meant to explain. Causal-role theories face a parallel problem at the level of systemic relevance. Biostatistical theories avoid some difficulties but cannot account for malfunction in pathological populations. The paper concludes that physicalist accounts of biological function face a forced choice between eliminativism and the acceptance of irreducible normativity.

Keywords: biological function, supervenience, physicalism, normativity, selected-effects theory, teleosemantics, malfunction, causal-role theory

1. Introduction

Functional explanation is central to contemporary biology. Hearts pump blood, kidneys filter waste, and molecular chaperones facilitate protein folding. Such explanations are not merely descriptive; they are evaluative. A heart that fails to pump blood is defective, not merely causally inert. A kidney that fails to filter waste is malfunctioning, not merely producing fewer effects than usual. Biological function therefore appears to involve standards of success and failure — features that philosophers standardly describe as normative.

Despite this, it is widely assumed that biological normativity poses no serious metaphysical problem. On standard physicalist views, once all physical facts are fixed — including microphysical constitution, laws of nature, and evolutionary history — facts about biological function are thereby determined. Functional facts are taken to strongly supervene on physical facts, even if they are explanatorily autonomous at higher levels of description. This assumption underlies much of contemporary philosophy of biology and philosophy of mind alike.

The assumption is most prominently defended by selected-effects theories of function. According to these views, the proper function of a trait is the effect for which it was selected by natural selection (Wright 1973; Millikan 1984; Neander 1991). Because selection history is a physical and historical matter, such accounts aim to ground functional normativity without invoking irreducible teleology or intentionality. When supplemented by counterfactual analyses and extended into teleosemantic frameworks for mental content, these views constitute the strongest available attempt to naturalize biological function.

This paper argues that the attempt fails. Functional attributions involve correctness conditions that are underdetermined by physical and historical facts alone, even when evolutionary history is fully specified. Selection explains why traits persist, but it does not by itself determine which of a trait's many effects is normatively privileged as its function. As a result, functional facts do not strongly supervene on physical facts. Physicalist accounts of biological function face a forced choice between eliminativism about function and the acceptance of irreducible normativity.

The argument proceeds as follows. Section 2 clarifies the normative character of proper function, distinguishing it from causal-role accounts and statistical accounts. Section 3 specifies the supervenience commitments at stake. Section 4 presents the central supervenience-failure argument. Section 5 examines selected-effects theory, biostatistical theory, and causal-role theory. Section 6 addresses teleosemantic implications and the deflationary response. The conclusion considers remaining metaphysical options.

2. The Normative Character of Proper Function

2.1 Causal Roles and Their Limits

One influential approach avoids normative commitments entirely. On Cummins' causal-role account (1975), to ascribe a function to a component is simply to identify its causal contribution to the capacities of the containing system. The heart's function is to pump blood because pumping blood is its causal contribution to the circulatory system's capacity to distribute oxygen. Functional facts are relational and analytical: they depend on how we individuate the relevant system and what capacities we choose to analyse.

Causal-role theory has genuine virtues. It handles the diversity of functional contexts gracefully and avoids commitment to selection history. However, it systematically fails to accommodate malfunction. On the causal-role account, a non-pumping heart is simply a component not contributing to the oxygen-distribution capacity in the usual way — it is not defective in any sense exceeding this purely causal description. A component either contributes or it does not; there is no normative residue.

This is precisely what biological discourse does invoke. A surgeon diagnosing cardiac failure identifies an error — a departure from what the organ should be doing. Medical diagnosis, pharmacology, and evolutionary biology alike presuppose that some states constitute failure relative to a standard, not merely reduced causal efficacy. Causal-role theory, by design, cannot capture this.

2.2 Statistical Normalcy and Its Limits

Boorse's biostatistical theory (BST) (1977; 2014) grounds function in statistically normal causal contributions to survival and reproduction within a reference class. Malfunction is statistical deviation from the species norm.

