Innate Mind Need Not Be Within

Abstract

It is a widely accepted thesis in the cognitive sciences and in naturalistic philosophy of mind that the contents of at least some mental representations are innate. A question that has popped up in discussions concerning innate mental representations is this. Are externalist theories of mental content applicable to the content of innate representations? Views on the matter vary and sometimes conflict. To date, there has been no comprehensive assessment of the relationship between content externalism and content innateness. The aim of this paper is to provide such an assessment. I focus on the notions of innateness that are employed in innateness hypotheses within the cognitive sciences and adjacent fields of philosophy, and on causal externalist theories of content. I distinguish between three accounts of what being innate might amount to in innateness hypotheses within the cognitive sciences, and between three types of causal externalism. I explain what the possibility of innate externalistically individuated representations depends upon given all nine combinations. I explain why causal externalism can be true of innate mental representations, given but one of these combinations.

1 Introduction

The concept of innateness has been criticized in recent decades. It has been criticized for being explanatorily idle, metaphysically suspect, conceptually confused, and dangerous in practice. Despite this, hypotheses to the effect that a trait is innate continue to be common in explanatory contexts both within psychology and biology. Among such hypotheses is the widely accepted thesis in the psychological sciences and naturalistic philosophy of mind that the contents of some mental representations are innate. For example, in addition to basic perceptual representations (e.g., of redness), our innate representational repertoire has been argued to include concepts like CAUSE,¹ OBJECT, AGENT (Carey 2009; Spelke 1990), FACE (Diamond and Carey 1986), DANGEROUS ANIMAL (Barrett and Broesch 2012), various taxonomic categories of the animate world (Atran and Medin 2008), and more. Call the view that there are innate mental representations “representational nativism”. One question that has popped up in philosophical discussions concerning representational nativism is this. Can the content of an innate representation be individuated in accordance with an externalist theory of content? Put more dramatically, can the content of innate representations be “outside the head,” as the externalist slogan goes? Views vary. For example, Pitt (2000) puts forward an argument for the view that any kind of externalism about mental content is incompatible with empirical representational nativist hypotheses. According to Pitt, a mental representation is innate only if what is inside the head is sufficient for it to have the content that it does, whereas externalism is precisely the thesis that what is inside the head is not sufficient for a representation to have the content that it does. Therefore, he concludes, externalism cannot be true of the content of innate representations.² Some kind of tension between representational nativism and one or another version of content externalism has also been noted by, for example, Lau and Deutsch (2019), Sterelny (1989), Adamson (2001), and Cowie (1999). Then again, other philosophers combine externalism with content innateness. Jerry Fodor (1975, 1981), a proponent of the view that all concepts are innate, is also a proponent of externalism about conceptual content as such (e.g., Fodor 1990). Gross and Rey (2012) argue that the feasibility of concept nativism depends in part upon whether certain versions of externalism about content are feasible. Cummins (1997) and Rupert (2001) share the view that it is precisely innate representations that are best suited to having their contents individuated according to a causal externalist theory.

These various, sometimes conflicting, views lie scattered across the literature, often appearing on the sidelines of debates that do not have the relationship between externalism and nativism as their main topic. To date, there has been no overarching assessment of the relationship between content externalism and content innateness. The aim of this paper is to provide such an assessment and thereby prevent and clarify possible misperceptions. My focus will be exclusively on the notion of innateness that is operative in the nativist hypotheses within cognitive science and adjacent philosophical research, and on causal externalist theories. This is a natural focus. Firstly, it is the cognitive sciences that provide us with empirically grounded hypotheses regarding the innateness of certain mental representations. Secondly, causal externalist theories of content are the most prominent kind of externalist theory in naturalistic philosophy of mind.³

As we will see, there is no one generic answer to whether, when, or why causal externalism can be true of innate representations. The label “causal externalism” covers a range of different theories. Likewise, there is no consensus regarding what “innate” means in the cognitive sciences. It is likely that in different explanatory contexts, “innate” means different things. The implications of different types of causal externalism and different notions of innateness for whether and under which conditions a representation can be both innate and have its content externally individuated differ and need to be assessed separately. I will perform this assessment as follows. I will outline three influential accounts of what being innate amounts to in the innateness hypotheses that appear within the cognitive sciences (in Section 2) and three types of causal externalism (in Section 3). I will discuss (in Section 4) what the possibility of innate externally individuated representations depends upon given all nine combinations. I will argue that innate mental representations can be individuated according to causal externalism given all but one of these combinations.

2 Being Innate

Whether being innate sets any constraints on what the correct theory of content is for an innate representation, and if so, what these constraints are clearly depends upon what it means for a representation to be innate. Before I turn to clarifying this, some preliminary notes are in place.

First, in this paper, I will take for granted the view (also known as the representational theory of mind) that contents are carried by non-semantically individuated organism-internal symbols.⁴ This view is assumed by most representational nativist hypotheses and by the kind of causal externalism that I will discuss in this paper. Within this framework, a representation is a symbol that has content. For example, a mental symbol M (for “mental”) is a snake representation, SNAKE, if it has content <snake> (hereafter, I will use this notation to denote contents). Assuming this framework, that a representation R possessed by an organism is innate means that some symbol M of the organism has its content innately. What needs clarification therefore is what it is for M to have its content innately.

Second, as mentioned, it has been argued that the concept of innateness is suspect in various ways and should be expelled altogether from both explanatory and lay discourse (Griffiths et al. 2009; Griffiths 2002; Linquist 2018). In this paper, I bypass discussions concerning the pitfalls and benefits of the innateness concept. The fact is that hypotheses positing innate traits, including innate mental traits, continue to be common. Here the goal is to clarify the meaning of “innate” as it appears in such hypotheses, whether we approve of this meaning and these hypotheses or not. Such clarification is of foremost importance when the aim is to prevent possible misinterpretations and misuses of existing nativist hypotheses (an aim that the critics of the innateness concept have).

Third, this clarification should not rely upon pre-theoretical intuitions regarding innateness. Such intuitions—for instance, that innate traits are present at birth, form the essence of an organism, are unmalleable, and are not acquired—align poorly with, and are likely to encourage misinterpretation of scientific innateness hypotheses (Griffiths 2002; Griffiths et al. 2009).

