The power of social norms: Why conceptual engineers should care about implementation

Abstract

Jennifer Nado has recently argued that conceptual engineers should focus on (re-)designing representations and may safely ignore issues of implementation. I make a general case for the methodological importance of implementation to conceptual engineering. Using the Social Norms Account as a foil, I argue for three claims. (1) Inquiring into methods of implementation is a theoretically challenging and philosophically worthwhile project in and of itself. (2) A sound theoretical understanding of implementation is imperative for theorists of conceptual engineering. It proves vital for their assessment of the prospects for engineering interventions and the political risks inherent in them. (3) A sound theoretical understanding of implementation is imperative for practitioners of conceptual engineering. The usefulness of their representational (re-)designs depends on respecting broadly cognitive and social constraints on the part of the target population. I conclude that conceptual engineers have very good reasons to seriously engage with issues of implementation.

1 Conceptual engineering and the theory of implementation

Conceptual Engineering as the philosophical program of ‘assessing and improving our representational devices’ (Cappelen, 2018, p. 3; see also Isaac/Koch/Nefdt 2022, p. 1; Fischer, 2020, p. 2) is widely held to face the so-called ‘implementation challenge’: Conceptual engineers are pressed to explain how their (re-)designed representational devices can be implemented (Cappelen, 2018, pp. 72–78, ch. 7; Cappelen, 2020; Jorem, 2021; Koch, 2021; Nimtz, 2021; Pinder, 2021). Jennifer Nado has recently argued that implementation isn’t really part of conceptual engineering per se; Koch (2024) takes a similar line.¹ Nado’s view is straightforward:

I would argue, however, that the real ‘meat’ of the engineering process consists in the formulation, rather than the implementation, of suitable successors: in finding useful ways of carving up reality, and in constructing definitions to express those useful carvings. Persuading others to adopt those definitions, and thereby changing some relevant subset of their cognitive-behavioral dispositions, is in many cases the ultimate goal – but this latter step is really ‘advertising’ rather than engineering per se. (Nado, 2023, p. 145)

In defense of her stance, Nado argues that ‘the primary job of a conceptual engineer’ (ibid., p. 146) consists in (re-)designing useful representational tools apt to serve whichever function or role they aim to fit, whereas the process of ‘getting folks to use the tools we’ve designed’ is an advertising task rather than ‘an engineering process in the normal sense of the word’ (ibid., p. 146; italics in the original). Bringing speakers to adopt (re-)designed representational tools is anyway ‘a psychological and sociological project rather than a philosophical one’ (ibid., p. 146), or so Nado claims. In effect, then, Nado assures us that there is no need for a philosophically minded conceptual engineers to concern themselves with implementation. Doing so lies beyond their proper job description.

There is a sense in which Nado is perfectly right. Conceptual engineering can be pursued independently of the practice of implementation. Conceptual engineers do not have to push for representational reform themselves; no conceptual engineer needs to be a conceptual activist.² However, things look rather different for the theory of implementation. I understand the theory of implementation to inquire into issues concerning deliberate representational change, such as, notably, ‘By what interventions may we bring subjects to adopt (re-)designed representational devices?’. I believe that Nado is wrong about the theory of implementation. Conceptual engineers have very good reasons to theoretically engage with implementation, irrespective of whether they work on conceptual engineering (and are theorists of conceptual engineering) or engage in conceptual engineering themselves (and are practitioners of conceptual engineering). This holds true even if Nado is right and conceptual engineering basically comes down to representational (re-)design.

In this paper, I make a general case for the methodological importance of implementation to conceptual engineering. I argue that conceptual engineers as a class have very good reasons to engage with the theory of implementation – they should think long and hard about deliberate representational change. As I will explain, it follows that siding with Nado won’t allow engineers to sidestep issues of implementation. If I am right, these issues simply won’t go away.

To make my case, I argue three points. First: Inquiring into the methods by which we may actually get subjects to adopt (re-)designed representations is a theoretically challenging and philosophically worthwhile project in and of itself. I argue this point throughout this paper by discussing issues in, aspects of, and approaches to the general theory of implementation. In § 2, I start by outlining the Social Norms Account of implementation that I have devised in Nimtz (2021), which I employ as a foil. I emphasize that it does three things – it analyses implementation as an exercise in social engineering, it puts forth social norms à la Bicchieri as an effective tool for this task, and it outlines what I call an ‘implementation template‘ meant to practically inform and guide conceptual activists in their quest for implementations. My subsequent discussion will add to this already complex picture.

Second, a sound theoretical understanding of implementation is imperative for theorists of conceptual engineering. Such a theoretical understanding proves vital when it comes to assessing claims about the prospects for conceptual engineering, or the political perils inherent in it. In § 3, I argue that a theoretical understanding allows us to diagnose where Koslow (2022) goes wrong when she argues from linguistic mechanisms governing language change to the claim that the prospects for ‘ambitious conceptual engineering projects of the sort hopes most ride on’ (ibid., p. 23) are bleak. In § 4, I argue that such an understanding lets us see why Queloz and Bieber overshoot the mark when their assessments of the political risks we run by engineering interventions lets them diagnose significant ‘inherent tensions between the ambition to overcome the implementation challenge and the ideals of liberal democracy’ (Queloz/Bieber 2021, p. 2). From this I conclude that theorists of conceptual engineering as a class should engage with issues of implementation.

Third, a sound theoretical understanding of implementation is imperative for practitioners of conceptual engineering. In § 5, I argue that even if Nado is right and the practice of conceptual engineering comes down to (re-)designing ‘useful ways of carving up reality’ (Nado, 2023, p. 145), conceptual design decisions will be constrained by representational ergonomics and by characteristic implementability – any useful representational tools must be such that the targeted users can grasp and apply it and it must be such that actual interventions may move the target population to adopt it. From this I conclude that practitioners of conceptual engineers as a class should engage with issues of implementation. In § 6, I conclude and rehearse the methodological importance of the study of implementation to the overall project of conceptual engineering.

Let me emphasize that my argument concerns conceptual engineers as a class. Individual conceptual engineers are prima facie free to focus on whichever engineering aspect they care for, just as aeronautical engineers may spend their whole careers studying or designing flaps. But some conceptual engineers had better think long and hard about implementation, just as your airplane design team had better make sure that someone deals with the wings.

