To reform or to eliminate an attractor?

Abstract

Machery, E. (2021). A new challenge to conceptual engineering. Inquiry: An Interdisciplinary Journal of Philosophy.https://doi.org/10.1080/0020174X.2021.1967190.) argues to have identified a novel reason why deficient concepts of a certain kind are better eliminated than reformed. Namely, if the deficient concept is an attractor—a concept that we are psychologically drawn to think with—then eliminating this concept is a more feasible alternative than reforming it. Machery illustrates his argument with the example of the scientific concept of innateness, which he considers to be an attractor. I argue two things against Machery's thesis. First, there is little reason to believe that attempts to eliminate the scientific concept of innateness will be more feasible than attempts to reform it. Second, there is little reason to believe that attempts to eliminate a concept, whether it is an attractor or not, are more feasible than attempts to reform it. Moreover, a concept being an attractor even diminishes—instead of increasing—the chances that eliminating it is more feasible than reforming it.

1 Introduction

Scientists and philosophers have long questioned the epistemic legitimacy of employing the concept of innateness in scientific contexts (see Griffiths & Linquist, 2022, for an overview). Despite this, the concept continues to be used in various subfields of biological and psychological science. Against this backdrop, discussions and disagreements persist regarding what to do about the concept. Some think that the currently used concept of innateness should be reformed to better serve some legitimate epistemic function (Ariew, 1999; Cofnas, 2017; Khalidi, 2007; Northcott & Piccinini, 2018; Samuels, 2008).¹ Others think it best to stop using it– to eliminate it (Griffiths, 2002; Machery et al., 2019; Machery, 2021; Mameli, 2008; Shea, 2012). In a recent paper, Machery (2021) provides an argument in favour of elimination that turns on the psychological feasibility of the two kinds of projects. He argues that since the scientific concept of innateness is an “attractor”– a concept such that our minds are drawn to use this concept– any reformed version of it is unlikely to catch on because the targeted concept users, “attracted” by the original concept, would slip back into using it. To avoid such slippage, it is better to eliminate the concept altogether.

Machery’s argument bears not only upon debates around the scientific concept of innateness, which is but one of the many and various concepts used in different scientific and non-scientific domains that have been argued to be in need of either reform or elimination. He uses this case study to support a more general thesis: if the deficient concept is an attractor, elimination is to be preferred over reform as the more feasible, more implementable option. His thesis therefore has broader relevance for ongoing discussions concerning the implementability of different kinds of conceptual enhancement projects (e.g., as discussed by Cappelen, 2020; Fischer, 2020; Isaac et al., 2022; Jorem, 2021; Koslow, 2022; Nimtz, 2021; Thomasson, 2021). The possible ramifications of this thesis are the more relevant as many of the concepts argued to be problematic are quite plausibly attractors (for instance, racial concepts, the concept of truth).²

I think Machery has identified a genuine obstacle to reforming certain kinds of concepts, the concept of innateness included. However, I will argue that he has not identified an efficient way to elude this obstacle by opting for elimination. As I will also be focusing on the concept of innateness as a representative example, let me first introduce this case study before outlining my project and argument in more detail. This will also allow me to outline some concepts and premises for the coming discussion.

First, let’s clarify what is meant by ‘the scientific concept of innateness’. Machery commits to a psychological account of concepts. So will I. According to this account, the scientific concept of innateness is the mental body of information (set of beliefs and belief like states) that scientists by default (automatically and across a broad range of contexts) retrieve (activate in their minds) when applying or interpreting the word ‘innate’ in the context of doing science. It is the body of information that by default guides scientists’ innateness judgements and underlies the inferences they draw from innateness ascriptions. For example, suppose that a scientist is to decide if some birdsong B counts as ‘innate’ or that she isinformed that birdsong B is innate to some species. The mental body of information that by default guides her judgement of whether ‘innate’ applies to B, and that determines her interpretation of ‘innate’ and thereby which further inferences she draws from ‘B is innate to species S’ would be her concept of innateness. Let it be noted that this body of information need not align with scientists’ explicit beliefs about what it means to be innate (see Machery, 2021, p. 14). From now on, I will be using ‘concept’ and ‘the scientific concept of innateness’ in this psychological sense. If a concept relates to an expression (e.g., ‘innate’) in the way just described, I will say that this concept is ‘associated’ with this expression.³

Let ‘INNATE’ designate the actual scientific concept of innateness, the body of information that consists in some cluster of attributes that scientists currently associate with ‘innate’. Machery’s argument for the greater feasibility of eliminating over reforming INNATE is premised on the hypothesis that INNATE is nothing but the folk concept of being caused by the inner essence of the organism (Griffiths, 2002; Griffiths et al., 2009, and Machery et al.,2019, advance the same view). According to a widely (albeit not universally) accepted theory known as psychological essentialism, we humans are disposed to reason about biological organisms as if they possess some fixed, unobservable, kind-defining, causally potent inner “essence” lying deep inside such organisms; and we are disposed to reason about some traits of organisms as if they are caused by this essence (Gelman, 2003, 2009; Medin & Ortony, 1989). Such essence-caused traits, evidence suggests, we view as fixed in the sense that their nature and development is insensitive to environmental variation; as typical to organisms of the same kind (for example, to a species); as having function. Now, Machery draws on empirical evidence that the innateness judgements of both lay people and scientists are guided by a concept very much similar to this folk concept of being caused by the essence of an organism. For example, Griffiths et al. (2009) and Machery et al. (2019) found that the judgements of both non-scientists and scientists across a broad range of disciplines about whether a birdsong is innate were predicted by beliefs about whether this birdsong is fixed, species typical, and (to a lesser degree) functional.⁴ This observation led them to speculate that INNATE is nothing but this ingrained pretheoretical concept.

This hypothesis of identity is relevant according to Machery, as it threatens to render attempts to reform INNATE unfeasible. As he points out, the concept of being caused by essence is generally thought to be what he calls an ‘attractor’– a concept such that ‘its influence on thought is unintentional’, ‘it is more likely to determine people’s trains of thought than the other psychological structures that could influence them’ (Machery, 2021, p. 12) and it manifests itself ‘in convergences across individuals within a culture or across cultures’ (ibid., p. 13).⁵ This, he argues, predicts that attempts to make scientists use some reformed version of INNATE would likely fail, as scientists would just slip back into using INNATE, their minds being drawn to do so. In order to avoid such slippage, one should eliminate the concept from scientific contexts altogether.