Fourth, one talks of innateness in different explanatory contexts in different research fields. No existing attempt to provide an explication of the concept of innateness that would accord with the usage of “innate” in all (or most) of these contexts has gained univocal recognition. Rather, it is plausible that in different contexts, “innate” has slightly different meanings that do not converge on the same phenomenon (Griffiths 2002; Mameli and Bateson 2006). Thus, a trait (e.g., a representation) that counts as innate in one explanatory context need not count as innate in a different explanatory context where some other concepts of innateness are operative. Therefore, when we evaluate the implications of a given innateness hypothesis, we should properly delineate the explanatory context of interest, and be careful not to conflate these implications with the implications of innateness hypotheses in other explanatory contexts where different innateness concepts are employed. My focus here is the implications of hypotheses that posit innate mental representations. The main locus of such hypotheses is cognitive science and the adjacent philosophical research where the innateness discourse has been largely shaped by Jerry Fodor’s (1975, 1981) and Noam Chomsky’s (1965) classic arguments for radical concept nativism and universal grammar nativism respectively.

However, there is no consensus on what the concept of innateness is in cognitive science and the adjacent philosophical research. There is so much that is agreed upon that whether a representation is innate or not depends upon how the representation is, or is not, acquired, where for something to be acquired normally means for it not to be possessed at birth and for it to be possessed at some later ontogenetic timepoint.⁵ But views differ on how the “how” should be filled in. Correspondingly, there are different accounts of what innateness in the field of cognition amounts to. I will outline three such accounts—primitivism, dispositionalism, and the “non-experiential acquisition” account of innateness. These, I believe, exhaust the range of alternatives that have been explicitly advocated and those that conform to at least some usages of “innate” in the relevant literature. I will not take a stand on which of the three accounts is correct. In fact, they could all be correct because different innateness hypotheses within cognitive science and philosophy might employ different innateness concepts so that each of the three accounts of innateness correctly captures the meaning of some of these concepts. Rather than taking a stand on which of the three accounts is correct, I will later (in Section 4) discuss the relationship of content innateness and content externalism for all three accounts.

According to what is probably the most widely endorsed and discussed account of psychological innateness, a representation R is innate if and only if R is a psychological primitive in that it has not been acquired through a psychological process (henceforth, not acquired psychologically). This position is known as primitivism. Primitivism is endorsed for instance by Cowie (1999), Fodor (1981), and Samuels (2002). What makes R’s acquisition psychological—correspondingly, non-psychological—is a somewhat contentious matter. Paradigmatic examples of non-psychological acquisition are acquisition by brute-causal triggering and acquisition by maturation. Correspondingly, psychological acquisition is acquisition that differs from these intuitively non-psychological acquisition processes. It is less clear which positive feature is common to psychological acquisition processes and which distinguishes them from triggering and maturation.⁶ According to one standard conception, a representation R is acquired psychologically if and only if acquiring R is representationally mediated (Fodor 1981; Carey 2009). A paradigmatic example of acquiring R in a representationally mediated way is acquiring R through inductive or abductive learning, i.e., by way of extracting information from experience, then representing this information, and then employing it as evidence when assessing mental hypotheses about R’s content.⁷ However, not everyone agrees that this notion of psychological acquisition (qua representationally mediated acquisition) is the notion employed by psychologists. For example, Shea (2011) points out that there are uncontroversially psychological ways of acquiring new representations that do not involve operations over representations (see also Gallistel 2000); Margolis and Laurence (2011). Rather, the feature that is common to uncontroversially psychological kinds of representation acquisition, according to Shea, is that they consist of the forming of new representations through the processing of environmental information, irrespective of whether this information is represented. I will adopt this latter, more inclusive, notion of a psychological process and I will interpret the primitivist understanding of innateness accordingly. Let us call the primitivist understanding of innateness so interpreted P-innateness.

P-innateness. R is P-innate iff R is not acquired through the processing of information.

An alternative account of psychological innateness that has gained traction is dispositionalism. This is defended by Khalidi (2002, 2007). According to Khalidi, P-innateness as an explication of the concept of innateness in cognitive psychology is extensionally inadequate. He points out that there are paradigmatic examples wherein a cognitive trait is deemed innate in cognitive science even though the trait is acquired through the processing of information (see also Margolis and Laurence 2013). According to Khalidi’s own dispositionalist account—an account that is meant to cover such examples—R is innate if and only if R would be acquired by an organism in an environment E that is informationally impoverished relative to R. This is the case if the information contained in R is (to some significant degree) richer than the information available in E (Khalidi 2007, p. 99). Call the dispositionalist notion of innateness D-innateness.

D-innateness. R is D-innate iff R would be acquired in an environment that is informationally impoverished relative to R.⁸

Khalidi offers no general criterion as to when some environment where R is acquired is informationally impoverished relative to R. Instead, he refers to concrete examples of how and when scientists have in fact recognized informational impoverishment (Khalidi 2002, pp. 263–268). If there is any general test for informational impoverishment, it seems to be the following: An environment E in which R is acquired is impoverished relative to R if information available in E is insufficient to explain how and why R was acquired on the assumption that R was acquired through a domain-general mechanism (suggested in Khalidi 2002, p.268; Margolis and Laurence 2013). This test could be passed by R in the following two circumstances. (a) R is acquired by a domain-specific information processing mechanism that contributes specific endogenous information to R’s acquisition and thus co-specifies R’s content. (b) R’s acquisition is not explicable in terms of information processing at all. This general test for informational impoverishment reveals two relevant things for our purposes. First, it reveals that R can be D-innate without being P-innate (i.e., not acquired via the processing of environmental information). As per (a), R can be acquired in an informationally impoverished environment and thus be D-innate even if its acquisition involves the processing of environmental information, i.e., if R is acquired by a domain-specific information processing mechanism. However, that R is P-innate implies that R is also D-innate. If R is not acquired through the processing of information and is thus P-innate then, as per (b), R is acquired in an informationally impoverished environment and is thus D-innate. This implication will later afford us a shortcut. Having later demonstrated that the content of P-innate representations can be individuated according to an externalist theory, we will have also thereby demonstrated that the content of D-innate representations can be individuated according to the externalist theory.