In what follows, I argue from methodological precedent rather than from methodological principles. In my Nimtz (2021), I develop an approach to implementation that is driven by ‘social norms’ in the technical sense of Bicchieri (2006, 2016). (I refer to social norms in Bicchieri’s sense as ‘social norms_CB’.) In this paper, I employ this Social Norms Account as an illustrative foil. I thereby avoid conducting the discussion in a speculative vacuum. I specifically employ the Social Norms Account because it aims to show that implementation is feasible by explaining how it can be done. This practical bend makes it particularly helpful for my purposes here; it also may explain why Isaac/Koch/Nefdt (2022) single out this approach for their discussion. Although I emphatically hold that the Social Norms Account is rather promising, I do see room for improvement and I will refine it where I see fit. This will bring the explanatory power of the Social Norms Account into sharper focus. Let me stress that my overall argument is not premised on the Social Norms Account. I also emphatically do not think that the Social Norms Account is the only game in town. There most likely is a plurality of viable ways to effect representational change. In fact, one of the key lessons of my argument will be that we need to consider different methods of implementation when we assess general views about conceptual engineering.

Before I begin, let me be clear on my general perspective. The picture I develop in Nimtz (2021) assumes two things: Engineers aim to change the linguistic meanings (as identified by a generic internalist meta-semantics) of words, and their efforts concern natural language. Here, I will adhere to the first idea. I take it as given that conceptual engineering primarily concerns linguistic expressions and their meanings. I will not adhere to the second idea, though. I now feel that our views need to be more inclusive; Nado (2023) rightly insists that conceptual engineers should sidestep debates about linguistic meaning or the nature of concept. I thus allow that conceptual engineers may target any community or population they care for, as long it has shared representations. I also allow that conceptual engineers may target any representational content associated with linguistic expressions that may be considered a meaning in a liberal sense of the term. In line with Koslow (2022) and Fischer (2020), I take this to imply that we are dealing with meanings in the sense in which ‘“myriad” and “girl” are used today with different meanings than they once were’ (Koslow, 2022, p. 5) and in which meanings guide how speakers ‘assess the correctness of categorization judgments, the accuracy of descriptions using the word, and the validity of inferences from utterances using the word’ (Fischer, 2020, p. 4). I take it that this liberal view is compatible with broadly internalist views in meta-semantics.³

This perspective on conceptual engineering covers Pinder’s (2021) idea that engineers should target speakers-meaning rather than natural language meaning. It also covers Nado’s own proposal according to which ’conceptual engineers should take themselves to be in the business of inventing (or perhaps better, discovering) classification procedures’ (Nado, 2023, p. 137). Nado introduces classification procedures as shareable abstract representations that encapsulate methods allowing users to sort things (ibid., p. 2, pp. 147–148). These procedures delineate categories – a determinate classification procedure picks out an ‘intension-like function from worlds to sets of entities’ (ibid., p. 148) – and speakers associate them with expressions or concept in order to decide what these apply to (ibid., p. 137). In short, then, classification procedures are representational contents that govern the application of associated expressions within a community. This means that they qualify as meanings in my liberal sense. Nado further thinks that classification procedures can be expressed by definitions (ibid., p. 149) and that representational (re-)design as a rule consists in stipulating a new definition for a linguistic expression ‘employed as definiendum’ (ibid., p. 149). This makes implementation an exercise in ‘[p]ersuading others to adopt those definitions, and thereby changing some relevant subset of their cognitive-behavioral dispositions’ (ibid., p. 145). For example, we may aim to change how subjects classify celestial objects by changing the definition they associate with their term ‘planet’. Nado thus sees conceptual engineering as primarily trying to change how meanings – or more specifically: classification procedures – hook up with linguistic expressions across a target population. This is precisely how the Social Norms Account construes things, as we shall presently see.⁴

2 The social norms account of implementation

Let us assume, as I will do throughout, that the conceptual engineer has successfully concluded the assessment-cum-improvement stage of their project to conceptually re-engineer a shared⁵ term ‘F’. That is to say, the engineer has (1) established that ‘F’ presently expresses M in the community C, (2) devised M* as a new meaning (the term taken in the liberal sense explained), (3) made a case for the normative claim that, all things considered, ‘F’ as used in C should express M* rather than M, where (4) this case is arguably strong enough to mandate a conceptual change: it licenses the engineer to bring it about that ‘F’ expresses M* rather than M in C. Having successfully navigated this exercise in theorizing, the conceptual engineer faces a practical challenge: How can the mandated change be realized? How can the term ‘F’ as used in the community C be made to express M* rather than M?

This is the implementation challenge. I construe this challenge as a request to come up with a demonstrably effective method of linguistic intervention (see Nimtz, 2021, p. 2). On a generic internalist meta-semantics⁶, bringing speakers to change their linguistic behaviour will reliably change what their shared expressions mean. So, if sufficiently many speakers can be made to adhere to the behavioral rule ‘Employ ‘F’ to exclusively express M*!’, this will bring their term ‘F’ to mean M* rather than M. On this perspective, the implementation challenge comes down to a problem in ‘social engineering’ (Nimtz, 2021, p. 8): How can conceptual engineers bring about a sufficiently broad change in linguistic behavior within the target population to yield the mandated change in meaning? If the community is small, well-connected, and overall sufficiently interested, the engineers may engage the members individually and try to change their linguistic behaviour by using the very reasons that mandate the change from M to M* in the first place. But what are the engineers to do if the community is fragmented, if overall interest in the matter is low, or if there are ‘just too many individuals one would have to affect’ (Jorem, 2021, p. 194)? In answer to these questions, the Social Norms Account enlists Cristina Bicchieri’s theory of social norms_CB (Bicchieri, 2006, 2016; Wietmarschen, 2021).

Social norms_CB in Bicchieri’s technical sense are behavioral rules underwritten by characteristic social expectations (Bicchieri, 2016, ch. 1; Bicchieri, 2006, ch. 1). Put simply, a social norm_CB is a behavioral rule that subjects comply with because others do so and think one ought to do so. Introducing the term ‘reference network’ to capture ‘the range of people whom we care about when making particular decisions’ (Bicchieri, 2016, p. 14), Bicchieri explains that R is a social norm_CB in a community C given that most subjects in C comply with R because they believe that (1) most people in their reference network comply with R (empirical expectation) and (2) most people in their reference network believe that one ought to comply with R (normative expectation) (Bicchieri, 2016, p. 35; Bicchieri, 2006, p. 11). For example, villagers in a development project may adhere to some inconvenient hygiene regiment because their communal leaders do so and insist that one ought to do so (Bicchieri, 2016, ch. 3, Bicchieri & Mercier 2014).

The fact that R is a social norm_CB thus comes with a characteristic stratification. There are those who adhere to R because they straightforwardly believe that one ought to do so. Their behaviour is guided by a normative conviction and thus is quite independent of what others think or do. In Bicchieri’s (2006, pp. 20–21, 2016, pp. 30–34) taxonomy, these subjects adhere to R as a moral norm. And there are those who adhere to R because, as they believe, this is what those they look to in this matter do and think one ought to do. Their behaviour is guided by social expectations and it very much depends upon those in their reference network. These subjects adhere to R as a social norm_CB. This stratification is important. What renders a social norm_CB communally efficacious is its power to project compliance with some rule R beyond the reach of the straightforward normative conviction that one ought to follow R. In fact, a community may adhere to a social norm_CBR even though only a minority shares this normative conviction.⁷ This will be so if most members of the community look to members of this minority as their reference network in the matter and comply with R because these members adhere to R and believe that one ought to do so.