Let’s grant that if a concept is an attractor then the targeted users of this concept are more likely to fall back onto using it than if the concept was not an attractor (I will say more about the plausible mechanisms accounting for this later). However, showing this much does not suffice to ground the further thesis that elimination is to be preferred to reformation as a more feasible option. What needs to be shown in addition is that scientists are significantly less likely to slip back into using attractors, such as INNATE, in response to attempts to eliminate these concepts. Conspicuously, Machery says close to nothing in the way of showing this. And as I will argue, there is little to support this thesis.

I will argue three things. First, Machery’s thesis that if a deficient concept is an attractor then the targeted concept users are less likely to revert to employing this concept in the case of elimination than in the case of reformation is as yet ungrounded. With some elimination and reformation projects, this might be the case, while with others it might not be. Importantly, there is nothing in the categories “elimination” and “reformation” that would suggest that the first alternative is more likely than the other. Second, I will argue that many (possibly all) actual and realistic proposals to eliminate INNATE are not more feasible in this regard than those of reforming INNATE. So, if INNATE is an attractor, this provides a counterexample to the general claim that if a concept is an attractor then eliminating the concept is more feasible than reforming it. Third, I will argue that the deficient concept being an attractor either does not affect the relative feasibilities of reformation and elimination or, in some cases, decreases the feasibility prospects of elimination relative to those of reformation. I begin in Sect. 2 by first reconstructing two (and what I believe to be the only) plausible reasons for thinking that concept elimination is more feasible than reformation, without distinguishing between attractors and non-attractors. In Sect. 3, I point out that these reasons rely on assumptions that need not be true in the case of individual or typical elimination and reformation projects. Whether they are or not depends on the nature of a concrete elimination scenario. In Sect. 4, I argue that these assumptions are not true in the case of many plausible proposals to eliminate or reform INNATE. In Sect. 5, I argue that these assumptions are less likely to be true in the case of attractors than in the case of non-attractors.

Upon which factors the psychological feasibility of some conceptual enhancement project depends can differ for scientific and lay contexts. However, despite focussing on the example of the concept of innateness used in scientific contexts, nothing in my discussion turns on the context being a scientific one.

2 Two plausible explanations why concept elimination is more feasible than concept reformation

Any explanation to why eliminating a concept might be more feasible than reforming it must have something to do with a feature or features that differentiate the two kinds of project. While the difference between the two is vague and rarely explicitly defined, the prescriptions of self-proclaimed concept eliminators and reformers typically diverge in two core aspects. I will display these aspects using the example of INNATE.

First, a reformer of INNATE argues that in some relevant scientific context, INNATE (the cluster of mentally represented attributes associated with ‘innate’) should not be used to draw inferences; instead, a different conceptual structure, CONCEPT*, should be used, such that CONCEPT* counts as a reformed version of INNATE (rather than a replacement of INNATE with an entirely new concept). An eliminator argues that neither INNATE nor any of its versions should be used in the relevant scientific context. Second, a reformer argues that we should continue using ‘innate’, the expression currently associated with INNATE in some scientific context, albeit to express the relevant reformed version of INNATE. The eliminator argues that we should not use ‘innate’, whether to express INNATE, some reformed version of INNATE, or any other concept. Some eliminators might insist that the terminological dimension is not essential to their project. What ultimately matters is ceasing to reason in terms of some mental category, regardless of how this category is labelled. But even so, most, if not all, proponents of eliminating some concept, including INNATE, also endorse terminological intervention alongside the conceptual, at least as an indispensable tool for achieving the more pressing goal. Furthermore, concept elimination is sometimes explicitly described as the thesis that some term should be abandoned (e.g., Haueis, 2021; Machery, 2009; Taylor & Vickers, 2017). Thus, I maintain that the terminological acceptance is a relevant difference between paradigmatic instances of concept elimination and reformation.

In the rest of the paper, I will be working with these characterizations. These characterizations are not intended as definitions. They are meant to demarcate what the agendas of self-proclaimed proponents of elimination or reformation with respect to a concept typically diverge on. Correspondingly, all generalisations made in this paper about concept elimination and reformation intend to be contingently true about actual projects pursued under these labels, and hypothetical projects insofar as they resemble these actual instances.

Prima facie, both of these differences can be argued to predict that scientists are more likely to slip back into using INNATE in response to attempts to reform the concept than to eliminate it. The crux of the argument is: by using ‘innate’ or some reformed version of INNATE we continue activating INNATE in the heads of scientists and thereby increase the risk that they will draw subsequent inferences in terms of INNATE rather than the relevant reformed version of it.

Begin with the reformer proposing to use a reformed version of INNATE in place of INNATE, and the eliminator proposing not to use any version of INNATE. The reader might have noticed that the above characterisation of concept reformation did not specify what makes CONCEPT* a reformed version of INNATE rather than an entirely new concept. This question of conceptual (dis)continuity is a vast topic in its own right but not one to be discussed at any length in this paper. However, I take this much to be uncontroversial that CONCEPT* is a reformed version of INNATE in virtue of being, in some relevant way and to some relevant degree, similar to INNATE. Given our assumption that concepts consist of bundles of represented attributes, such similarity plausibly amounts to CONCEPT* sharing some significant number of the core attributes (fixity, typicality or functionality) with INNATE. This means that whenever one mentally entertains some reformed version of INNATE, one also necessarily activates at least one of these core attributes of INNATE. This is relevant. For recall, according to our assumption, to possess a concept is to be disposed to co-activate certain attributes constitutive of the concept. Therefore, when individuals who possess INNATE activate some of its attributes– as they would do when mentally entertaining some reformed version of INNATE– this risks laterally activating the other constitutive attributes of INNATE as well, therefore INNATE as a whole (see Fischer, 2020, pp. 13–14). Call this argument for the greater feasibility of elimination compared to reformation Conceptual Similarity Argument.