BST captures more than causal-role theory: it introduces a standard against which deviation constitutes malfunction. Nevertheless, it faces two structural problems. First, statistical normalcy and normative correctness can come apart: in populations where pathological conditions are widespread — endemic myopia, near-universal parasitic infection — the statistical norm may encode what we would ordinarily classify as impaired function. Statistical deviation is neither necessary nor sufficient for malfunction in the normative sense. Second, BST cannot explain why the statistically normal contribution counts as the success condition. Statistical regularities are descriptive; correctness conditions are not reducible to them without remainder.

2.3 Malfunction as the Test Case

Proper function involves correctness conditions that exceed both causal-role descriptions and statistical normalcy. Malfunction is not merely unusual causal behaviour — it is causal behaviour that fails to meet a standard. This entails: (i) a standard of success independent of what the system actually does; (ii) the system being subject to that standard — it ought to operate in the relevant way; (iii) deviation constituting failure, not merely difference.

Crucially, malfunction is independent of reproductive consequences in particular cases. A heart that misfires during an otherwise successful reproductive life is still malfunctioning, even though the misfire has no negative fitness consequences. The correctness condition is not defined by actual fitness outcomes — it is defined by a standard the system is subject to regardless of consequences. This independence is what makes malfunction genuinely normative rather than merely causal.

2.4 The Normative Commitment

To attribute a proper function is to commit to the existence of a correctness condition — a standard of success — relative to which the system can operate correctly or incorrectly. This correctness condition is not reducible to: (a) the system's actual causal contributions, (b) the statistical profile of the population, or (c) any conjunction thereof. This is a minimal normative claim: it asserts only that the correctness conditions involved in proper function outstrip purely extensional, causal-statistical descriptions.

3. Supervenience and the Physicalist Commitment

3.1 Strong and Weak Supervenience

Weak supervenience holds that, within any possible world, physical duplicates are duplicates in all other respects. As Kim (1984) notes, this is compatible with cross-world pairs of physical duplicates differing in non-physical properties — it cannot capture the claim that physical facts necessitate all other facts. Strong supervenience holds that, necessarily across all possible worlds, physical duplicates are duplicates in all respects. This is what physicalism requires.

(S) Necessarily, for any two possible worlds w₁ and w₂, if P(w₁) = P(w₂), then F(w₁) = F(w₂),

where P is the complete physical description (including evolutionary history) and F is the totality of functional facts.

3.2 Why Weak Supervenience Is Insufficient

Retreating to weak supervenience concedes the point: if functional facts require non-physical supplementation to be fixed, functional normativity is not fully grounded in physical facts. Expanding the physical base is legitimate insofar as those additional facts are themselves physical. The question is whether P, however expanded, suffices to fix F. The argument of Section 4 is that it does not.

3.3 Supervenience, Grounding, and the Scope of the Argument

The paper's claim can be stated in two ways. As a supervenience claim: Thesis (S) asserts modal covariation — no two possible worlds share the same physical facts while differing in functional facts. This is what the supervenience-failure argument directly targets. As a grounding claim: A stronger version holds that physical facts do not ground functional facts, where grounding is an explanatory, non-symmetric dependence relation (Fine 2001; Rosen 2010). Supervenience and grounding come apart in both directions.

The present argument is primarily targeted at supervenience: the modal argument of Section 4 attempts to construct possible worlds that are physical duplicates but functional diverges. If the thought experiment succeeds, supervenience fails and grounding fails a fortiori. However, even if a physicalist were to resist the modal argument — by insisting that the possible worlds imagined are not genuinely possible — the grounding question would remain independently pressing. For the physicalist to claim that physical facts ground functional facts, they must specify an explanatory account of how the normative properties of correctness conditions are derived from the non-normative properties of physical history. The sections that follow argue that no such account is available.

4. The Supervenience Failure Argument

4.1 The Underdetermination Problem

Let T be a biological trait type — the vertebrate heart. T produces many effects throughout its evolutionary history: pumping blood, generating electrical signals, producing characteristic sounds, generating heat, contributing mass to the thoracic cavity. All are real causal effects, equally part of the complete physical description P(T).