Reference networks can be shared. Still, they are fundamentally personal – the reference network for a subject in the matter of R is the range of people that subject looks to when it comes to R. It follows that those reference networks may take different forms. If everyone in the community looks to the same few subjects, compliance with R may depend on a tightly interconnected small group of agents – think of the communal leaders mentioned above. (Call this an expert structure.) However, if most subjects look to one of their neighbours, compliance with R may depend on a vast, fragmented, and loosely knit web that is just held together by the fact that its members share the belief that one ought to follow R and live by it. (Call this a civic structure.) Note that ‘personal’ does not entail freedom of choice. S’s reference network may simply be dictated by socio-economic realities. Neither does it entail direct contact. If S looks to Q who looks to T, and so on, who looks to those in group G, then S’s behaviour will align with what those in G do and believe right.

That a subject S looks to a reference network when it comes to R does not yet explain why S should be motivated to do as these others do. This is especially puzzling since actual social norms_CB often require behavior that is beneficial to the community, but costly or inconvenient to the individual. Bicchieri (2006, pp. 23–24) identifies three factors that may motivate subjects to comply with a social norm_CB. Besides the fear of negative sanctions and a possible desire to please others, Bicchieri points out that recognizing someone else’s ‘normative expectation as well founded’ (Bicchieri, 2006, p. 23) can motivate a subject to comply: ‘If I recognize your expectations as reasonable, I have a reason to fulfil them’ (Bicchieri, 2006, pp. 23–24). Subjects may thus be motivated to act in a certain way because they accept that those in their reference networks hold their normative belief for what appear to be legitimate reasons, even though the subjects themselves do not straightforwardly believe that one ought to act in the way prescribed – say, they themselves have not (yet) sufficiently engaged with the issue in question to make up their minds.

I call this ‘oblique motivation’ (Nimtz, 2021, p. 12). Oblique motivation contrasts with the direct motivation exerted by a straightforward normative belief. The perceived legitimacy of a normative belief held by those in one’s reference network might for instance stem from their (perceived) normative authority in the matter, from their good if subjectively not quite persuasive arguments, or from a pluralist conviction that people take different normative stances together with the absence of compelling reasons against complying with the rule in question (see Nimtz, 2021, pp. 11–12). That said, acting on oblique rather than straight motivation may reveal less-than-ideal epistemic rationality on the part of the agents, notably so if they comply with R even though they, whilst sufficiently competent, have never even bothered to make up their own normative mind.

The Social Norms Account adapts Bicchieri’s general theory of social norms_CB to linguistic behaviour. Its key contention is that conceptual engineers ‘can implement specific conceptual changes by instituting social norms_[CB] adding to or revising the ones already shaping our linguistic behaviour’ (Nimtz, 2021, p. 14) in a way that renders social norms_CB an ‘effective, specific, and feasible means to effect conceptual changes mandated by conceptual assessments, both in principle and in practice’ (ibid., p. 19) – even when the engineers face a large or fragmented linguistic community where overall interest in the matter is low. The Social Norms Account moreover offers practical guidance. Bicchieri (2016) combines her theory of social norms_CB with empirical insights to devise practical procedures for diagnosing, creating, and changing social norms_CB. The procedures are meant to guide actors like the UN who strive to determine, say, ‘[h]ow to set up successful AIDS campaigns, convince mothers to nurse their children from birth, or induce people to build and use latrines’ (Bicchieri, 2016, p. xi; see also Bicchieri et al., 2014).

In the same vein, I sketch a schematic procedure (see Nimtz, 2021, p. 20), which I will here call the ‘implementation template_SN’, meant to practically inform and guide conceptual activists in their attempt to institute a revisionary linguistic rule R as a social norm_CB in a community.

Step 1

Find a group N such that (1) those in N accept R or lean towards R and (2) most speakers in the community either directly or indirectly look to (members of) N in the matter of R, or may plausibly be convinced to do so.

Step 2

Convince the subjects in N that one ought to follow R. Count on direct motivation to make them follow R themselves.

Step 3

Convince most speakers in the community to accept (members of) N as their reference network for R, either directly or indirectly, and inform them that those in their reference network follow R, and believe that one ought to do so.

Step 4

Convince most speakers that the members of their reference network hold the normative belief that one ought to follow R for good reasons and count on oblique motivation to bring the speakers to comply with R.⁸  

Let me stress three things. First, the template_SN exclusively relies on rational persuasion. Conceptual engineers are to convince those in N to believe that one ought to follow R, convince most speakers in the community to accept (members of) N as their reference network for R, and convince most speakers that those in their reference network hold this normative belief for sufficiently good reasons to provide oblique motivation. As I stress, this means ‘that instituting R as a social norm will be easier than instituting R as a moral norm to the extent that convincing speakers that someone else’s normative belief that one ought to follow R is legitimate is easier than convincing speakers that this normative belief is right’ (Nimtz, 2021, p. 19). Second, the template_SN provides an idealized picture. Even if the engineers exclusively proceed by reason and argument, the real-world effects of their intervention is likely to be at least in part driven by the likes of favorable biases, or worries about the social costs of non-conformity. Third, the template_SN provides contextually bound guidance-cum-inspiration rather than a general recipe. Bicchieri’s procedures require favorable social conditions to work. Likewise, the implementation template_SN will be helpful only if social conditions are structurally receptive to the rule R in question. Gibbons (2022) brings out very clearly that misaligned material, social, or moral incentives may make it instrumentally rational for sub-communities to resist a representational change, rendering direct attempts at implementation futile. Conceptual engineers may try to improve the odds by relying on social norm_CB, since this may change especially social incentives. Still, the incentives that make subjects resists direct representational changes may often prohibit instituting the corresponding social norms_CB. As I concede, ‘for expressions such as ‘liberal’ whose interpretations mark fault lines in politics, there simply won’t be a potential reference network (…) to exploit’ (Nimtz, 2021, p. 20).

The Social Norms Account provides a general and complex picture of implementation. First, it analyses implementation as an exercise in socially engineering a sufficiently broad change in linguistic behavior in order to yield a change in meanings. Second, it highlights that social engineers employ social tools in their intervention and advises conceptual engineers to rely on social norms_CB. Third, the Social Norms Account highlights that these insights may yield practical guidance for conceptual engineers and puts forth the implementation template_SN to this effect. As we shall see over the next three sections, using this general picture as our foil will bring out rather vividly why conceptual engineers should care a lot about issues of implementation.