Two actual proposals to reform INNATE illustrate this argument: that of Ariew (1999) and that of Cofnas (2017). According to Ariew, a trait (e.g., birdsong) is to be classified as innate in a species if it is environmentally canalised, i.e., its development is invariant across some relevant range of environmental variation. According to Cofnas, a trait should be classified as innate if it is a genetic adaptation. A trait being environmentally canalised is one way for the trait to be fixed. A trait’s being a genetic adaptation is one way for the trait to have a function. Thus, both Ariew’s and Cofnas’ reformed innateness concepts– INNATE^CANALISED and INNATE^ADAPTATION– share at least one core attribute with INNATE. Is the coactivation potential between fixity, typicality and functionality strong enough for the activation of, say, the attribute of fixity and thus INNATE^CANALISED to trigger the activation of these other attributes of INNATE as well? The studies of Griffiths et al. (2009) and Machery et al. (2019) suggest it is. These studies found that when birdsong was said to exhibit but one of these three attributes, subjects often judged this birdsong to be ‘innate’. Since all existing and realistic reformations of INNATE retain at least one of these attributes (see Griffiths et al., 2009), these findings suggest that they all pose a risk of triggering INNATE. Of course, this is not to claim that such triggering of INNATE by the activation of one or more of its attributes will happen in all contexts. As a minimal condition, the context would have to be one of reasoning about the development, kind membership, and inheritance patterns of biological organisms and their traits. However, this is guaranteed to be the context for employing any reformed version of INNATE.

Let’s now consider the second difference between elimination and reformation: the reformer proposes we continue using ‘innate’, while the eliminator proposes that we stop. This difference may also seem to predict that eliminating INNATE is more feasible than reforming it. Recall that ‘INNATE’ refers to the cluster of attributes that scientists by default activate in response to hearing, reading, mentally entertaining ‘innate’. It is therefore to be expected that if scientists continue to hear, read and mentally entertain the symbol ‘innate’– as they would in a reformation scenario– this will trigger them to activate INNATE due to the established psychological association between ‘innate’ and INNATE. Given that in an elimination scenario ‘innate’ is not used, this trigger is absent and the risk of activating INNATE thereby reduced. Call this second argument for the greater feasibility of elimination Linguistic Association Argument.

Note that the fact that INNATE continues to get activated in scientists’ minds is not a problem in itself. The problem, as both the reformer and eliminator see it, arises if the subsequent thoughts and inferences that scientists draw end up being thought and drawn using this concept. There are different ways to prevent this. One option is to suppress the deficient concept (or some of its problematic attributes) after its activation, thereby blocking it from influencing one’s further reasoning. Knobe and Samuels (2013) argue that this is precisely what occurs when scientists interpret innateness ascriptions in scientific contexts: upon encountering ‘innate’, they first retrieve the lay concept of innateness but then successfully ‘filter out’ some of its contextually irrelevant attributes.⁶ Alternatively, one might aim to prevent the problematic concept from being activated altogether. Both Conceptual Similarity Argument and Linguistic Association Argument focus on this earlier phase of cognitive processing.⁷

At first glance, both Conceptual Similarity Argument and Linguistic Association Argument may seem to provide decent reason to believe that attempts to reform a concept are less likely to succeed than attempts to eliminate it. However, two points need to be noticed about these arguments. Firstly, both arguments pertain to concepts in general, rather than attractor concepts specifically. Yet, Machery’s thesis is that elimination is more feasible than reformation with attractors in particular. This thesis is naturally interpreted as claiming that CONCEPT being an attractor increases the likelihood that the elimination of CONCEPT succeeds, while reformation doesn’t. For this to be so, the hypothesized effect of these two differences between elimination and reformation would need to be somehow amplified in the case of attractors. I address this matter in Sects. 6 and 7, arguing that this is not the case. Moreover, even if elimination enjoys some feasibility advantage over reformation with concepts in general, this advantage is reduced in the case of attractors. As a prerequisite for this, however, another point must be noticed about Conceptual Similarity Argument and Linguistic Association Argument. Namely, both arguments rely on a hidden substantive assumption which, as I will now argue, is not guaranteed to be true.

3 Assumptions of Conceptual Similarity Argument and Linguistic Association Argument

Take another look at the characterization of concept elimination from the previous section that Conceptual Similarity Argument and Linguistic Association Argument built upon. This characterisation is entirely negative. It specifies what should not to be the case: INNATE, any reformed version of INNATE, and ‘innate’ should not be used in scientific contexts. But it says nothing about what would or should be instead. Yet, many different things could be instead. Compare concept elimination with the abolition (i.e., elimination) of the institution of monarchy. Abolishing monarchy will always manifest in implementing some other than monarchic political system. But it can manifest in the implementation of significantly different political systems. Likewise, not using CONCEPT and ‘concept’ will necessarily manifest in some positive alternative scenario and can manifest in different positive scenarios. Call all such scenarios ‘elimination scenarios’. All scenarios of eliminating CONCEPT, as those of reforming CONCEPT, will involve revising the theory that currently includes CONCEPT, accompanied by a change in how the revised theory is linguistically expressed. These revisions would then manifest in targeted concept users employing some concepts other than CONCEPT and its reformed versions, and expressions other than ‘concept’ (see Fig. 1). However, the nature and scope of these revisions, and which other concepts and expressions would be employed instead of CONCEPT and ‘concept’, can vary across particular elimination scenarios.

Fig. 1

Concept reformation and concept elimination. (1.1) According to the reformer of CONCEPT, we shouldn’t be using CONCEPT but CONCEPT*, where CONCEPT* is a reformed version of CONCEPT; the term ‘concept’, currently associated with CONCEPT, should continue to be used but to express CONCEPT*. A reformer might support pluralism and argue that ‘concept’ should be used to express different reformed versions of CONCEPT in different contexts.In this case, CONCEPT*₁ and CONCEPT*₂ represent different reformed versions of CONCEPT meant to be used in different contexts. Alternatively, a reformer might argue that ‘concept’ should be used to express the same reformed version of CONCEPT in all contexts. In this case, CONCEPT*₁ and CONCEPT*₂ represent different rivalling candidates for this reformation. (1.2) The eliminator proposes that we stop using both CONCEPT and ‘concept’. Instead, we should start or continue to use some concept CONCEPT* that is not a reformed version of CONCEPT and express CONCEPT* with some different expression ‘concept*’. An eliminator might argue that in different contexts, CONCEPT should be replaced with different concepts, each expressed with different expressions. In this case,CONCEPT*₁, CONCEPT*₂, ‘concept*’₁ and ‘concept*’₂ represent different replacements of CONCEPT and ‘concept’ in different contexts. Alternatively, the eliminator might prescribe replacing CONCEPT with the same concept in all contexts. In this case, CONCEPT*₁, CONCEPT*₂, ‘concept*’₁ and ‘concept*’₂ represent rivalling candidates for an appropriate replacement.