The difficulty is not that physical facts fail to correlate with functional attributions. Pumping blood ranks highest in causal relevance to differential reproduction throughout the lineage. The physicalist is right to point to this correlation. The difficulty is that correlation is not determination. Moving from "E₁ was more causally relevant to reproduction" to "E₁ is what hearts are for, and failure to produce E₁ constitutes malfunction" requires an additional normative step — a judgment that causal relevance to reproduction is what matters in determining success conditions. That judgment is not contained in P.

4.2 Modal vs. Epistemic Underdetermination

The claim is not epistemic uncertainty. Two possible worlds, physically and historically identical in every detail, are genuinely open to different functional attributions depending on which normative standards are operative. There is a possible world w₂ physically identical to our world w₁ in which the correctness condition for hearts is sound production rather than blood pumping. Nothing in the physical-evolutionary history rules out this evaluative standard. Since w₁ and w₂ share the same P but differ in F, (S) fails.

4.3 Defending Premise (P2): Beyond the Is-Ought Gap

The modal argument rests on:

(P1) is established by Section 2. (P2) requires careful defence, which must be distinguished from two weaker claims that will not suffice.

The first weaker claim is the epistemological point that causal explanation is interest-relative: any physical history supports multiple explanatory framings depending on which contrasts the investigator finds salient. This is true but insufficient. It would show only that we cannot read off the function from physical history unaided — an epistemic limitation — not that physical history fails to fix the function as a matter of metaphysical fact.

The second weaker claim is the bare Humean observation that 'is' does not entail 'ought' as a matter of formal logic. This too is insufficient on its own: a naturalist can accept Hume's point at the level of formal inference while holding that normative facts are constitutively identical to certain physical facts, so that the 'ought' is not derived from but just is the relevant 'is'. The Humean gap does not by itself rule out this identity claim.

What (P2) requires is the stronger claim that no physical-historical pattern can constitute a correctness condition, even constitutively. A correctness condition is not merely a causal pattern or a statistical regularity; it is a standard that makes a certain kind of claim on the system — it is what the system is answerable to, not merely what the system typically does or was disposed to do. The difference is precisely the normative residue that survives the subtraction of all causal-statistical description: the system's being subject to evaluation as succeeding or failing, not merely as instantiating or not instantiating a pattern.

To constitute a correctness condition, a physical pattern would have to be constitutively identical to the property of being a standard relative to which success and failure are defined. But the property of being a standard relative to which failure is possible is not a causal, statistical, or historical property. It is a property that determines what counts as going right and what counts as going wrong — a property that belongs to the normative order, not the causal order. No causal-historical fact, however richly described, has this property constitutively, rather than merely tracking it from the outside.

This claim may be contested by higher-order physicalists, who argue that normative properties supervene on physical properties at a sufficiently abstract level of description. This response is addressed in Section 5.1.5. The point here is that no such identification has been vindicated, and the analysis of Sections 2.1–2.4 gives independent reason to think none is forthcoming: each candidate physical property — causal efficacy, statistical frequency, selection history — has been shown to underdetermine the correctness condition without normative supplementation.

5. Physicalist Responses and Their Limits

5.1 Selected-Effects Theory

The dominant physicalist response is SE theory (Wright 1973; Millikan 1984; Neander 1991; Godfrey-Smith 1994): T has function F iff (i) T produces F and (ii) T exists because past instances produced F — F was the effect selected for. Both conditions are physical and historical. SE theory aims to close the is-ought gap by reidentifying the normative standard with a physical-historical fact.

5.2 The Covariation Problem and Millikan's Response

Traits produce many effects covarying with reproductive success throughout the lineage. SE theorists respond that the function is the effect that causally explains selection — the one that made the difference. This is the heart of Neander's (1991) and Godfrey-Smith's (1994) refinements.

Millikan (1989) presses this line more carefully. She distinguishes between an effect that is merely 'normal' for members of a reproductively established family and an effect that is the direct proper function — the effect type the item was 'designed' by selection to produce. On her account, the proper function of a token heart is determined by the function of the type to which it belongs, where type-membership is fixed by reproductive lineage. The selection explanation for the persistence of the type thereby fixes the proper function without requiring a separate normative input.