Both P-innateness and D-innateness allow that a representation is innate even if its acquisition is caused by an environmental stimulus, for instance, by way of brute-causal triggering. Some might see this as a violation of a conceptual constraint on any adequate definition of innateness—namely, that such a definition should render being innate incompatible with having been environmentally caused.⁹ In order to attend to such intuitions, as well as to possible usages along these lines of “innate” in the scientific and philosophical literature, I introduce a third conception of what innateness amounts to. Call it the non-experiential-acquisition notion of innateness (NEA-innateness).

NEA-innateness. R is NEA-innate iff the acquisition of R has not been caused by an environmental stimulus.¹⁰ 

Under what conditions is R’s acquisition caused by an environmental stimulus, and under what conditions it is not, is a further question. Here is not the place to defend an answer to this question. Instead, I will adopt a view that I believe to be sufficiently uncontroversial for current purposes. According to a prominent view about scientific explanation (e.g., Woodward 2003), x counts as a cause of y if y counterfactually depends on x in such a way that if x had not occurred, then y would not have occurred. According to a popular elaboration of this basic view, claims of the form x is a cause of y are at their core contrastive claims that include explicit or implicit reference to the relevant alternative(s) to x or y or both. That is, they specify what would have occurred instead of x if x had not occurred and/or what would have occurred instead of y if y had not occurred (e.g., Hitchcock 1996; Northcott 2008; Van Fraassen 1980). A causal claim of the form x is a cause of y is true if and only if in the case of the specified alternative to x, y would not have occurred but instead some specified alternative would have occurred. This is the view I subscribe to. Assuming this view, the acquisition of R by an organism is caused by some actual environmental stimulus E only if, had the organism experienced some alternative environmental stimulus E*, the organism would not have acquired R. Accordingly, the acquisition of R is not caused by E (i.e., R is NEA-innate) if the organism would have acquired R even if the organism had experienced some relevant E* rather than E (I will add more flesh to this schema in Section 4). This interpretation of NEA-innateness does not only align with how causation is understood in causal explanations in the empirical sciences in general. It has also been argued to be the correct interpretation of “cause” in innateness hypotheses in particular (Birch 2009; Northcott 2012; Northcott and Piccinini 2018; O’Neill 2015).

3 Causal Externalist Theories of Mental Content

Representational nativist hypotheses are empirical hypotheses about how a representation is acquired. Theories of mental content are metaphysical theories about what makes it the case that a representation is the representation that it is, i.e., about what makes it the case that a mental symbol M has a particular content. What brings certain theories of content under the label “causal externalism” is the thesis that M has its particular content in virtue of standing in some appropriate causal relation to some feature(s) of the world that are extrinsic to the subject that has M.¹¹ Different kinds of causal externalist theories differ in terms of how they specify what this appropriate causal relation is. I will distinguish between three types of causal externalist theory on the basis of what kind of causal relation each theory says individuates the content of M: nomological covariation theories, historical covariation theories, and causal acquisition theories. This taxonomy might not neatly map onto other existing classifications. However, this is not important, since the taxonomy is tailored to group externalist theories specifically according to their implications for their applicability to innate representations. I take this taxonomy to exhaust the array of existing causal theories of content.

Causal covariation theories, also known as informational theories, are the most commonly advocated. According to causal covariation theories, the content of M is determined by what reliably causes M to be tokened (i.e., by what M causally covaries with). There are two kinds of covariation theories: nomic and historical. According to nomic covariation externalism (hereafter NCE), the content of M is what would cause an organism to token M in a lawful manner. Fodor’s asymmetric dependency account is an example of NCE (e.g., Fodor 1990). For instance, suppose that Roberta’s M has content ˂snake˃ and is thus SNAKE and that NCE is true of SNAKE. This is so if and only if it is true that if Roberta were to encounter a snake, then she would token M. Clearly, every theory of content must leave room for the possibility that, sometimes, SNAKE tokenings are caused by things that are not in the extension of SNAKE, for example, by meandering snake-like cracks on a dry surface, utterances of the word “snake,” and twigs on the road. Therefore, not everything that would cause M to be tokened can determine the content of M. Different versions of NCE specify differently which of the possibly different causes of M tokenings determines M’s content. These details are irrelevant for our taxonomic purposes. For these purposes, the relevant characteristic of NCE is that, according to NCE, snakes need not have actually caused M to be tokened in order for M to have content ˂snake˃. It suffices that snakes would cause M to be tokened.

In this respect, NCE differs from historical covariation externalism (hereafter HCE) (e.g., Dretske 1981; Maloney (1994; Rupert 1999; Ryder 2004; Usher 2001). According to HCE, the content of M is determined by what has actually caused a subject to token M. For example, if HCE is true of Roberta’s M, then M has content ˂snake˃ and is SNAKE only if snakes have in fact caused Roberta to token M. Again, not everything that has actually caused Roberta to token M can determine its content lest it be impossible that, sometimes, mental representations are caused to be tokened by things that they are not about. Views vary on which of the possibly different actual causes of M tokenings determines M’s content. For example, according to Dretske’s (1981) version of HCE, only those things that caused M tokenings during the period of learning the corresponding representation can determine the content of M. According to Rupert, the content of M is what has caused M tokenings most efficiently (in a certain technical sense of the word) etc. I will address some of these specifics in Section 4.3.

HCE should in turn be distinguished from causal acquisition externalism (hereafter AE) (e.g., Prinz (2004), pp. 251–252). According to AE, the content of a representation is determined by what caused its acquisition, that is, by what caused it to be the case that a mental symbol M of a subject has a particular content. Therefore, AE is true of Roberta’s SNAKE only if it is snakes that caused her to acquire SNAKE, that is, caused it to be the case that some mental symbol M of hers means <snake>. Sometimes, HCE and AE may assign to M the same content. This happens when the causes of M tokenings (for instance, snakes) that according to a version of HCE determine the content of M (as <snake>) are at the same time the causes of the acquisition of the corresponding representation (SNAKE). However, this need not always be the case, as I will explain in Section 4.3.