The fact that eliminating CONCEPT always has a specific, albeit not always clearly articulated, positive counterpart is important. Because whether eliminating CONCEPT is more feasible than reforming it– as per Conceptual Similarity Argument and Linguistic Association Argument– depends entirely on what this positive counterpart is. For instance, in a scenario where INNATE is eliminated and no reformed version of INNATE is used, the risk of activating INNATE is reduced only if none of the concepts used in place of INNATE in this scenario are as likely to trigger INNATE as the relevant reformed version of it suggested by a reformer. Similarly, refraining from using ‘innate’ reduces the risk of activating INNATE only if none of the expressions used in a given elimination scenario is as likely to trigger INNATE as ‘innate’. In general terms, not using a reformed version of CONCEPT and ‘concept’ reduces the risk of targeted concept users activating CONCEPT only if No Conceptual Trigger and No Linguistic Trigger are true of concrete elimination and reformation scenarios.

No Conceptual Trigger

No concept used in an elimination scenario is as likely to trigger CONCEPT as the relevant reformed versions of CONCEPT.

No Linguistic Trigger

No expression used in an elimination scenario is as likely to trigger CONCEPT as ‘concept’.

And there is no a priori reason why No Conceptual Trigger and No Linguistic Trigger couldn’t be false in the case of some or even most elimination and reformation scenarios. All concepts share attributes with many other concepts (KITTEN shares attributes with CAT, as does WILDCAT and TIGER). If sharing attributes with CONCEPT is the reason why some CONCEPT* triggers the activation of CONCEPT, all concepts are potentially triggered by multiple other concepts. It is also evident that the same concept can be triggered by different expressions (‘cat’, ‘kitty’, ‘wildcat’ can all trigger CAT). And there is no reason yet to rule out the possibility that a scenario of eliminating CONCEPT involves using concepts and expressions that are among those likely to trigger CONCEPT.

One might object that if CONCEPT* is a reformed version of CONCEPT, it is necessarily more similar to and thus more likely to trigger CONCEPT than if CONCEPT* is not a reformed version of CONCEPT. But this objection is not convincing. It assumes that the kind and degree of similarity that defines CONCEPT* as a reformed version of CONCEPT is the same that determines how likely CONCEPT* is to trigger CONCEPT. But why assume this? One option is to make it true by definition: define CONCEPT* as a reformed version of CONCEPT only if it is similar to CONCEPT in such a wat that suffices for CONCEPT* to trigger CONCEPT. In this case, Conceptual Similarity Argument becomes a conceptual truth. But this truth would be boring because it would be uninformative regarding the relative feasibility prospects of actual projects pursued as ‘concept reformation’ and ‘concept elimination’. Most of these projects do not define (if they define it at all) ‘is a reformed version of CONCEPT’ in terms of how likely CONCEPT* is to trigger CONCEPT. Moreover, defining the relation ‘reformed version of CONCEPT’ in this manner would get the extension of ‘reformed version of CONCEPT’ scandalously wrong. It is well known that other factors beyond conceptual similarity contribute to whether one concept activates another: semantically related concepts like HAMMER and NAIL can activate each other; antonyms like DARK and LIGHT can activate each other etc. No account that renders NAIL a reformed version of HAMMER or DARK a reformed version of LIGHT can be correct.

Secondly, one might object that the eliminator of CONCEPT always proposes to abolish the expression that is most strongly associated with and thus most likely to trigger CONCEPT– the expression that has CONCEPT as its dominant psychological meaning. But again, whether this is actually the case is an empirical question rather than something warranted or even made likely by the label ‘elimination’. Moreover, strength of association between an expression and a concept is not always the main determinant of which concept the expression triggers on a given occasion or in some special contexts (e.g., some scientific contexts). Expressions often trigger concepts that are not their dominant psychological meanings, namely when the context of using some expression encourages the activation of such concepts.

In summary, whether No Conceptual Trigger and No Linguistic Trigger are true of concrete pairs of elimination and reformation scenarios, or even typically true, is an open question that cannot be assessed without investigating the specifics of particular elimination and reformation scenarios. Next, I will exemplify the claims made in this section with some actual and plausible proposals to eliminate and reform INNATE. In the same instance, I will argue that many representative proposals to eliminate INNATE do not meet No Conceptual Trigger and No Linguistic Trigger, and thus are not more feasible than existing proposals to reform INNATE.

4 No Conceptual Trigger and No Linguistic Trigger in the case of INNATE

Mameli and Bateson (2006, 2011) endorse the elimination of the scientific concept of innateness. They argue that in different theoretical contexts the currently used concept of innateness is appropriately replaced with different concepts: in some contexts with CANALISED, in other contexts with ADAPTATION, to name some main examples. Now, it should not escape notice that these two proposed replacements for INNATE– CANALISED and ADAPTATION– are not dissimilar to INNATE. In fact, they are just as similar to INNATE as the earlier discussed reformations of INNATE proposed by Ariew and Cofnas, INNATE^CANALISED and INNATE^ADAPTATION, where either retains at least one attribute of INNATE (fixity and functionality respectively). As such, these replacements of INNATE should be equally likely to trigger INNATE as INNATE^CANALISED and INNATE^{GENETIC ADAPTATION}. I suspect that this example illustrates a more general phenomenon. I suspect that often we will find little, if any, difference between the concepts that a reformer on one hand and an eliminator on the other hand propose to use in place of INNATE (similarly with other concepts), so that any substantial disagreement between the two boils down to disagreement about which term to use to express the same concept. Even if these concepts differ in some respects, they are unlikely to differ in terms of their similarity to INNATE.