3 Why theorists of conceptual engineering should care about implementation: Koslow on empirical linguistics and meaning change

Koslow (2022) argues that empirical insights into language change effectively sink conceptual engineers’ practical ambitions for revisions in meaning. Drawing on the analysis of implementation provided by the Social Norms Account, I explain why her argument is flawed. Rather than mimicking nature, conceptual engineers strive to socially engineer linguistic reality by an active intervention. So, any assessment of the prospects for implementing revised meanings had better take the social force – which the Social Norms Account will equate with the influence of social norms_CB – into account that conceptual engineers themselves exert in intervening. There is a methodological moral in this: Conceptual engineers should care about implementation because doing so brings out crucial factors that affect the general prospects of their engineering projects.

Many authors have recently argued that scientific results from linguistics, psychology, or the social sciences bear on conceptual engineering (Machery, 2021; Fischer, 2020; Nimtz, 2023). Koslow (2022) adds linguistics to this list. Arguing from ‘empirical pragmatics and diachronic semantics, including the study of linguistic innovation’ (Koslow, 2022, p. 2), Koslow proposes an empirically substantiated ‘method for (…) assessing the feasibility of proposed revisions’ (ibid., p. 2) by conceptual engineers. Her method pivots on ‘predictive generalizations’ (ibid., p. 3) about ‘how word usage evolves over time’ (ibid., p. 22) given the overall forces at work:

Questions about the feasibility of conceptual revision are like questions about whether a particular product will succeed in a new market, or whether a nonnative species will thrive in our ecosystem. We can study the likelihood of one expression becoming more popular than another in an area of discourse, rather like we can study the likelihood of one frog species outcompeting another in an area of the rainforest. While an expression’s usage may be inexhaustible, it is also shaped by various communicative pressures that exert themselves in foreseeable ways. (Koslow, 2022, p. 3)

Koslow argues that empirical insights into communicative pressures allow linguists to predict which word-meaning pairs are ‘likely to catch on in which areas of discourse’ (ibid., p. 4). Her view of communicative pressures is broadly Gricean. The fact that language serves conversational goals puts pressure on the selection of words and meanings, for how well the word-meaning pairs used realize some speakers’ conversational goals will depend on whether these make it comparatively easier or harder for the speakers to abide by conversational maxims such as ‘Be clear!’, or ‘Don’t raise irrelevant issues!’ (see ibid., p. 11). Koslow here highlights four generalizations. First, homonymy avoidance. Speakers tend to steer clear of homonyms. Second, enduring conversational desires. If a word becomes unsuitable even though it serves an enduring communicative purpose, speakers tend to use a different word to express the old meaning rather than to give it up. Third, loading. Speakers tend to avoid ‘words that are potentially offensive, uncomfortable, awkward, merely distracting, and especially words that are taboo (…) in favor of expressions that are less loaded’ (ibid., p. 14). Fourth, opacity. ‘Words that are unclear to audiences, such as technical terms and ones whose meanings are particularly difficult to recall, tend to be replaced by those that are clearer’ (ibid., p. 16).

Koslow takes these generalizations to provide pronounced obstacles to the idea of implementing (re-)designed meanings (ibid., pp. 12–18). She concludes that empirically well-established linguistic mechanisms effectively sink the ambitions of conceptual engineers:

Things do not look good for ambitious conceptual engineering projects of the sort hopes most ride on. They risk being like semantic perpetual motion machines, or blueprints for a bridge elegant in theory but unbuildable in practice. Correcting our vocabulary might have beneficial effects but is vastly harder than is usually presumed. (Koslow, 2022, p. 23)

I think that Koslow is right to emphasize the importance of empirical work in linguistics for conceptual engineering (see Nimtz, 2023). Still, I believe that her argument is flawed. The Social Norms Account analyses implementation as an exercise in socially engineering a collective change in linguistic behavior via an active intervention, recommending social norms_CB as the social tool of choice. Yet, once we see that implementing is intervening, it becomes apparent that we need to consider standard communicative pressures together with the social force exerted by the engineer’s intervention when we predict language dynamics in an implementation context. Since Koslow’s model ignores this social force, it misjudges the prospects for conceptual engineering.

To begin with, Koslow’s ‘market model’ of language dynamics relies on what ‘(…) is known about meaning change as it occurs in the wild’ (ibid., p. 2; my italics). Koslow takes this to be the proper methodological foundation: ‘There could be an applied science of how to increase a word’s share of use. It would be rooted in facts about how word usage evolves over time, identifying more fine-grained generalizations than I have here’ (ibid., p. 22, my italics). This misses that implementation is about socially engineering collective linguistic behaviour. Engineers neither aim to mimic naturally-occurring meaning change, nor do they intend to leave the success of their reform to standard linguistic market forces. Conceptual engineers rather envisage a proper engineering solution for the challenge they face – that is, a solution that combines empirical insight with creative design, technical innovation, and active intervention. For an analogy, consider flying. Humanity didn’t master flying by mimicking how birds do it. This proved an utter failure. The key to success was to learn about aerodynamics from nature and then to devise a decidedly non-natural engineering solution.

Once we appreciate that conceptual engineers conceive of implementation as intervention, it becomes apparent that it is not enough to aggregate the communicative pressures that shape language dynamics in the normal course of events. When we want to assess the prospects for deliberate linguistic change, we need to add the social force conceptual engineers themselves exert by their intervention. If the Social Norms Account is right, conceptual engineers will rely on social norms_CB. Do these create a social force of a new kind? When aeronautical engineers solved the problem of flying by using fixed wings and an internal combustion engine, they clearly relied on innovative elements and additional forces. I believe the analogous may often hold true for conceptual engineers. Pace Thomasson (2021), I cannot see that the norms governing which meanings we express with our terms are in general social norms_CB in Bicchieri’s technical sense. I don’t use ‘circle’ to express circle because others think one ought to do so, but because I know that this is right. That said, things are arguably different when I defer in my usage. I do employ ‘fish’ to express fish precisely because the experts in my reference network do so and think that one ought to do so.⁹ In any case, the important point to keep in mind is that the conceptual engineers’ interventions change the social forces that determine collective linguistic behaviour, either by bringing an additional social force to the mix, or by altering one of the social forces already in play.

We now see where Koslow’s model goes wrong. Her assessment of the prospect for changing meanings leaves out the social force that conceptual engineers exert by intervening. Let us assume that conceptual engineers rely on specific social norms_CB. Ignoring the social force these norms exert basically guarantees that we misjudge the prospects for the engineers’ reform. Consider collective behavioral changes more generally. Empirically well-supported generalizations about social behavior, addiction, and conveniences would have us predict that bringing people to stop smoking in pubs is a hopeless enterprise – until we see that we can ban this behavior by law and thus recruit the social force of legal norms to counterbalance common behavioral tendencies. By the same token, well-supported generalizations about language development, conversational goals, and communicative efficiency would have us predict that bringing speakers to change their linguistic behaviour in line with the engineers’ aims is a hopeless enterprise – until we see that conceptual engineers pursue active interventions that recruit the force of social norms_CB to counterbalance well-established common tendencies in linguistic behavior.