This is a more sophisticated position than the bare SE formulation, and it cannot be dismissed without direct engagement. Two difficulties arise, however. First, Millikan's account relies on the notion of the selection explanation for a trait's persistence. But as the causal-explanation problem shows, there is no single, mind-independent selection explanation for any trait: the same causal history licenses multiple correct explanatory descriptions, depending on what contrast classes are salient. Millikan's account does not explain how the physical history itself determines which description picks out the proper function. The appeal to 'the reason for the type's persistence' presupposes that there is a uniquely privileged reason — and that presupposition is precisely what needs to be grounded.

Second, and more fundamentally, Millikan's framework still requires that we can distinguish effects that a type was selected for from effects that merely accompanied selection. She is explicit that not every effect correlated with reproductive success constitutes a proper function. But this distinction — the 'for' that separates proper function from fortuitous effect — is normative in the sense argued above: it requires treating one causal contribution as the success condition and others as incidental. The reproductive-lineage apparatus determines which lineage is relevant; it does not generate the normative distinction between what the lineage was selected for and what came along for the ride. Millikan's account inherits, rather than dissolves, the presupposition charge.

5.3 Counterfactual Refinements

Sophisticated SE variants appeal to counterfactual robustness (Godfrey-Smith 1994; Schulte 2013): F is T's function if T would have been selected for producing F across a range of nearby possible environments. This presupposes a normative criterion for what counts as success across those environments — specifying which effect would have been "selected for" across possible histories already requires a normative standard. It also cannot account for malfunction independently of actual outcomes: a heart misfiring in an individual whose reproduction is unaffected is malfunctioning, yet counterfactual analysis generates this verdict only by tacitly importing the normative standard it was supposed to derive.

5.4 The Presupposition Charge

To make the problem vivid, consider a symmetry case. Suppose early vertebrate hearts evolved in an environment in which efficient blood circulation and efficient heat distribution were equally causally relevant to differential reproduction — both contributed independently and approximately equally to survival across the relevant ancestral populations. The physical-evolutionary history is, with respect to these two effects, symmetrical: neither has greater causal explanatory weight.

In this case, SE theory has no physical-historical resources to determine which effect is the proper function. Any answer requires a normative commitment — a judgment about which effect counts as the one the trait is for — that the physical history does not supply. The SE theorist might respond that in this case there are two proper functions. But this concedes the point: 'the function' is not read off the physical history but is determined by whichever normative framing we bring to it. Multiple framings are compatible with the same physical history; what distinguishes them is not a physical fact but a normative choice.

SE theory, far from reducing functional normativity to physical history, takes functional normativity as an input to its analysis of selection. An account that presupposes normativity cannot ground normativity.

5.1.5 Higher-Order Physicalism

A more sophisticated physicalist response argues that functional normativity supervenes on physical facts at a higher level of abstraction. On this view, the property of being a correctness condition is realized by a higher-order physical property such as being the effect type that has historically explained the persistence of the trait type across selective environments. This higher-order property is physical — definable in terms of causal history and differential reproduction — and the normative property of being a correctness condition is constitutively identical to it.

This view deserves more attention than it typically receives. It avoids the covariation problem by moving to a highly abstract level of description at which the relevant causal-historical property may be unique. Nevertheless, the response faces a version of the same problem. Even at the higher level of abstraction, the property of being the effect type that historically explained persistence must be individuated — and individuation requires specifying what counts as an explanatory unit, what the relevant causal-historical grain is, and which environmental contexts are included. Each of these choices involves normative judgment about what matters for functional attribution. The higher-order property does not uniquely determine itself; it requires normative input to be specified.

More fundamentally, the identification of 'being a correctness condition' with 'being the historically persistence-explaining effect type' is not a discovery but a stipulation. To treat these as constitutively identical is to legislate that the normative standard just is the causal-historical pattern — not to explain why it is. The physicalist owes an account of why this identification holds, and any such account will, on the present argument, need to appeal to normative resources that the physical base does not supply.

5.5 Biostatistical Theory

Boorse's BST (1977; 2014) grounds functional normativity in statistical normalcy relative to a reference class. A part or process functions normally when its causal contribution to survival and reproduction lies within the statistically normal range for the organism's age-sex class. BST avoids the selection-history dependency of SE theory and handles some cases more intuitively. Nevertheless, BST faces two decisive objections.