The elimination scenario of Mameli and Bateson is one where the target theory is revised “locally”: INNATE is expelled from a theory, a different concept (such as ADAPTED, CANALIZED, etc.) then occupies its position while the rest of the theory remains largely intact. But we also find more revisionary projects pursued under the agenda of eliminating INNATE. For example, Griffiths and Gray (1994) and Oyama (1985) argue that the elimination of INNATE should follow from adopting the developmental systems theory (DST), which advocates for a more encompassing conceptual restructuring of the established theories of development and evolution. Now, ceteris paribus, the more revisionary an elimination scenario the greater its prospects of not containing concepts as likely to trigger INNATE as its reformed versions. CONCEPT* can only come to occupy a position in a theory formerly occupied by INNATE– a position defined by relations to other concepts, explanatory role or extension– if it is in some way similar to INNATE. Therefore, any CONCEPT* that assumes the position formerly occupied by INNATE will likely share some attributes with INNATE. In a more thoroughly revised theory where no concept replaces INNATE specifically, one might argue, all concepts would be less similar to and thereby less likely to trigger INNATE. In this light, insofar as DST refrains from using any such categories that have replaced the dichotomous innate-acquired distinction specifically, it has, everything else being equal, better prospects of meeting No Conceptual Trigger than, for instance, the elimination scenarios of Mameli and Bateson (this reasserts the earlier point that whether an elimination scenario meets No Conceptual Trigger can vary). But ultimately, whether a specific revised theory contains concepts that share those attributes with the deficient concept that risk activating it can only be determined by scrutinising the specifics of that theory. For instance, while in many respects a substantial theoretical revision of the theory currently embedding INNATE, DST might still utilize concepts that refer to the fixity, typicality, or functionality of traits (e.g., the concept of being an adaptation, of being environmentally canalized). One could even reasonably argue that no plausible revision of a theory currently incorporating INNATE (such as some theory concerning the development and functioning of the immune system or some cognitive capacity) can do away with such concepts. As I will elaborate in Sect. 6, this possibility is exacerbated if INNATE is an attractor.

Furthermore, expressions that one is likely to use in a scenario of eliminating INNATE, expressions suitable to name the replacements of (or leftovers from) INNATE, are likely to trigger INNATE if ‘innate’ is. Consider some empirical evidence for this.

Mameli and Bateson (2006, 2011) observe that in different scientific contexts, ‘innate’ is used synonymously with expressions such as ‘canalized’, ‘genetically caused’, ‘evolved’, ‘present at birth’, ‘Darwinian adaptation’, and others. This observation can be interpreted in two ways. Either the synonymous use of ‘innate’ with these expressions is explained by the fact that ‘innate’ is sometimes used and interpreted to express CANALIZED, sometimes GENETICALLY CAUSED, sometimes EVOLVED, etc. This hypothesis predicts that ‘innate’ will occasionally activate one or other of these concepts. Alternatively, it is explained by the fact that ‘canalized’, ‘evolved’, ‘genetically caused’ etc. are sometimes used and interpreted to express INNATE. This second hypothesis predicts that these other expressions sometimes trigger INNATE, just like ‘innate’. How much of such synonymous use of ‘innate’ with each of these expressions either hypothesis accounts for is an empirical question. But quite likely, both hypotheses account for some proportion of it. This in turn predicts that at least sometimes these other expressions trigger INNATE just like ‘innate’ (I will return to this in Sect. 7).

Another body of evidence comes from a study by Linquist et al. (2011). Linquist and colleagues tested the extent to which ‘innate’, ‘in the DNA’, ‘part of nature’ are interpreted in terms of the folk concept of being caused by an organism’s inner essence (which we assume to be identical with INNATE). They found that ‘in the DNA’ was interpreted through this concept more frequently than ‘innate’. This finding is consistent with other research on how people interpret information about genes and genetically caused traits. For instance, according to a prominent view called “genetic essentialism”, people in contemporary societies– both laypeople and scientists– implicitly believe genes to be the material locus of an organism’s inner essence (Dar-Nimrod & Heine, 2011; Gould & Heine, 2012; Heine, 2016). Correspondingly, they tend to view traits they perceive as genetically caused as if caused by this essence and therefore as being fixed, typical of a group, and functional. Evidence for this hypothesis largely comes from studies comparing how people interpret the information that a trait has genetic causes versus the information that it has environmental causes. These studies do not directly test whether, and which, specific expressions encourage essentialist interpretation of genetic information. However, since information about genetic causes is presented in natural language, the findings of these studies suggest at least this much: some expressions (e.g., ‘has genetic causes’) activate the concept of originating from inner essence, i.e., INNATE.

Different factors can explain these observations, probably acting together. First, these and other expressions might activate INNATE in virtue of being associated with, and are thus prone to trigger, concepts like CANALIZED, ADAPTED, GENETICALLY CAUSED, where the latter, in turn, are prone to trigger INNATE– as hypothesized earlier. Alternatively, these expressions might be directly associated with INNATE. Many natural language expressions are polysemic in that they are associated with multiple semantically related concepts. Which of these concepts an expression triggers on a given occasion is influenced by two main factors. First, it is influenced by the strength of association between the expression and a given concept, where strength of association is a function of how frequently the expression has been used and interpreted to express this concept in the past. The second factor is context– sometimes an expression triggers a concept that is not most strongly associated with it, or perhaps not associated at all. This happens if the context of using the expression contains some feature that independently encourages the activation of this concept (whether it is some feature of the physical environment, other expressions uttered nearby the target expression, background beliefs of the interpreter, or other) (Fischer, 2020b; Giora, 2003). Now, quite likely, INNATE is already among the set of concepts associated with ‘in the DNA’, ‘canalized’ etc., even if it is not the concept most strongly associated with them. This means that if the context is conducive to it, these expressions would activate INNATE. The context of interpreting the substitutions for ‘innate’ in an elimination scenario will likely share at least three features that are conducive to the activation of INNATE. The first is the general theoretical context which at least in the local kinds of elimination scenarios would not have changed much. The second feature is the subject matter. According to the received psychological essentialism theory, one general trigger of employing the concept of inner essence is simply engaging in reasoning about issues concerning the development, inheritance patterns, kind membership of biological organisms. This course-grained subject matter is inevitably part of any elimination scenario in which one or another replacement for ‘innate’ would be interpreted. The third feature, assuming INNATE is an attractor, is the very cognitive bias of the interpreters of these replacements to think in terms of INNATE rather than related alternatives. All this makes it plausible that just like ‘innate’, the different expressions considered sometimes do trigger, and would trigger, INNATE.