The Social Norms Account brings out that implementation poses a problem in social engineering which the conceptual engineer strives to solve by an engineering intervention that itself projects a social force. This invalidates Koslow’s model alongside with her bleak prediction. True, the Social Norms Account won’t make Koslow’s obstacles go away. But it puts them into the context of a social intervention where forces beyond those operating in standard communication are in play.¹⁰ In fact, the Social Norms Account allows conceptual engineers to be rather optimistic about implementation. Social norms_CB have proven efficacious at, for example, bringing people ‘to build and use latrines’ (Bicchieri, 2016, p. xi). As I have argued (Nimtz, 2021, pp. 18–21), since changes in linguistic behavior are contextually adjustable and easily reversible and since subjects are often free to choose their linguistic reference networks, social norms_CB should even work better for linguistic behavior.

Let me turn to the methodological moral. Whatever else my argument in this section has done, it has made vivid that a sound theoretical understanding of the methods by which we can actually get people to adopt (re-)designed representations will bring out factors and guiding ideas that manifestly affect the general prospects of engineering projects – in our cases, the effects of the engineer’s own intervention, and the interventionist outlook of engineering per se. I take it that this provides a good reason for conceptual engineers to care for the theory of implementation.

4 Why theorists of conceptual engineering should care about implementation: Queloz and Bieber on political risks of conceptual reform

Queloz and Bieber (2021) diagnose an ‘inherent tensions between the ambition to overcome the implementation challenge and the ideals of liberal democracy’ (Queloz/Bieber 2021, p. 2; see also pp. 1, 3, 13). Having explained how a society could effectively manage representational changes via a central institution powerful enough to ‘coordinate and enforce uptake’ (ibid., p. 2) thus regulating ‘which concepts people in fact come to use’ (ibid.), Queloz and Bieber argue forcefully that doing so would violate liberal democratic ideas. They conclude that politically speaking, it is a good thing that ‘conceptual engineering is hard to implement’ (ibid., p. 3). I explain why Queloz and Bieber’s argument does not quite do what they think it does. I grant that if implementation proceeds via institutionalizing top-down control, it is most likely politically untenable. Yet implementation need not proceed through institutionalizing top-down control. The implementation template_SN underwritten by the Social Norms Account outlines a rather different and arguably effective method that is, judged by Queloz and Bieber’s own rational persuasion standard, likely to be politically tenable. There is a methodological moral in this: Conceptual engineers should care about implementation because which political effects conceptual engineering will have varies with the method(s) of implementation we envisage. In fact, differences in political effects arise on the level of procedures. If we stick to the Social Norms Account but switch from the implementation template_SN to what I call the ‘dark template_SN’, we arguably lose liberal-democratic viability.

Queloz and Bieber see a way to solve the implementation challenge. They maintain that we as a society could decide to create a central institution tasked to control which terms with which meanings speakers in fact use. (This is how I will understand Queloz and Bieber’s talk of ‘concepts’¹¹.) We could moreover endow this institution with the power to police and enforce linguistic behavior, thus arriving at something like a ‘powerful centralized Ministry of Conceptual Engineering capable of exercising a great deal of control over which concepts people come to use in any given area of life’ (ibid., p. 13). Queloz and Bieber believe that thus institutionalizing control over uptake could ‘catalyze or even come close to guaranteeing the uptake and enforcement of conceptual innovations’ (ibid., p. 3). For support, they point to fields where we find similar mechanisms at work:

Authority over which concepts are to be used in a community of concept-users, such as a profession, can be centralized, and the processes of education, training, and continued admission and advancement in that community can be hierarchically organized and tightly regimented. We find such regimented, centralized, and hierarchical practices ensuring control over certain concepts in the medical profession, for instance, (…). (Queloz & Bieber, 2021, pp. 3–4)

Given that the implementation of (re-)designed representations could be all but guaranteed by institutionalizing control over uptake, should we make the political decision to create such an institution? Queloz and Bieber’s answer to this is an emphatic ‘No’. They argue that institutionalizing top-down control over uptake in a society carries intolerable inherent risks and is therefore incompatible with that society being a liberal democracy. More specifically, Queloz and Bieber see a liberal and a democratic rationale for ‘making and keeping conceptual uptake hard to control’ (ibid., p. 10).

For their liberal rationale, Queloz and Bieber embrace Shklar’s (1989) ‘liberalism of fear’ and tie the extent to which a society is liberal to how well it ‘avoids unchecked concentrations of power to protect the less powerful against abuses of power by the more powerful’ (ibid., p. 3). Institutionalizing control over uptake goes against this key liberal ideal ‘because any such power is likely to end up being appropriated by the already powerful or abused by the newly powerful’ (ibid., p. 10). Hence, ‘the power to effectively implement conceptual engineering projects, however ameliorative, across a society’s conceptual repertoire is incompatible with that society being a liberal society’ (ibid., p. 3).

For their democratic rationale, Queloz and Bieber argue that ‘democratic processes of governance and deliberation (…) cannot be genuinely democratic unless the power to determine in what terms people think is itself democratically distributed across society’ (ibid., p. 14). The faction in control of a controlling institution could ‘cloud and warp people’s understanding of their own interests, stint their sense of what is possible, and bias their electoral choices as well as their collective deliberation’ (ibid., p. 13). Queloz and Bieber see the risk that this faction could dictate ‘through which concepts the governed distinguish between legitimate and illegitimate exercises of power’ (ibid., p. 13) and thereby ‘effectively engineer the legitimacy of any exercise of power whatsoever’ (ibid., p. 14). They conclude that ‘[t]he institutionalization of that kind of power fatally undermines democracy on just about any conception of it, whether economic, epistemic, or deliberative’ (ibid., p. 14). ¹²

In sum, Queloz and Bieber argue that institutionalizing top-down control over linguistic behavior or conceptual choices is demonstrably efficacious, yet it would invite systematic abuse and thereby create intolerable risks to liberal democracy. Queloz and Bieber take this to be a structural problem. They think that we cannot reconcile institutionalizing top-down control with liberal democracy by adding democratic safeguards such as, say, independent oversight; at least, we cannot do so without jettisoning the efficacy. Let me grant all this for the sake of the argument. I still deny that Queloz and Bieber have established ‘inherent tensions between the ambition to overcome the implementation challenge and the ideals of liberal democracy’ (Queloz/Bieber 2021, p. 2, my italics), given that these are significant enough to make it a good thing that implementation is hard. The reason is simple. Queloz and Bieber do not consider alternatives to implementation through institutionalized top-down control. But there may well be an effective alternative to institutionalized top-down control that is arguably politically viable by liberal-democratic standards, or at most incurs tolerable risks to liberal democracy. Indeed, I believe that the implementation template_SN underwritten by the Social Norms Account is a strong candidate for precisely such an alternative.