4 Can Causal Externalism Be True of Innate Representations?

We are now equipped to turn to our main question: Are causal externalist theories applicable to the content of innate representations? The thesis that a representation R is innate is a thesis about how R is acquired, i.e., about how an organism came to meet the conditions under which her mental symbol M has its particular content and is a particular representation R. Causal externalism is a theory about what these conditions are. Therefore, whether a causal externalist theory is applicable to innate representations boils down to whether it is possible for an organism to come to meet these conditions in a way that conforms to a definition of innateness. In the rest of the paper, I will identify which considerations this possibility turns upon, given the nine combinations of innateness definition (i.e., the three outlined in Section 2) and causal externalism (i.e., the three outlined in Section 3). As I will explain that R is innate does not rule out that causal externalism is true of R’s content, except when we combine AE with NEA-innateness.

4.1 NCE and Innate Representations

Let us begin with NCE. NCE can indeed be true of the content of innate representations regardless of which of the three definitions of innateness we assume. In order to see why and when, let us first outline what it is to possess a representation such that NCE is true of it, and then we will consider what it takes to acquire (i.e., come to possess) such a representation.

According to NCE, Roberta’s mental symbol M has content <snake> and is thus SNAKE only if snakes would cause Roberta to token M in virtue of a lawful dependence of M tokenings upon snakehood instantiations, i.e., only if M lawfully covaries with snakehood instantiations. Therefore, if NCE is true of SNAKE, then for Roberta to possess SNAKE is for her to be in a state where some M of hers lawfully covaries with snakehood instantiations. For Roberta to acquire SNAKE is for her to come to be in this state. Now, in order for Roberta’s M to lawfully covary with snakehood instantiations, there has to be something in place—a mechanism of some sort—that sustains this covariation, a mechanism that guarantees that snakehood instantiations would cause Roberta to token M. It is common to call such a mechanism a “sustaining mechanism”. Assuming NCE, the presence of some appropriate sustaining mechanism is, causally speaking, both necessary and sufficient for SNAKE possession. Acquiring SNAKE is the process whereby this sustaining mechanism comes into existence (Laurence and Margolis 1999; Margolis 1998; Margolis and Laurence 2011). Given this picture, SNAKE can be both innate and individuated according to NCE only if it is possible that an appropriate sustaining mechanism comes into existence in a way that conforms to a definition of innateness.

Whether this is so or not depends upon what such a sustaining mechanism consists of. For different representations, the sustaining mechanisms may consist of different things (for examples and discussion, see Cowie 1999, p. 132; Margolis and Laurence 2011; Sarnecki 2006). I will discuss two types of sustaining mechanism, which are commonly mentioned in discussions of NCE. These two types of mechanism do not exhaust the options. However, they will suffice for us to see that, when, and why representations which are individuated according to NCE can indeed be acquired in conformity with P-innateness, D-innateness, and NEA-innateness.¹² The first type of sustaining mechanism is as follows. For many of Roberta’s representations, the role of the sustaining mechanism is plausibly played by Roberta’s beliefs about what kinds of thing a given representation represents. Such beliefs guarantee that a mental symbol M is reliably activated in the presence of the property that R represents in virtue of supporting inferences from certain kinds of perceptual input to the conclusion that the property is instantiated. The second type of sustaining mechanism is as follows. In the case of developmentally basic representations (for example, sensory representations like RED), M–property covariation is likely to be sustained by non-cognitive neurophysical features of Roberta’s neural and sensory architecture (e.g., Cowie 1999, pp. 133 ff; Cummins 1997; Pylyshyn 1984; Rupert 2001).

If the sustaining mechanism for a representation R consists of Roberta’s beliefs then, indeed, R is unlikely to be P-innate and unlikely to be NEA-innate. Recall that, given NCE, for Roberta to acquire a representation is for an appropriate sustaining mechanism to come into existence. However, given the kind of entity that a belief is, a biological organism is unlikely to come to possess a belief in any other way than through experience-based evidential reasoning, for example, through learning in the form of hypothesis testing (e.g., Cowie 1999, Ch. 6; Sarnecki 2006). Hypothesis testing, however, is a paradigmatic instance of information processing. Therefore, if the sustaining mechanism for R consists of Roberta’s beliefs, then it is almost certain that Roberta acquired R by means of processing information and, consequently, that R is not P-innate. Moreover, if the sustaining mechanism for R consists of Roberta’s beliefs, R is also unlikely to be NEA-innate. Suppose that the sustaining mechanism for Roberta’s representation R is some belief B with content b. Most plausibly, what explains why Roberta acquired B rather than some different belief B* with content b* (or no belief at all) is that she experienced some environmental stimulus E that served as evidence confirming the hypothesis that b is true. Had Roberta experienced some significantly different stimulus E* instead of E, a different hypothesis would have been confirmed (say that b* is true) and Roberta would not have acquired B but, instead, some different belief, say B* with content b* (or no belief at all). Given the counterfactual notion of causation that we are here assuming, E thus counts as the cause of her acquisition of B and thus the cause of her representation R for which B functions as a sustaining mechanism. R therefore is caused by an environmental stimulus and so is not NEA-innate.

Note, however, that representations sustained by beliefs are much more likely to be D-innateness. In the case of some representations, the beliefs that form the sustaining mechanism for those representations, although learned, may well be learned by a dedicated learning mechanism and therefore in an informationally impoverished environment.

So, if the sustaining mechanism for R consists of Roberta’s beliefs, then R acquisition is very likely to be caused by an environmental stimulus and via a psychological process and so R is unlikely to be P-innate and NEA-innate. Things are different with representations whose sustaining mechanism consists of neurophysical features of Roberta’s sensory mechanism. For instance, it is plausible that a mechanism which ensures that redness experiences would reliably cause Roberta to activate a certain mental symbol would have been brought about by biological and neurophysiological maturation processes rather than through the processing of information. It is also plausible that the acquisition of such a mechanism is not caused by any environmental stimulus. This latter claim is more controversial. To motivate it, let us first consider how someone might argue for the opposite.

Suppose that Roberta possesses a sensory mechanism that, due to its neurophysical constitution, guarantees that red things would cause her to token M. Call this mechanism “Redness-Detector”. Given the counterfactual dependence account of causation, Roberta’s possession of Redness-Detector is caused by an environmental stimulus E if it is true of E that, first, Roberta experienced E and, second, that had Roberta experienced some alternative stimulus E* instead of E, she would not possess Redness-Detector. This seems to be the case with all sensory mechanisms. For instance, surely Roberta would not possess Redness-Detector if immediately after birth she had been deprived of nutrition and water. Had Roberta been deprived of nutrition and water, she would have died and would not possess anything. Nutrition and water therefore count as causes of Roberta’s possession of Redness-Detector.