Let me summarise my reasoning thus far. I began the paper by outlining two reasons why one might argue that concept elimination is more feasible than concept reformation. I made explicit two assumptions underlying these reasons (No Conceptual Trigger and No Linguistic Trigger), and argued that these assumptions may not be true in the case of individual elimination and reformation proposals. I then illustrated this claim with the example of INNATE, arguing that in the case of some paradigmatic proposals to reform or eliminate INNATE, these assumptions don’t seem to be true. This serves to show that Machery’s claim that eliminating INNATE is more feasible than reforming it is false if disambiguated as ‘all existing and plausible proposals to eliminate INNATE are more feasible than proposals to reform INNATE’, unlikely to be true if disambiguated as ‘most existing and plausible proposals to eliminate INNATE are more feasible than proposals to reform INNATE’, and probably true only if interpreted to claim something as uncontroversial as ‘some existing and plausible proposals to eliminate INNATE are more feasible than proposals to reform INNATE’.

However, Machery also defends a second, more general thesis: if the deficient concept is an attractor, the strategy of eliminating the concept is more feasible than reforming it. Up until now, my arguments have been indifferent to whether the deficient concept is an attractor and thus have not directly addressed this thesis. Indeed, if INNATE is an attractor, we have a counterexample to this thesis if read as a universal generalisation. However, could the fact that a concept is an attractor still increase the chances that eliminating this concept succeeds while reforming it does not? In other words, does being an attractor increase the feasibility of elimination relative to reformation? Given the set-up of this discussion, this is so if No Conceptual Trigger or No Linguistic Trigger are more likely to be true in the case of attractors than non-attractors. I now proceed to argue that they are not. If anything, they are less likely to be true in the case of attractors. I begin with a short note about the category “attractor” that my arguments will assume.

5 What is an attractor?

The thesis that if CONCEPT is an attractor then eliminating CONCEPT is to be preferred to reforming it as a more feasible alternative assumes that the category “attractor” is cohesive enough to support a generalization like this. Machery’s explicit characterization leaves it opaque which set of features it is that most attractors have in common and that support this generalisation. According to his characterization, a concept is an attractor if ‘its influence on thought is unintentional’, ‘it is more likely to determine people’s trains of thought than the other psychological structures that could influence them’ (Machery, 2021, p. 12) and it manifests itself ‘in convergences across individuals within a culture or across cultures’ (ibid., 13). But this characterization is vague in several respects: In how large convergence? How much more likely? More likely than which ‘other psychological structures’ (for, supposedly, many actually used concepts are more likely to determine people’s trains of thought than some other concepts)? What explains this greater likelihood? Answers to these questions potentially have implications not only for the extension of ‘attractor’ but also for which generalizations about attractors are true. This includes generalizations regarding whether projects of eliminating an attractor are more feasible than those of reforming it, and why.⁸ Here, my aim is not to resolve this vagueness. My aim is to acknowledge that it can be resolved in different ways and pin down which features my following discussion will assume attractors to share. I will assume that if CONCEPT is an attractor then:

• 1) the disposition to activate the cluster of attributes that constitute CONCEPT is stronger than the dispositions to activate most other contextually viable alternative clusters of attributes; stronger in the sense that, under similar circumstances, this cluster is more likely to get activated than these other clusters.

• 2) the existence and strength of this disposition is explained by some universal endogenous feature of human cognitive architecture. This in turn predicts that the disposition manifests in convergences across individuals within and across cultures.⁹

According to the dominant view, the folk concept of being caused by organism’s inner essence (thus INNATE) meets (1) and (2). Other plausible candidates are concepts sometimes included in ‘core cognition’, for example AGENCY, CAUSE, FACE (Carey, 2009), and concepts central in our folk theories such as FORCE, RACE, BELIEF.

6 Is No Conceptual Trigger more likely to be true in the case of attractors than in the case of non-attractors?

As argued in Sect. 3, a scenario of eliminating CONCEPT is more feasible than a scenario of reforming it if No Conceptual Trigger is true of these scenarios, that is, if no concept used in the particular elimination scenario is as likely to trigger CONCEPT as the relevant reformed version of CONCEPT. I also argued that whether this is true cannot be predicted based on knowing that one is dealing with elimination rather than reformation. With some elimination-reformation scenario pairs, No Conceptual Trigger may be true, with others it might not. Let’s now suppose that CONCEPT is an attractor. Does this somehow increase the chances that no concept used in a scenario of eliminating CONCEPT is as likely to trigger CONCEPT as its reformed versions? At first glance, it is hard to see why it would make any difference at all. And here’s the reason.

Throughout the paper I have assumed that a conceptual structure CONCEPT* used in either an elimination or reformation scenario triggers CONCEPT in virtue of sharing some core attributes with CONCEPT. I did not specify how many and which attributes. This is an empirical matter. However, if CONCEPT is an attractor, we should expect that CONCEPT* needs to share fewer attributes with CONCEPT to trigger CONCEPT than if CONCEPT* were not an attractor. This follows from the above notion of attractorship. According to this notion, CONCEPT is an attractor only if some relevant group of people (e.g., all humans) across some relevant set of contexts are more disposed to activate CONCEPT rather than some plausible alternative concepts. This amounts to saying that, within this relevant group and set of contexts, the prior probability of activating an attractor is higher than that of activating a non-attractor. Given this, the conditional probability of activating an attractor on activating any number n of its attributes is also higher, meaning that the probability that CONCEPT* sharing n attributes with CONCEPT triggers CONCEPT is higher if CONCEPT is an attractor than if it is not an attractor. This predicts that if CONCEPT is an attractor, it would be triggered also by some such concepts that share relatively few attributes with it and wouldn’t trigger it if it were not an attractor. This prediction is consistent with the observation that probable attractor concepts appear to be triggered by impoverished stimulus. For instance, we appear to retrieve the concept of agency (a plausible attractor) merely when observing as-if goal oriented physical movement (Barrett, 2004; Keil & Newman, 2015); we appear to activate the concept of a face (a plausible attractor) merely when observing higher stimulus density in the upper part of a configuration (Simion & Giorgio, 2015), to list but some examples. It is also consistent with the findings of Griffiths et al. (2009) and Machery et al. (2019) that the concept of being caused by inner essence (INNATE) (by our assumption an attractor) was often triggered by the activation of but one of its core attributes. This observation may well be explained by INNATE being an attractor and the prior probability of activating INNATE in the context of biological reasoning being relatively high.