The Social Norms Account proposes to change linguistic behavior by instituting the respective rule R as a social norm_CB in the target population and the implementation template_SN details a specific four-step-procedure for how we may go about doing this. I take it for granted that implementation template_SN is effective. To begin with, I have argued at some length before that the template outlines an effectual way to implement (re-)designed representations, both in principle and in practice (see Nimtz, 2021, pp. 12–21). What is more, the implementation template_SN does not rely on top-down control. It does not come with a top-down structure, and, as we have seen above, the respective reference network could well take the civic form of a large and loosely knit web of ordinary subjects, rather than a small cohesive group of experts. Finally, the implementation template_SN does not see uptake as driven by control and enforcement. It relies on conviction and oblique motivation instead.¹³  

So, does the implementation template_SN create intolerable risks to liberal democracy? I for one don’t think so. Queloz and Bieber appear to grant that any implementation procedure which thoroughly relies on the ‘rational power of reason and argument’ (Queloz/Bieber 2021, p. 15) has a good claim to be politically viable. This is good news for the implementation template_SN, which is designed to work by rational persuasion. Engineers following the template are required to convince the subjects they deal with in order to effectively institute their revisionist rule R as a social norm_CB. They do not need to convince everyone of the same thing, though. Having identified a potential reference network N, the engineers need to convince those in N that one ought to follow R. They then need to convince most speakers to accept (members of) N as their reference network for R, before they turn to convincing these speakers that those in their reference network hold the normative belief that one ought to follow R for good reasons and act on it. One may feel uneasy about the second step. Here the engineers set out to convince speakers to accept (members of) N as their reference network rather than to convince them that one ought to follow R. This may appear less than ideal. But I cannot see that it undermines liberal democracy or violates liberal-cum-democratic standards, given that these do not require ideal political conditions, as Queloz and Bieber (2021, pp. 15–16) are happy to concede. In fact, the risks we incur by employing the template seem very much tolerable given that the mode of propagation employed still is that of convincing. Our engineers rely on reason and argument to make people accept N as their reference network, they do not revert to sanctions or social coercion. Kitsik (2023) fails to see this when she cites the Social Norms Account as a programmatic instance of ‘paternalistic conceptual engineering’ (ibid., p. 18). Her own standards should have told her otherwise: ‘I contrast paternalistic cognitive engineering with explicit rational persuasion, informing, and educating’ (ibid., p. 8; my italics).

Although my discussion has been far from exhaustive, I take it to show that there is a straightforward case to be made that the implementation template_SNqua method comes out politically viable on Queloz and Bieber’s own standards; at the very least, it poses a far less severe risk to liberal democracy than institutionalized top-down control. This comparative upshot is all I need. Queloz and Bieber show that implementation through institutionalized top-down control creates intolerable risks for liberal democracy. As we have seen, this is in all plausibility not true for implementation via the implementation template_SN. This provides ample support for a general claim I deem highly plausible anyway: The type and severity of the political risks we incur by an intervention will systematically vary with the method we employ in intervening. But if that is so, then you cannot demonstrate an inherent tension between implementation and liberal democracy per se by looking at just one method of implementation. You need to consider all pertinent alternatives; at the very least, you need to identify democracy-undermining traits of your method that manifestly generalize to all such alternatives. There is a general methodological moral in this. Whatever else my argument has done, it has made vivid that a sound theoretical understanding of the methods by which we can actually get people to adopt (re-)designed representations will highlight factors that affect the type and severity of the political risks we incur in implementation. This provides a good reason for conceptual engineers to care for the theory of implementation.

Let me stress that my argument concerns methods of implementation. There are other factors that affect political viability. If you engage in an implementation project and commit some ‘metalinguistic injustice’ (Catapang Podosky, 2022) or rely on a ‘conceptual domination’ (Shields, 2021), your project most likely thereby degenerates into a ‘meaning perversion’ (Marques 2020, p. 263), which is Marquez term for ‘revisionary projects that are politically or morally illegitimate’ (ibid.). I fully grant that. My point is that the method you use is an important determinant of the risk you run from the very outset and that different methods per se come with different types and levels of risk. Löhr (2022) argues that the moral legitimacy of an engineering intervention cannot be read off from the change in representation it aims for; our moral calculus also needs to heed the social and conceptual disruption caused by the intervention itself.¹⁴ I agree. Still, since socially disruptive effects will vary with the chosen method of implementation, any ethics of conceptual disruption needs to pay close attention to such methods and the differences between them, or so I insist.

I said that the pertinent political risks depend on the method of implementation we employ. In actual fact, the risks vary with the specific procedure, rather than the general method. The evocative foil for Queloz and Bieber’s case is the fiction of a ‘powerful centralized Ministry of Conceptual Engineering’ (ibid., p. 13). I take it that much of their argument develops the political perils inherent in any centralized institution poised for top-down control over people’s more private choices, regardless of whether these govern, say, dietary habits, sexual practices, or linguistic uses. The remedy appears obvious: switch to a method of implementation that relies on bottom-up influences of behavior, say, via social norms_CB. However, switching methods is not good enough; we need to employ the right procedure. Otherwise we might just trade perils for equally bad ones.

As Bicchieri is the first to admit, social norms_CB are not intrinsically tied to rational persuasion. Quite the opposite. You may look to a group as your reference network because social pressure forces you to and you may be motivated to comply with the behavioral rules these people deem obligatory because you will be socially punished if you don’t (Bicchieri, 2016, pp. 188–194). Linguistic rules are no exception to this. Suppose that although most of your peers disagree, you have convinced yourself that speakers ought to follow the linguistic rule R. Here is a procedure that may guide your attempt to establish R as a social norm_CB. I call it the ‘dark template_SN’:

Step 1*

Find a group N of likeminded speakers who agree (1) that one should adhere to R and (2) that since they themselves are in a position of normative authority in this matter, everyone should look to (members of) N as their reference network for R.

Step 2*

Count on direct motivation to make the members of N actually follow R themselves. Create a consensus in N that not following R amounts to an intolerable normative transgression.

Step 3*

Portray N to the wider public as the proper normative authority with respect to R. See to it that the public knows that those in N follow R, believe that one ought to do so, and deem not following R an intolerable normative transgression.

Step 4*

Establish a credible threat known to most speakers that they may be severely socially sanctioned – that is, be subjected to campaigns of social degradation across all available channels – if they fail to comply with R, or do not look to those in N in this matter.