But this conclusion does not follow. Not every x that some effect y counterfactually depends upon counts as the cause of y. For some x to count as a cause of y, the alternative to x, x*, in the case of which some different effect y* would have occurred, has to meet certain constraints. This is also the case when assessing whether a representation is NEA-innate, i.e., not caused by an environmental stimulus. With each trait (representational traits included), one can find an environmental stimulus such that in the case of some alternative to this stimulus, the trait would not have been acquired. So, if there were no constraints upon which E* is relevant for deciding whether a representation is NEA-innate, NEA-innateness would, trivially, have no instances—all traits would be caused by an environmental stimulus. Here is not a place to discuss in detail what the constraints for a relevant E* might be when assessing whether a representation is NEA-innate. The following consideration suffices. In the biological and psychological sciences, innateness hypotheses often serve to explain the phenotypic development of organisms in their normal environments. Given this, the relevant values for E*, when deciding whether a representation is NEA-innate, are plausibly those that count as developmentally normal for a given organism. Admittedly, the notion of a normal developmental environment is notoriously difficult to spell out. But, surely, no such environment is developmentally normal for an organism if the environment predictably results in the death or severe malfunctioning of the organism. An environment where Roberta is denied nutrition and water results in exactly this and therefore does not qualify as developmentally normal. So, even though it is true that if Roberta had been deprived of nutrition and water after birth then she would not possess Redness-Detector, it does not follow that Roberta’s Redness-Detector is caused by nutrition and water. Being deprived of food and nutrition simply is not a relevant contrast according to the abovementioned minimal standard of relevance. At the same time, it is plausible that in the case of most such E–E* pairs where E* qualifies as normal according to the abovementioned standard, Roberta would develop Redness-Detector regardless of whether she experienced E or E*; consequently, Redness-Detector is not caused by any environmental stimulus and RED is NEA-innate. If these considerations are sound, many sensory, and plausibly some primitive non-sensory, representations are eligible for being both individuated according to NCE and innate, given all three accounts of innateness.

4.2 AE and Innate Representations

According to NCE, what caused Roberta to acquire R (i.e., caused the existence of the appropriate sustaining mechanism) is orthogonal to what the content of R is. The latter is solely determined by what would cause the tokenings of the symbol M that functions as the vehicle of R. Not so, according to AE. According to AE, the content of Roberta’s representation R is what (beyond Roberta’s skin) caused Roberta to acquire R. Thus, R can be both innate and individuated according to AE only if it is possible for the causes of R acquisition to determine R’s content even if R is acquired in a way that conforms to a definition of innateness. If the relevant innateness definition is NEA-innateness, this is not possible. The content of R can be determined by the environmental stimulus that caused Roberta to acquire R only if the acquisition of R was in fact caused by an environmental stimulus. But R is NEA-innate only if the acquisition of R is not caused by an environmental stimulus. Therefore, if R is NEA-innate then, by definition, AE cannot be correct of R’s content (given that AE assumes the same notion of being a cause that matters in the context of NEA-innateness).

However, that R is P-innate or D-innate does not have the same implication. Recall that neither P-innateness nor D-innateness rules out the possibility that the acquisition of an innate representation was caused by an environmental stimulus. That R is P-innate or D-innate only rules out the possibility that R’s acquisition was caused by an environmental stimulus in a certain manner. So, the applicability of AE to P-innate and D-innate representations will depend upon certain details concerning what the process of R acquisition would have to be like if the causes of R acquisition are to be eligible for individuating R’s content. Let us focus on P-innateness (bearing in mind that being P-innate entails being D-innate).

That R is P-innate rules out that AE is true of R, and vice versa, if the cause of R acquisition can individuate R’s content only if the process whereby R is acquired involves information processing. Two reasons (at least) have been proposed for thinking that, for most representations, this is likely to be the case. The first is this. The thesis that the content of R is determined by what caused R acquisition presumes the existence of some suitable representation acquisition mechanism that explains how the causes of R fixed R’s content. For example, the thesis that SNAKE represents snakes because snakes caused Roberta to acquire SNAKE relies on the assumption that there was a SNAKE-acquisition mechanism that accounts for how and why snakes qua causes of SNAKE acquisition fixed the content of SNAKE. Some authors have argued that in the case of most representations, the only mechanism that can plausibly account for this is one or another kind of information processing mechanism: For instance, that SNAKE acquisition consisted in extracting information from experiences of snakes and then building this information into the content of the acquired representation (for more details, see Cowie 1999; Margolis and Laurence 2011; Shea 2011).

Here is the second reason. Since every snakehood instantiation is also a vertebratehood instantiation, whenever the acquisition of R is caused by a snake, it is also caused by a vertebrate. But why then is the content of R <snake> rather than <vertebrate>? According to Devitt and Sterelny (1987), with most representations, this is explicable only if the acquisition of the representation was mediated by the subject’s intention to fixate on only one of these properties, for example, on snakehood rather than vertebratehood in the case of SNAKE. Any intentional activity, however, is an activity of information processing (for argumentation along these lines, see Cowie 1999; Cummins 1997; Loar 1991; Sterelny 1989).¹³

Both reasons, if valid, show that in the case of most representations, AE cannot be true of a representation R unless R acquisition involved information processing and consequently, unless R is not P-innate—but only in the case of most representation. In the case of some developmentally primitive representations, the mechanism by which the cause of R acquisition fixes R’s content might as well be some brute-causal representation acquisition device (Cummins 1997; Rupert 2001). This device could have evolved to output R rather than R* in response to a specific kind of environmental stimulus, for instance, SNAKE rather than NATRIX NATRIX in response to snakes and OBJECT rather than SURFACE in response to macrophysical objects. Representations acquired by such devices would be both P-innate (and by implication D-innate) and eligible for individuation according to AE. Do such representation acquisition devices in fact exist, and if they do, which representations are acquired by such devices, are empirical questions.

4.3 HCE and Innate Representations

I begin by considering the compatibility of HCE with NEA-innateness. This will allow me to clarify some relevant differences between AE and HCE vis-à-vis their applicability to innate representations.