So, if CONCEPT is an attractor, the likelihood that some conceptual structure CONCEPT* sharing n attributes with CONCEPT triggers CONCEPT is higher compared to when CONCEPT is not an attractor. As a result, all reformed versions of CONCEPT sharing n attributes with CONCEPT are more likely to trigger CONCEPT than if CONCEPT were not an attractor and, consequently, all plausible proposals to reform it are to some degree more likely to fail. And from this one might be tempted to infer that the chances that some reformation of CONCEPT is more likely to trigger CONCEPT than its replacement in an elimination scenario are also higher. If this inference were correct, it would be true that No Conceptual Trigger is more likely in the case of attractors than non-attractors and thus elimination more feasible than reformation. But this inference is not correct. Because as much as being an attractor increases the chances of some reformed versions of CONCEPT triggering CONCEPT, it also, and for the same reasons, increases the chances of some CONCEPT* employed in an elimination scenario doing so. For example, if INNATE is an attractor, the chances are increased that also those concepts used in more revisionary elimination scenarios like DST share enough attributes with INNATE to trigger it, even if sharing the same number of attributes with INNATE would not be enough to trigger it if this concept were not an attractor. In sum, CONCEPT being an attractor does not seem to affect how likely No Conceptual Trigger is to be true of attempts to reform or eliminate CONCEPT. The difference between the feasibility of an elimination and a reformation project (if one there is in the first place) stays the same regardless of whether or not CONCEPT is an attractor.

But a case can be made for an even stronger thesis: that No Conceptual Trigger is even less likely to be true in the case of attractors than non-attractors. I have been stressing it as a plausible possibility that with many pairs of proposals to eliminate or reform CONCEPT, No Conceptual Trigger is not true. However, I also acknowledged that with some such pairs it might be (remaining agnostic about how likely either alternative is). Let’s suppose that we are dealing with a pair of elimination and reformation scenarios of which No Conceptual Trigger is indeed true. Specifically, let's suppose that:

• i) The proposed reformation of CONCEPT shares enough attributes with CONCEPT to definitely trigger its activation and

• ii) no concept used in the elimination scenario shares enough attributes with CONCEPT to trigger its activation.

Let’s stipulate that CONCEPT is not an attractor. Now imagine the consequences if CONCEPT was turned into an attractor. As explained above, if CONCEPT is an attractor, we can expect that sharing fewer attributes with it suffices for some other conceptual structure to trigger its activation compared to if CONCEPT were not an attractor. This predicts that if CONCEPT in our described scenario were turned into an attractor, at least some of the concepts used in the above elimination scenario that didn’t pass as similar enough to CONCEPT to trigger it when CONCEPT was yet not an attractor, would now do so. This scenario is an example where being an attractor reduces the likelihood that No Conceptual Trigger is true. It is an example where the deficient concept, CONCEPT, being an attractor increases the likelihood that some CONCEPT* used in an elimination scenario triggers CONCEPT, however, does not increase the corresponding likelihood for the reformations of CONCEPT. This is because (i) stipulates that the relevant reformation of CONCEPT would trigger CONCEPT anyway, even if it were not an attractor. If this example is representative, and I believe it is, No Conceptual Trigger is less likely to be true in the case of attractors than in the case of non-attractors.

7 Is No Linguistic Trigger more likely to be true in the case of attractors than in the case of non-attractors?

As in the case of No Conceptual Trigger, the fact that the deficient concept is an attractor either does not impact whether No Linguistic Trigger is met, or makes it less likely. Take INNATE. Since INNATE being an attractor increases the base probability of its activation, it also increases the chances that using ‘innate’ will continue to trigger INNATE. But it also increases the chances that the other expressions used in an elimination scenario to name plausible replacements of INNATE (e.g., ‘adaptation’, ‘heritable’, ‘canalized’) would do so. Whatever the respective probabilities of ‘innate’ triggering INNATE or some relevant other expression, for example ‘canalized’, triggering INNATE on the condition that INNATE is not an attractor, one would assume (in the absence of argument to the contrary) that these probabilities increase equally for both ‘innate’ and ‘canalized’ on the condition that INNATE is an attractor. But similarly to what was said in relation to No Conceptual Trigger in the previous section, a case can be made that INNATE (or any other concept) being an attractor even diminishes the prospects of No Linguistic Trigger being met. It does so by making it more likely that the alternatives to ‘innate’ used in a scenario of eliminating INNATE would trigger INNATE. For two reasons.

As mentioned earlier, natural language expressions are typically associated with multiple semantically related concepts. The first reason why a scenario of eliminating an attractor is more likely to involve using expressions that would trigger its activation is that attractors are likely to be associated with (and thus be triggered by) more linguistic expressions than non-attractors. As per our assumption, attractors are often cultural or cross-cultural universals partly in virtue of some universally shared cognitive disposition. Given this, we can assume two things about attractors: (i) They are employed in a relatively broad range of discursive contexts (both lay and scientific contexts), (ii) They are employed by many (possibly all) people from a variety of subgroups of speakers of a language or languages. In different discursive contexts and among different subgroups of speakers of a language, language use varies– different “dialects” are spoken, different expressions used. Therefore, in these different contexts, the same concept is likely to latch onto different expressions and thus become associated with multiple expressions (see Christiansen & Chater, 2008; Neeleman & van de Koot, 2012; Rose et al., 2021, for some indirect evidence). (i) and (ii) thus predict that the number of expressions associated with an attractor is typically larger than that associated with a non-attractor which are typically used in fewer and less heterogenous linguistic contexts, or so the hypothesis. The more there are semantically related natural language expressions that speakers of a language associate with CONCEPT, the more likely it is that one or more of them will be employed in a scenario of eliminating CONCEPT to name the replacements of CONCEPT.