The dark template_SN is just as bottom-up and potentially grassroot-driven as the implementation template_SN. Still, it differs from the original implementation template_SN in three key respects: It does not require that its reformist agenda is underwritten by a clear mandate for implementing R, it may invoke self-proclaimed rather than discursively underwritten normative authority, and it relies on social coercion rather than on reason and argument.

The dark template_SN is arguably effective. In fact, it may often be more effective than the implementation template_SN. Yet, unlike the implementation template_SN, the dark template_SN arguably violates liberal and democratic standards. This is not to say that any instance of the dark template_SN is by necessity politically destructive. It may incidentally guide a truly ameliorative campaign, especially if there happen to be compelling reasons which de facto mandate the respective conceptual change. To say so is rather to insist that the dark template_SNqua procedure does not meet plausible criteria for political tenability. This holds true for the standard Queloz and Bieber invoke. Allowing the dark template_SN into our social practice inherently invites systematic abuse including coercive conceptual control. This also holds true for Kitsik’s standard. Subjects made to comply with a rule R by social coercion can indeed reasonably complain that ‘their sovereignty over their belief-formation’ (Kitsik, 2023, p. 621) has been violated. It thus is the dark template_SN in particular, rather than the Social Norms Account in general, that faces the charge of paternalistic conceptual engineering.

5 Why practitioners of conceptual engineering should care about implementation: Nado on conceptual engineering and representational design

Why should conceptual engineers care about implementation? I have argued that they should do so because a sound theoretical understanding of implementation methods proves vital when it comes to gauging the prospects for engineering interventions and it allows us to identify – and thus potentially avoid – the specific political risks we run by intervening in specific ways. These reasons concern the theory of conceptual engineering. In effect, I have argued that theorists of conceptual engineering as a class had better think long and hard about implementation. I take this to be an important result. Philosophers in the field mostly theorize about rather than practice conceptual engineering; actual attempts at deliberate representational amelioration are few and far between.

These reasons are unlikely to impress Nado. Nado appears to be primarily concerned with the practice of conceptual engineering. As I read her, she argues that conceptual engineering as a practice does not involve engaging with implementation; any conceptual engineer who does goes beyond their proper job description. I have granted from the outset that practitioners of conceptual engineering need not be conceptual activists. Still, I submit that they have very good reasons to theoretically engage with issues of implementation. Useful representational tools are constrained by representational ergonomics and by characteristic implementability – they must be such that normal subjects can grasp and apply them and they must be such that actual interventions may get the target population to adopt them. This directly impacts the design decisions of conceptual engineers. There is a methodological moral in this: Practitioners of conceptual engineers as a class should engage with issues of implementation because representational design decisions are constrained by ways and means to make subjects adopt the representations thus (re-)designed.

Nado (2023) maintains that the proper business of conceptual engineers – here understood as philosophers who engage in the practice of conceptual engineering – consists in devising or ‘finding useful ways of carving up reality, and in constructing definitions to express those useful carvings’ (ibid., p. 10). In fact, representational (re-)design thus understood very much demarcates the proper practice conceptual engineers engage in, or so Nado has it. She goes on:

If this general approach is right, then the majority of the actual cognitive work that goes into conceptual engineering will be at the level of designing the tools – that is, in inventing or discovering ways of classifying, and in thinking through how suitable those classifications are for various purposes, functions, and roles we might recruit them for. (Nado, 2023, p. 10)

Nado maintains that whilst representational (re-)design is at the heart of the practice of conceptual engineering, implementation is but an optional afterthought. Making speakers adopt (re-)designed representational tools is a matter of persuasion, rather than of reasoning, and so comes down to ‘’advertising’ rather than engineering per se (…)’ (Nado, 2023, p. 145) or ‘in the normal sense of the word’ (ibid., p. 146). Given that people are not always swayed by good reasons, this task anyway is ‘a psychological and sociological project rather than a philosophical one’ (ibid., p. 146).

I have granted from the outset that practitioners of conceptual engineering need not engage in the practice of ‘conceptual activism’ (Cappelen, 2018, p. 199). Turning to the theory of implementation, Nado’s first reason for thinking that philosopher-engineers may properly ignore issues of implementation appears to be this: They may do so because implementation is a second-rate, non-engineering, and non-philosophical subject, at least when compared to representational (re-)design – it’s ‘really advertising’ (Nado, 2023, p. 145). I believe that the fuller picture we have arrived at makes it abundantly clear that this misappraises the field. The theory of implementation marks a highly complex discipline bearing all the hallmarks of the ‘applied science’ Koslow (2022, p. 22) alludes to. It inquires into all factors that affect speakers’ representational choices, be they cognitive abilities, motivational structures, or social pressures, it aims to design social tools to manage linguistic behavior, and it draws on these insights and tools to provide practical guidance to those actually campaigning for representational reform. I can see no sensible measure on which this discipline comes out theoretically less demanding than representational (re-)design. I also can see no reason why it should be any less of an engineering subject. Engineering is characterized as ‘the study of using scientific principles to design and built machines (…) and other things’ (Cambridge Dictionary), where these may include ‘structures, or other complicated systems and processes’ (OED).¹⁵ By these criteria, inquiring into social tools apt to effect representational change qualifies as engineering in the normal sense of the word – as social engineering if you insist, but as engineering nonetheless.

But isn’t Nado right to insist that implementation falls to sociology or psychology rather than to philosophy? Bicchieri provides a powerful philosophical theory of the nature of norms that highlights the pivotal role of social norms_CB. That Bicchieri’s theory can provide practical guidance for actual social amelioration speaks to its explanatory power; it does not indicate that we have left philosophy. There is no denying that implementation raises intriguing empirical issues in sociology, psychology, or linguistics. But as we shall see below, the same holds true for any representational (re-)design that aims to be eventually useful.

This brings us to Nado’s second and central claim about how implementation relates to the practice of conceptual engineering. I understand her to argue that philosopher-engineers can leave issues of implementation to one side because they are primarily in the business of ‘finding useful ways of carving up reality, and in constructing definitions to express those useful carvings’ (ibid., p. 10). This is what effectively demarcates their practice; at heart, they are philosopher-designers. This yields a straightforward standard: Engaging with some intellectual field or other is part of conceptual engineering as a practice just in case philosopher-designers need to do so, or have very good reasons to do so, given that they aim to devise useful ways of carving up reality. I will grant this standard. For I think that philosopher-designers need to engage with issues of implementation, too.

Engineering processes are typically taken to be segmented into stages – they lead from an assessment stage via a design stage to an implementation stage (see Isaac/Koch/Nefdt 2022, § 2). The activities of designing and implementing are independent of one another in the sense that engaging in one does not require engaging in the others. There can be a division of labor between designers and implementers – for example, you design a new definition for ‘oppression’ and I then intervene to have it adopted within the target population. However, this does not mean that the stages of the process are utterly independent of one another. They are not. A philosopher-designer exclusively contributes to the design stage. Still, their work is constrained by factors pertaining to the implementation stage and studied by the theory of implementation. For in order to make informed design decisions, the philosopher-designer needs to know about factors that systematically shape representational trajectories in the target population, notably cognitive abilities, motivational structures, and social pressures, just as the material engineer ‘must be able to determine the forces and thermal environment that a product, its parts, or its subsystems will encounter’ (Columbia University).