According to HCE, the content of a mental symbol M is determined by what has caused (in some appropriate way) an organism to token M. For instance, if HCE is true of the content of Roberta’s M, then M has content <snake> and Roberta possesses SNAKE if and only if snakehood instantiations have caused (in some appropriate way) her to token M. What this appropriate, content-determining, way of causing M to be tokened is, is defined differently by different versions of HCE. HCE is akin to AE in that according to both HCE and AE, Roberta can possess SNAKE only if she has in fact causally interacted with snakehood instantiations and thus with environmental factors. Because of this shared feature, HCE may at first glance also seem to have the same implications as AE for whether or not it is applicable to NEA-innate representations. For example, one might reason as follows: “If HCE is true of Roberta’s SNAKE, then Roberta must have causally interacted with snakes. Therefore, Roberta’s acquiring and possessing SNAKE must have been caused by snakehood instantiations and therefore by environmental stimuli. In contrast, SNAKE is NEA-innate only if the possession of SNAKE is not caused by an environmental stimulus. Therefore, if SNAKE is NEA-innate, then HCE, just like AE, cannot be true of SNAKE.”

But this reasoning by apparent analogy with AE equivocates. It equivocates between two ways in which an environmental stimulus, e.g., snakehood instantiations, could be necessary for SNAKE possession: constitutively or causally necessary. Indeed, if HCE is true of SNAKE, then an environmental stimulus is necessary for acquiring and possessing SNAKE. It is constitutively necessary in that what (metaphysically) makes it the case that the content of Roberta’s mental symbol M is <snake>, and that Roberta possesses SNAKE, is that snakehood instantiations have caused Roberta to token M. However, that SNAKE is NEA-innate only implies that an environmental stimulus has not caused Roberta to acquire and possess SNAKE.¹⁴ So, that HCE is true of SNAKE entails that SNAKE cannot be NEA-innate (as per the above reasoning) only given the following condition. Take the snakehood instantiations that, according to HCE, individuate M as SNAKE (by having caused Roberta to token M in some appropriate way) and therefore are constitutively necessary for Roberta to possess SNAKE. These snakehood instantiations must be what caused Roberta to possess SNAKE. But as I will now argue, they need not be.

In order to see this, let us spell out in more detail what possessing a representation of which HCE is true amounts to, and what it is to cause an organism to possess such a representation. Assuming HCE, Roberta’s mental symbol M has content <snake> and Roberta possesses SNAKE if and only if snakes have caused Roberta to token M in such a way that, according to a given version of HCE, determines M’s content. That snakehood instantiations have caused Roberta to token M in such a content-individuating way simply is what Roberta’s possession of SNAKE consists in. In other words, for Roberta to possess SNAKE is for Snake-History to obtain.

Snake-History. Snakehood instantiations have caused Roberta to token M in the content-individuating way.

To cause Roberta to possess SNAKE is to cause Snake-History to obtain. Therefore, an interest in what caused Roberta to possess SNAKE is an interest in what caused Snake-History to obtain. Here we are specifically interested in whether Roberta’s possession of SNAKE is necessarily caused by the same snakehood instantiations that individuate the content of SNAKE by having caused Roberta to token M in some content-individuating way. Given that Roberta’s possession of SNAKE consists in the obtaining of Snake-History, this question amounts to the following: are the snakehood instantiations that have caused Roberta to token M in the content-individuating way and constitute Snake-History thereby also necessarily the causes of Snake-History?

There are prima facie reasons to think that they are not. It is commonly taken for granted that x is a cause of y only if, first, x and y are two separate things, and, secondly, x precedes y. The snakehood instantiations that constitute Snake-History meet neither condition in relation to Snake-History. As constituents of Snake-History, they are not separate entities from Snake-History nor do they occur before it. However, let us grant that these considerations can somehow be bypassed in the case at hand. There are still other reasons why the snakehood instantiations that constitute Snake-History need not be its causes.

To see this, let us take a look at the particular version of HCE defended by Rupert (1999). On Rupert’s account, Roberta’s M has content ˂snake˃ if and only if snakes are more “efficient” in causing Roberta to token M than members of any other natural kind. This is so, according to Rupert’s definition, if the ratio of the number of times snakes have caused Roberta to token M to the number of times snakes have caused Roberta to token any other M* is greater than the ratio of the number of times members of any other natural kind K have caused Roberta to token M to the number of times members of K have caused Roberta to token any other M* (Rupert 1999, pp. 323–325). So, given Rupert’s specification, snakes have caused M tokenings in the content-individuating way if and only if snakes have caused Roberta to token M most efficiently in this technical sense of the word. Therefore, given Rupert’s version of HCE, for Roberta to possess SNAKE is for Snake-History^Rup to obtain.

Snake-History^Rup. Snakehood instantiations have caused Roberta to token M most efficiently (in the above-described sense).

Now, what we want to know is whether the snakehood instantiations that have caused Roberta to token M most efficiently and thereby constitute Snake-History^Rup are, necessarily, also the causes of Snake-History^Rup.

In order to answer this question (negatively), we first need to specify the relevant sense of being a cause. What it means for one thing to cause another can vary with context. Our current context is the applicability of HCE to representations that are NEA-innate, i.e., to representations that are not caused to be acquired by an environmental stimulus. Hence, when assessing if the snakehood instantiations that constitute Snake-History^Rup (and thus SNAKE possession) are necessarily also the causes of Snake-History^Rup (and thus SNAKE possession), the relevant sense of being a cause is one that is relevant when assessing if SNAKE is NEA-innate. For only if the snakehood instantiations that constitute Snake-History^Rup are indeed the causes of Snake-History^Rup in this sense does it follow that SNAKE is not NEA-innate. As I have already argued (in Section 2), NEA-innateness should be understood in terms of a contrastive counterfactual dependence account of causation. Given this account, the snakehood instantiations that constitute Snake-History^Rup are also the causes of Snake-History^Rup insofar as whether Snake-History^Rup rather than some alternative history obtains (for instance, that elongated lizards, meandering cracks on a dry surface, or nothing at all have caused Roberta to token M) counterfactually depends upon these snakehood instantiations. The following scenario serves to demonstrate that it need not do so.