First, as noted in Section 2.2, statistical normalcy and normative correctness can come apart. Widespread pathology — near-universal myopia, endemic parasitic infection — can establish a statistical norm while leaving intact the intuitive judgment that the sub-norm condition represents malfunction. Boorse (2014) acknowledges this challenge and responds by refining the reference class — arguing that the relevant norm is indexed to the species design rather than to any actual population distribution. However, this refinement reintroduces the problem: the 'species design' is not a purely statistical notion; it imports a normative conception of what the species is supposed to be like, which is precisely what BST was meant to ground rather than presuppose. Wakefield (1992) and Schwartz (2007) develop related objections, arguing that BST cannot distinguish genuine dysfunction from mere statistical abnormality without invoking a prior conception of natural function that the statistical framework cannot supply.

Second, and more fundamentally, BST reduces biological normativity to statistical facts, which are a subset of physical facts. The supervenience argument therefore applies: fixing the statistical norm does not fix the normative standard — the correctness condition — without a further judgment that the statistical norm is the standard of correctness. That judgment is normative, not statistical. BST is best understood as an operationalization of biological normativity for medical purposes — a useful heuristic that tracks the normative standard in typical cases but does not constitute it.

5.6 Causal-Role Theory Revisited

Cummins' causal-role theory (1975) avoids the problems of both SE theory and BST by foregoing any attempt to ground normativity. Functions are simply causal contributions relative to a chosen systemic description. There are no standards of success; hence no malfunction; hence no normative residue. As noted in Section 2.1, this means causal-role theory is not a competitor to the normativity analysis — it is a revision of the concept of function that eliminates the phenomenon at issue.

Even within causal-role theory, the functional description of a system is relative to a systemic decomposition — a choice not fixed by physical facts alone. This shows that even the most deflationary form of function-attributing discourse introduces an interest-relative element that physical facts do not determine. The physicalist who retreats to causal-role theory has not escaped the underdetermination problem; they have relocated it.

6. Teleosemantics and the Deflationary Response

6.1 Teleosemantic Implications

Teleosemantics (Millikan 1984; Dretske 1988; Papineau 1993) grounds the intentional content of mental states in biological function: a mental state M has the content C if it is M's biological function to be tokened in the presence of C. Since biological function is grounded in selection history, content is naturalized.

If, however, biological function is not fully naturalized — if it presupposes irreducible normativity — then teleosemantics inherits the problem it was designed to solve. Intentional content requires correctness conditions; teleosemantics grounds correctness in biological function; but biological function already involves irreducible correctness conditions. The regress is not terminated. Teleosemantics may illuminate the evolutionary history of representational systems and the conditions under which representation can go wrong. But it cannot, on the present analysis, ground the normativity of content in physical facts. It can at best describe how normative standards are tracked by biological systems, taking those standards as given.

6.2 The Deflationary Response

The most sophisticated physicalist response accepts that 'function' talk in biology is interest-relative and pragmatic, while denying that this threatens physicalism. On the deflationary view, when biologists say the heart's function is to pump blood, they are not making a claim about an irreducible normative fact but about how the heart is usefully described relative to the explanatory interests of biological science. 'Function' talk is a useful organizing tool, not an assertion about sui generis normative facts.

This response has genuine force. A quasi-realist about biological function can maintain all the functional discourse of biology — attributing functions, diagnosing malfunctions, explaining traits teleologically — while holding that these attributions do not describe mind-independent normative facts but rather reflect the perspective of investigators with particular interests.

However, it is important to be clear about what this response achieves. The deflationary move does not rebut the supervenience failure argument — it concedes it. The original physicalist ambition was not merely that we can continue using function-talk; it was that functional facts are genuinely grounded in physical facts, that the world contains malfunctions and functions as real features of biological systems, not merely as projections of human explanatory interests. The deflationary response abandons this ambition. It is therefore not a way of saving strong physicalism about biological function; it is a way of giving it up gracefully.