I will highlight two sets of constraints. First, (re-)designed representations had better have the right representational ergonomics – they had better be such that they can be grasped, processed, and applied by the target subjects. I understand representational ergonomics to mostly concern cognitive capacity. In order to accommodate representational ergonomics, designers need to adapt their products to what the individuals concerned can do. Suppose, for example, that the individuals in question cannot reliably process highly disjunctive representations. This militates against designing highly disjunctive definitions. Second, (re-)designed representational tools had better have the right characteristic implementability. That is to say, they had better be such that actual interventions can effectively get them adopted within the target population. I understand characteristic implementability to mostly concern motivation. In order to accommodate characteristic implementability, designers need to adapt their products to what subjects individually, and the target population collectively, may be moved to accept. Suppose that the target population won’t be moved to accept definitions that involve technical vocabulary from, say, economics. This militates against using such terms in the definitions proposes by the designer. Given that representational design is constrained by representational ergonomics, representational (re-)design is deeply intertwined with empirical fields such as cognitive psychology. Given that it is constrained by characteristic implementability, it is deeply intertwined with empirical fields such as sociolinguistics.

Let me emphasize the general point: Representational (re-)design is constrained by representational ergonomics and characteristic implementability because – or more precisely: to the extent that – it aims to be eventually practical. This is not a gratuitous assumption that Nado can drop. Nado stresses that at the heart of conceptual engineering lies the aim of ‘inventing or discovering ways of classifying, and in thinking through how suitable those classifications are for various purposes, functions, and roles we might recruit them for’ (ibid., p. 11) in order to provide ‘useful ways of carving up reality’ (ibid., p. 10, my italics). But if that is so, then the designer needs to make informed design decisions that heed representational ergonomics and characteristic implementability. Representational devices that the target individuals cannot reliably grasp, process, or apply, simply aren’t useful. By the same token, representational devices that the target population just cannot be made to adopt aren’t useful, either. Per se, however, both these dimensions concern the implementation stage rather than the design stage of the engineering process.

The usefulness of a (re-)designed representational device depends on who is intended to do what with it. That is, it depends on its intended role or function and on the target population the device is designed for. Philosopher-designers aiming at useful designs for a target population hence have very good positive reasons to be well-informed about representational ergonomics and characteristic implementability with respect to that population, for this will allow them to make better design decisions. The specifics will vary with the specifics of their project and notably with the target population. It matters a lot if their definitions target, say, all speakers of a natural language or a small group of like-minded economists; as Cull (2021) brings out, it is also rather important whether the designers themselves belong to the target population. They may also want to make trade-offs between representational ergonomics and characteristic implementability on the one hand and representational accuracy on the other. Neither undercuts my epistemic point, for in order to see the variations or to explore potential trade-offs, the philosopher-designers need to be well-informed. There is a methodological moral to be drawn: Whatever else my argument has done, it has made vivid that practitioners of conceptual engineers as a class should engage with issues of implementation because representational design decisions are constrained by issues pertaining to implementation, notably by the demands of representational ergonomics and characteristic implementability.

Let me add a thought. Can’t philosopher-designers sidestep all issues of implementation by insisting that their designs target rational thinkers in general? I fail to see this. Suppose ‘targeting rational thinkers in general’ means that their designs are meant to be useful to any rational thinker. This will not allow the philosopher-designer to sidestep issues of implementation. You and I most plausibly are rational thinkers. Yet there are limits to our memories and to what we can be made to accept and any (re-)designed representation that is meant to be useful to us will need to respect these. Suppose ‘targeting rational thinkers in general’ means that philosopher-designers exclusively derive their criteria for usefulness from the concept of a rational thinker. This would allow them to abstract away from contingent constraints such as limits of memory or social malleability. However, since these constraints are very much operative for you and me, the resulting representational designs will most likely prove useless for rational thinkers like us.

6 Conclusion

In this paper, I have made a general case for the methodological importance of implementation to conceptual engineering. I have argued that theorists and practitioners of conceptual engineering alike have very good reasons to engage with the theory of implementation – as a class, they should think long and hard about deliberate representational change. I have used my Social Norms Account as a foil in order to defend three general claims about conceptual engineering and implementation.

First, inquiring into the methods by which we may bring people to adopt (re-)designed representations is a theoretically challenging and philosophically worthwhile project in and of itself. I take it that my argument over the paper as a whole has made it abundantly clear that the study of implementation makes for an exciting theoretical discipline and an engineering enterprise in the full sense of the term at that. I guess that my argument has also brought out rather nicely that the Social Norms Account marks an analytically and explanatorily powerful proposal within the field of theorizing about implementation.

Second, a sound theoretical understanding of implementation is imperative for theorists of conceptual engineering. Inquiring into deliberate representational change highlights factors that affect engineering success, notably the social force that the engineers themselves exert by intervening, and shows how the effects of an engineering intervention depend on the method employed. Taking implementation seriously thus allows us to better assess claims about conceptual engineering such as Koslow’s (2022) claim that empirical linguistics proves conceptual reform a lost cause, or Queloz and Bieber’s (2021) contention that conceptual engineering per se carries intolerable risks for liberal democracy.

Third, a sound theoretical understanding of implementation is also imperative for the practitioner of conceptual engineering. Inquiring into deliberate representational change brings out how representational design decisions are constrained by issues pertaining to implementation, notably those I have summarized under the headings of ‘representational ergonomics’ and of ‘characteristic implementability’. This holds true even if Nado (2023) is right and conceptual engineering basically comes down to representational (re-)design aimed at devising useful ways to carve up reality.

If I am right, advocates of conceptual engineering cannot sidestep issues of implementation by siding with Nado and equating conceptual engineering with representational (re-)design. The question of how we can actually bring people to adopt (re-)designed representational devices just won’t go away that easily. Let me add two thoughts. I feel that implementation is typically viewed as a problem for conceptual engineering. I don’t share this view. I rather think of implementation as a promising yet underappreciated source of theoretical inspiration and methodological insight for conceptual engineering. At the same time, I suspect that conceptual engineers tend to tacitly assume that implementation is essentially trivial – we know how to convince an individual to adopt a (re-)designed representation, and scaling-up to the level of the population is easy, since we simply go on the same way. I don’t share this view either. Even if we knew the former, which I believe we don’t, I think that bringing a whole community to adopt a (re-)designed representation comes with complexities and challenges of its very own.