This scenario has three components: one actual, and two counterfactual. In all three components, Roberta is an organism with but one mental symbol M. As things actually stand, Roberta has been living in an environment where she has encountered many snakehood instantiations but no lizardhood instantiations. More specifically, she has encountered snakehood instantiations three hundred times. Whenever Roberta encountered a snakehood instantiation, and only when she encountered a snakehood instantiation, the snakehood instantiation caused her to token M. Snake-History^Rup therefore obtains (and Roberta possesses SNAKE, given Rupert’s account). The first counterfactual component of the envisaged scenario is then as follows. Had Roberta been living in a different environment, an environment where she had encountered lizardhood instantiations three hundred times but snakehood instantiations only five times, only snakehood instantiations would have caused her to token M—as in the actual component of our scenario. And just as in the actual component, Snake-History^Rup would obtain. Finally, the second counterfactual component of the envisaged scenario is as follows. Had Roberta been living in an environment where she encountered snakehood instantiations three hundred times but lizardhood instantiations only five times (just like in the actual component of the scenario), however, had she carried a different genome to the one she actually carried, one that is this time lizard-prone, only lizardhood instantiations would have caused her to token M. Therefore, Lizard-History^Rup would obtain.

In this scenario, whether or not Snake-History^Rup obtains does not depend upon the nature of Roberta’s environment. Snake-History^Rup would obtain and Roberta would possess SNAKE even if Roberta had encountered five snakehood instantiations instead of the three hundred snakehood instantiations that she actually encountered and that in the actual world constitute Snake-History^Rup and thus Roberta’s possession of SNAKE. So, in this scenario, the snakehood instantiations that constitute Snake-History^Rup do not count as the causes of Snake-History^Rup. Rather, the cause of Snake-History^Rup seems to be some genetic predisposition of Roberta to selectively token M in response to snakes rather than any other features of the environment, for example, lizards. That actually only snakehood instantiations caused Roberta to token M and thus that Snake-History^Rup obtains is the effect of this genetic predisposition. For had Robert had a different, lizard-prone genome, Lizard-History^Rup rather than Snake-History^Rup would obtain and Roberta would possess LIZARD.

Of course, it is necessarily true that had Roberta encountered no snakes, Snake-History^Rup would not obtain (nor would any other version of Snake-History). It is also necessarily true that had Roberta lived on Twin-Earth and encountered instantiations of twin-snakes (i.e., creatures superficially identical with snakes, but with a different chemical composition and which share no ancestry with earthly snakes) rather than snakes, then instantiations of twin-snakes rather than of snakes would have caused her to token M, and thus Twin-Snake-History^Rup would obtain rather than Snake-History^Rup. Does this not show that Snake-History^Rup is necessarily caused by the snakehood instantiations that constitute it? No. It only shows that whether Snake-History^Rup is caused by its constituent snakehood instantiations depends upon what the relevant causal alternative is to these snakehood instantiations. If the relevant alternative is, say, “no snakehood instantiations,” then the obtaining of Snake-History^Rup is indeed caused by the very snake instantiations that constitute it. The same is true if the relevant alternative is “twin-snakehood instantiations”. However, the following considerations serve to show that “twin-snakehood instantiations” and “no snakehood instantiations” need not be the relevant alternatives.

Here’s what speaks against twin-snakehood instantiations as the relevant alternative in an assessment of whether a version of Snake-History (including Snake-History^Rup) is caused by the snakehood instantiations that are its constituents. We are interested in whether the snakehood instantiations that constitute Snake-History are also necessarily the causes of the obtaining of Snake-History insofar as we are interested in whether a representation individuated in accordance with HCE (e.g., SNAKE) can be NEA-innate, i.e., not caused to be possessed by an environmental stimulus. Consequently we are interested in whether the snakehood instantiations that constitute Snake-History (and thereby SNAKE possession) are necessarily the causes of Snake-History in a sense that would make it impossible for SNAKE to be NEA-innate. This is so only if the alternative to the snakehood instantiations that constitute Snake-History (and in the case of which Snake-History would not obtain) is an alternative that would also be relevant when deciding whether SNAKE is NEA-innate. I discussed one plausible constraint on the relevant alternatives when deciding whether SNAKE is NEA-innate in Section 4.1. Here is another one. Nativist hypotheses, including the hypothesis that SNAKE is innate, are empirical hypotheses. As with empirical hypotheses in general, nativist hypotheses are concerned with what is possible in the actual physical world. So, the relevant alternatives that appear in empirical hypotheses are plausibly those that are at least nomically and biologically possible in the actual world (see e.g., Northcott (2008), p.118). This renders twin-creatures of little interest to such hypotheses.

Two considerations speak in favor of the possibility that sometimes “no snakehood instantiations” is not a relevant alternative in an assessment of whether the snakehood instantiations that constitute Snake-History are also Snake-History’s causes. First, to insist that “no snakehood instantiations” is always and necessarily a relevant alternative in such an assessment would lack principled motivation. For suppose that “no snakehood instantiations” is indeed a relevant alternative in an assessment of whether the snakehood instantiations that constitute Snake-History are also Snake-History’s causes. In this case, the claim “Snake-History is caused by the snakehood instantiations that constitute Snake-History” reads as “If instead of the snakehood instantiations that caused Roberta to token M in the content-individuating way no snakehood instantiations had caused Roberta to token M, it would not be the case that snakehood instantiations caused Roberta to token M in the content-individuating way.” This claim is true by definition. Therefore, to argue that “no snakehood instantiations” is necessarily a relevant alternative in an assessment of whether Snake-History is caused by the snakehood instantiations that constitute Snake-History is to argue that, necessarily, Snake-History is caused by the snakehood instantiations that constitute it. But it is hard to see what would motivate such a claim. As a rule, we recognize that phenomena need not be caused by their constituents. So why would Snake-History and its analogs be an exception? Second, to insist that “no snakehood instantiations” is necessarily a relevant alternative in an assessment of whether the snakehood instantiations that constitute Snake-History are also Snake-History’s causes and, therefore, that Snake-History is necessarily caused by the snakehood instantiations that constitute Snake-History amounts to insisting that SNAKE possession is necessarily caused by the snakehood instantiations that individuate the content of SNAKE. If generalized to other representations, this reasoning implies that HCE is but the thesis that the content of a representation is individuated by what causes its acquisition and is, therefore, but an instance of AE. So, insofar as we want to maintain that there is a meaningful distinction between AE and HCE, we should allow that “no snakehood instantiations” need not always be a relevant alternative when assessing whether the snakehood instantiations that constitute Snake-History and thus SNAKE possession are also its causes. If we allow this much, then we also allow that Snake-History need not be caused by the snakehood instantiations that constitute it, and thus that SNAKE can be NEA-innate (mutatis mutandis for other representations).