Moreover, the deflationary response faces a second difficulty. Either the interests that generate functional attributions are themselves normatively significant — it matters that we attend to survival and reproduction rather than to sound production and heat distribution — or they are merely contingent pragmatic preferences with no further normative weight. If the former, the normativity has been relocated rather than eliminated: we now need an account of why these interests are the right ones to have. If the latter, then 'malfunction' attributions are indexed to contingent preferences that have no claim on the organisms themselves — hearts are not really malfunctioning when they fail to pump blood, any more than a stone is malfunctioning when it fails to skip across water in the way a skipping-stone enthusiast wants. This second reading collapses into full eliminativism: not the graceful quasi-realist position, but the costly one.

7. Conclusion

Biological proper function is irreducibly normative: it involves correctness conditions that exceed causal-role descriptions and statistical regularities. The physicalist commitment to strong supervenience of functional facts on physical facts fails because physical and evolutionary history underdetermine which of a trait's effects constitutes its correctness condition. Selected-effects theories presuppose rather than ground the normative distinction their reduction requires, even in Millikan's more sophisticated reproductive-lineage formulation. Higher-order physicalism relocates the problem rather than resolving it, since individuating the relevant higher-order property still requires normative input. Biostatistical theories face an analogous difficulty, as Boorse's own refinements reveal. Causal-role theories eliminate the phenomenon at the cost of eliminating malfunction. Teleosemantic theories inherit the problem they were designed to solve. The deflationary response either collapses into eliminativism or relocates the normative commitment without eliminating it.

The metaphysical upshot is a forced choice: either functional discourse is eliminable — a useful fiction with no normative import — or normativity is genuinely irreducible to physical facts. Physicalists who wish to preserve the explanatory resources of biological science, including the indispensable concept of malfunction, must acknowledge that their physical base, however expanded, does not suffice to fix the normative facts on which biological explanation depends.

This conclusion is deliberately limited in scope. It does not follow that normativity in general — moral, epistemic, or aesthetic — is metaphysically discontinuous with the physical world. The argument concerns the specific form of normativity involved in biological proper function, and its implications extend directly only to teleosemantic theories that take biological function as their foundation. Whether similar arguments apply to other forms of normativity, and what positive account of biological function is available to those who accept irreducible normativity, are questions for further work.

References

• Boorse, C. (1977). Health as a theoretical concept. Philosophy of Science, 44(4), 542–573.
• Boorse, C. (2014). A second rebuttal on health. Journal of Medicine and Philosophy, 39(6), 683–724.
• Cummins, R. (1975). Functional analysis. Journal of Philosophy, 72(20), 741–765.
• Dretske, F. (1988). Explaining Behavior: Reasons in a World of Causes. MIT Press.
• Fine, K. (2001). The question of realism. Philosopher's Imprint, 1(1), 1–30.
• Godfrey-Smith, P. (1994). A modern history theory of functions. Noûs, 28(3), 344–362.
• Kim, J. (1984). Concepts of supervenience. Philosophy and Phenomenological Research, 45(2), 153–176.
• Millikan, R. G. (1984). Language, Thought, and Other Biological Categories. MIT Press.
• Millikan, R. G. (1989). In defense of proper functions. Philosophy of Science, 56(2), 288–302.
• Neander, K. (1991). Functions as selected effects: The conceptual analyst's defense. Philosophy of Science, 58(2), 168–184.
• Papineau, D. (1993). Philosophical Naturalism. Blackwell.
• Rosen, G. (2010). Metaphysical dependence: Grounding and reduction. In R. Hale & A. Hoffmann (Eds.), Modality: Metaphysics, Logic, and Epistemology. Oxford University Press.
• Schulte, P. (2013). Mapping the world: The teleosemantic approach to the content of perceptual states. Synthese, 190(8), 1609–1635.
• Schwartz, P. H. (2007). Decision and discovery in defining 'disease'. In H. Kincaid & J. McKitrick (Eds.), Establishing Medical Reality. Springer.
• Wakefield, J. C. (1992). The concept of mental disorder: On the boundary between biological facts and social values. American Psychologist, 47(3), 373–388.
• Wright, L. (1973). Functions. Philosophical Review, 82(2), 139–168.