The second reason has to do with the context dependency of linguistic interpretation. As also pointed out, which of the multiple concepts associated with an expression, ’concept’, this expression triggers on a given occasion depends on two factors. First, it depends on how strongly ‘concept’ is associated with each of these concepts– the concepts that are more strongly associated with ‘concept’ would, ceteris paribus, be triggered more frequently. Second, it depends on the context of interpretation– sometimes, a given expression can trigger a concept that it is not the one associated with it most strongly, namely if something about the context of interpretation encourages employing this concept. Suppose that ‘concept’ (the expression to be eliminated) and ‘concept*’ (an expression used in an elimination scenario instead of ‘concept’) are both associated with CONCEPT, i.e., both have CONCEPT among their psychological meanings. And suppose that the context of using ‘concept*’ in an elimination scenario is conducive to employing CONCEPT instead of other concepts that are also associated with ‘concept*’. If so, ‘concept*’ might trigger CONCEPT just like ‘concept’ despite the fact that it is more strongly associated with some of these other concepts. If CONCEPT is an attractor, this scenario is more likely than if CONCEPT is not an attractor. A concept is an attractor only if there is an endogenously caused universal disposition to preferentially use this concept across some relevant range of contexts. As a feature of interpreting subjects, this disposition is a constant feature of the context of interpreting any expression used in an elimination scenario that will travel with them from one interpretation occasion to the next. Secondly, the external contextual cues that encourage the activation of an attractor are likely to be coarse-grained. As such, they are more likely to constitute the context of interpreting whichever alternative to'concept' gets to be used in an elimination scenario. For instance, as mentioned earlier, one contextual feature that according to the received version of psychological essentialism encourages the use of INNATE is the subject matter of the development of biological organisms. This subject matter, just like the endogenously rooted cognitive bias to think in terms of INNATE, is bound to be part of the context of interpreting any expressions used in a scenario of eliminating INNATE and nudge individuals to interpret even those expressions that are less strongly associated with INNATE in terms of INNATE.

In this light, let’s reconsider the observation of Mameli and Bateson (2006, 2011) that in different scientific contexts, ‘innate’ is often used and interpreted synonymously with expressions like ‘canalized’, ‘genetically caused’, ‘evolved’, and others (see Sect. 4). I considered two possible complementary explanations to this observation: (1) all these expressions are sometimes used and interpreted to express INNATE, and (2) ‘innate’ is sometimes used and interpreted to express CANALIZED, GENETICALLY CAUSED, EVOLVED etc. Given the reflections just given, the hypothesis that INNATE is an attractor suggests that (1) accounts for a relatively larger segment of the synonymous use of these expressions. This, in turn, increases the likelihood that these expressions, when used in an elimination scenario, would trigger INNATE.

All this suggests that if CONCEPT is an attractor, the chances are higher that even those expressions used in some elimination scenario that would not function as triggers of CONCEPT if CONCEPT is not an attractor would function as such if it is, either because of being associated with it and/or because of context effects. Moreover, recall that, as per common assumption, strength of association between an expression and a concept is a function of how frequently this expression has been employed together with this concept in the past. Therefore, if ‘concept*’, initially not strongly associated with CONCEPT, keeps triggering CONCEPT due to context effects, this can result in ‘concept*’ becoming more and more strongly associated with CONCEPT, up to the point of becoming its primary meaning. Grossi (2017) gives an example. ‘Hardwired’, now widely employed in mind and brain sciences, was initially introduced to replace ‘innate’ and to express a concept different from that expressed by ‘innate’. However, as Grossi argues, it is now used to mean something very similar to ‘innate’. This example should in general curb the eliminator’s hope that replacing ‘concept’ with ‘concept*’ would result in us not thinking and reasoning with some defective concept currently associated with ‘concept’. In reality, what might change instead is which expressions we associate with the same deficient concept. Terminological change might even veil the fact that the problematic concept continues to be used, thereby facilitating, rather than blocking the use of this concept.¹⁰

8 Conclusion

This paper examined and questioned the thesis recently defended by Machery that if some problematic concept is an attractor, such as the concept of innateness, then eliminating this concept is a more feasible and thereby preferable way to stop people from using this concept than reforming it. I argued three things in response to this thesis.

First, while Machery’s claim that a concept being an attractor is a challenge to reforming it is convincing, it is far from clear that it is less of a challenge to eliminating the concept. Is this so or not depends on the specifics of concrete elimination and reformation scenarios. In particular, it depends on whether a given scenario of eliminating a problematic concept involves using concepts and expressions as likely to trigger the activation of this concept as its relevant reformed versions and the expression the eliminator proposes to ban. This can vary with different elimination and reformation scenarios. And there is no reason why an average elimination scenario would not involve using such concepts and expressions. I also argued that the problematic concept being an attractor either does not impact how likely an elimination scenario is to involve the use of such concepts and expressions, or quite plausibly increases the likelihood that it does, thereby reducing the chances that the elimination of this concept would be more feasible than its reformation.

I illustrated these claims with Machery’s own case study– the scientific concept of innateness. I showed that some paradigmatic (and, by extension, plausible) scenarios of eliminating the concept of innateness from scientific discourse do involve the use of concepts and expressions as likely to trigger the retrieval of this concept as many of its proposed reformations and'innate'. This makes these elimination projects equally vulnerable to the risk that scientists slip back into thinking in terms of the original, deficient, concept. It also means that regardless of whether a given attempt to remove the currently employed concept of innateness from circulation counts as an attempt of elimination or reform, it faces the need of finding means to either prevent certain expressions and concepts from activating the deficient concept or supress it once activated. Going for elimination might even do the disservice of veiling this need and thus the fact that the old problematic concept insidiously continues to influence our reasoning.

Let me stress that all this is not to deny that some conceptual enhancement projects are psychologically more feasible than others, nor the relevance of this when deciding which of these projects are preferable. What I have suggested is that appealing to the elimination-reformation alternative in making such decisions is unproductive. This alternative fails to track the factors upon which the feasibility of different conceptual enhancement strategies depends, both in the case of attractors and non-attractors. Organising the discussion of the relative feasibility advantages of different conceptual enhancement projects around these categories only distracts from giving due diligence to examining and locating these factors.