So, we have seen (by considering Rupert’s HCE) that from the fact that even if HCE is true of SNAKE then causal interaction with snakehood instantiations is constitutively necessary for SNAKE possession, it does not follow that according to HCE, causal interaction with snakehood instantiations is necessarily the cause of SNAKE possession.¹⁵ We have also seen (by considering Rupert’s HCE) that HCE can be true of SNAKE even if SNAKE possession is not caused by any other environmental stimulus, but rather by some genetic susceptibility to token M in response to snakes rather than anything else. So, we have seen (by considering Rupert’s HCE) that HCE can be true of SNAKE even if SNAKE is NEA-innate.

This is not to deny that there could be other versions of HCE that cannot be true of NEA-innate representations. A version of HCE cannot be true of NEA-innate representations if either (a) or (b) is true.

• (a) The version of HCE entails that if snakehood instantiations have caused Roberta to token M in the content-individuating way and thereby constitute Snake-History, then they are the causes of Snake-History. Such a version of HCE is but a version of AE.

• (b) (a) is not the case; however, the version of HCE entails that Snake-History and other representation-constituting histories are necessarily caused by some other environmental stimuli. For example, the theory might accommodate the possibility that Snake-History is caused by some prior disposition of Roberta; however, this disposition must be caused by an environmental stimulus. By transitivity of causation, Snake-History would then also be caused by an environmental stimulus.

Whether either (a) or (b) is true of a version of HCE depends upon how the particular version of HCE spells out what the “content-individuating way” is, and should be assessed on an individual basis. However, whether this is so or not will have nothing directly to do with the externalist aspect of the theory but other additional nuances of the particular version of HCE.

Let us also quickly discuss the relationship between HCE and P-innateness. Again, whether HCE can be true of P-innate representations depends upon how a given version of HCE specifies the “content-individuating way” in Snake-History. A version of HCE cannot be true of P-innate representations if, having filled in what the content-individuating ways is, Snake-History must be caused in a way that involves information processing. This is so if either (c) or (d) is true:

• (c) According to the version of HCE, snake instantiations have caused M tokenings in the content-individuating way only if they have caused M tokenings in a way that involves information processing.

• (d) (c) is not the case; however, there are some indirect reasons why Snake-History and other representation-constituting histories must be caused in a way that involves information processing. For example, the version of HCE allows that the obtaining of Snake-History is caused by some prior disposition of Roberta; however, this disposition itself must be acquired by means of processing information. By transitivity of causation, Snake-History would then also be caused in a way that involves processing information and, consequently, would not be P-innate.

One can find examples of HCE that fall under (c). Such includes all those instances of HCE according to which the content of M of an organism is individuated by what caused an organism to token M during the period in which the meaning of M was learnt (Dretske 1981; Ryder 2004).¹⁶ Learning is a paradigmatic example of acquisition via information processing. So, if the content of M is individuated by what caused the tokening of M in the context in which the meaning of M was learnt, then the corresponding representation that M is a vehicle of is acquired in a way that involves information processing and cannot be P-innate. However, it is far from true that all versions of HCE are learning theories in this sense. Consider again Rupert’s theory, according to which, M means <snake> if snakes have been most efficient (in the statistical sense outlined above) in causing Roberta to token M. As we saw, Rupert’s theory allows that the snakehood instantiations that caused Roberta to token M more efficiently than the instantiations of other natural kinds and thus individuate the content of M as <snake> did so because of some genetically caused sensitivity to snakes and without any psychological mediation. Allowing this, Rupert’s theory offers an example of HCE that implies neither (c) nor (d) and thus is applicable to P-innate representations.

5 Concluding Remarks

There is no inferential shortcut from externalism to non-innateness or from innateness to non-externalism. The implications of different types of causal externalist theory of mental content differ regarding their applicability to innate representations and need to be assessed on an individual basis. Only causal externalist theories according to which the content of a representation is determined by what caused its acquisition cannot be true of innate representations for conceptual reasons, and even then, this is only on the assumption that innate representations are those whose acquisition is not caused by an environmental stimulus. But this kind of causal externalism is not the strongest game in town and such a notion of innateness is rarely operative in existing representational nativist hypotheses. As regards more popular covariation theories of content and more prevalent innateness notions, the applicability of such a theory of content to such innate representations depends upon additional, often empirical, details pertaining to how plausible it is that a biological subject comes to be in the conditions that, according to such a theory of content, make it the case that her mental symbol has its particular content. As I argued, in the case of most such theories, a representation’s being innate is consistent with its content being externally individuated.

It is worth stressing that my discussion has not shown that R’s being innate never has any consequences for whether an externalist theory is true of R’s content or vice versa. It might. For example, having good reason to believe that the content of R is determined by some causal relation between the organism and its environment is a reason to believe that R is not P-innate insofar as the relevant causal relation can only be established via information processing. However, considerations like these do not specifically pertain to the applicability of externalist theories to innate representations. The organism-internal properties that, according to an internalist theory of content, determine the content of R—be it some head-internal inferential or causal roles, prototypes, definitions, or something else—are no more likely to be established by a brute-causal process than are such organism-environment relations that according to an externalist theory individuate the content of R.

That a representation can be both innate and have its content externally individuated has implications that may be counterintuitive. For example, this implies that two genetically identical individuals can possess different innate representations insofar as their environments differ in relevant aspects. This implication is a soft bullet to bite. Lay intuitions have little evidential standing in assessments of the implications of innateness hypotheses in science and naturalistic philosophy. And given the explanatory role that innateness plays in the psychological and biological sciences, the possibility that in different environments genetically identical organisms possess different innate representations is not a problem. Indeed, one explanatory role for hypotheses that state that some representations are innate is to explain why certain cognitive and behavioral traits of organisms of a given kind develop invariantly across a range of different environments; however, this range is always limited. Thus, that a given innateness hypothesis does not generalize to all environmental conditions is not in itself a shortcoming of the hypothesis.