6.1 Introduction

Other than its particular social nature, a key aspect of human life is the fact that it is characterized by its heavy reliance on tools and artifacts. We humans are not just dependent on others to make our way through the world, but we are also building technology to aid us in this. As it turns out—and as will be made clearer later in this chapter—there are some important connections between the social nature of human living and its technological underpinnings. However, even intrinsically, assigning functions to artifacts is very important.

First, functional ascriptions can be used to describe and identify the artifact. A hammer is defined by what it is for: to drive nails into objects (say). Lots of things can be used, more or less well, as hammers (sticks, screwdrivers, flutes, keyboards), but that does not mean that they are hammers. Only objects that have the function to drive nails into objects are actually hammers.

Second and most importantly, functional ascriptions to artifacts play important social scientific roles. On the one hand, they can explain why certain artifacts are stable parts of society, and others not. The function of music tapes was to record and play music, not to be book weights. This explains why they went out of existence when better ways of distributing music became available, and were not affected by the existence of better book weights. On the other hand, artifact functions can also be used to justify interventions. If X is for F, then its failing to do F justifies fixing it so that it does F. A hammer that fails to drive nails into objects needs fixing. A lightbulb that has the same “problem” does not need fixing.

However, it is not yet fully clear how to justify a particular functional ascription to an artifact. As this chapter (together with the previous ones) makes clear, the two main ways to do so—the intentionalist and the historical selectionist accounts—have problems that mean these accounts are not fully compelling as they stand. This is particularly important when it comes to the intentionalist account, which is central in the case of artifacts, and is thus especially well illuminated here.

To alleviate this situation, the chapter applies presentist social functionalism to this case. This not only provides a more compelling way to assign functions to artifacts, but, by making the connections between artifacts and social institutions clearer, it brings out novel aspects of the discussion surrounding technology. It also again clearly displays the fruitfulness of the presentist social functionalist perspective.

The chapter is structured as follows. In sect. II, we present the two major existing accounts of artifact functions and bring out their problems. In sect. III, we develop the presentist social functionalist alternative and note its benefits. In sect. IV, we apply this presentist social functionalist treatment of artifacts to some contemporary cases: ChatGPT (and other large language models) and off-label uses of pharmaceuticals. Section V concludes.

6.2 Existing Accounts and their Problems

Artifacts bear the mark of intentional design, and much of the study of artifacts has emphasized this fact—elaborating, for example, the ways that intentions are communicated to users (Dipert 1993, 1995; see also Norman 2013) and the role that intentions may play in individuating artifact kinds (Evnine 2016). It is often taken as self-evident that talk of the functions of technology must refer to the intentions behind its design: “No agent, no purpose, no function” (McLaughlin 2000, p. 60). Intentions therefore form the basis of the prevailing theory of artifact functions: the intentionalist theory (Vermaas and Houkes 2003).

Intentionalist theories (see e.g. Dipert 1993, 1995; Hilpinen 1993; McLaughlin 2000; Houkes and Vermaas 2004, 2010; Vermaas and Houkes 2003) take intentions of designers or users to be necessary to justify proper function ascriptions to artifacts. In this sense, intentionalist accounts are often described as “agent-oriented” rather than “property-oriented” (Van Eck and Weber 2014, p. 1371).Footnote 1

One might also take a designer’s intention to be sufficient for an artifact to have a concomitant function. Most views, however, put further conditions on justified function ascriptions, which have to do with the ways that artifacts are epistemically situated. Houkes and Vermaas (2010), for example, suggest a support requirement and a communicative requirement. Designers are assumed to understand the artifacts they create more deeply than most users, and to communicate uses to users, facilitating a social division of knowledge. For Houkes and Vermaas (2010), in order for a function ascription to be justified, a designer must have a support belief, according to which an artifact has the physical properties necessary to be at least minimally capable of performing the function that they assign to it. They can thereby justify a use plan, a set of instructions to produce the functional effects with the artifact.

The communicative requirement also contributes to this social division of labor. If the intentions of designers occupy such a privileged position, then users must have some way of knowing them. Functions can be communicated by word of mouth, advertising, instruction manuals, and other media. They can even be signaled by the artifact itself—for example, an ergonomic grip on a hammer can signal how it is meant to be held.Footnote 2 However this communication takes place, it is often taken to be a core aspect of artifact functions, since it maintains the epistemic (understanding-focused) structure described above. Indeed, Dipert (1995) explicitly requires that artifacts intentionally signal their functions. Houkes and Vermaas (2010) likewise require that functions originate from the aforementioned “justifiers”; if one is not a justifier herself then she must rely on communications from other users.

In some cases, this communication will be superfluous, namely when one makes an artifact for one’s own use. Suppose I create a cake that looks like a plant pot. This artifact would not signal its function to users—in fact, it may misleadingly signal functions that it is not capable of performing. Yet it still seems capable of performing its function—that of a cake—when I eat it. For Houkes and Vermaas (2010), communication will not be necessary because I occupy the epistemic position of a justifier, at least at the time of baking. Nevertheless, even here there may be an epistemic asymmetry between designer and user. At the time of baking, I must know a recipe and various practices associated with baking. I need not know any of these things at the time of consumption. I am “permitted” to occupy a different epistemic position as a user than that which I must occupy as a maker. If communication is unnecessary here, it is because of my epistemic position as the maker. In general, however, communication is critical for artifact use, since it allows agents to use artifacts according to their purposes without fully understanding how they achieve those purposes.

Similarly, users often use artifacts in nonstandard ways, which may vary in their effectiveness. A novel use does not count as a proper function unless a user has sufficient knowledge about an artifact to rationally justify her new use plan in terms of the artifact’s physical properties. Houkes and Vermaas (2010) call such users justifiers and suggest that they participate in the artifact’s design since they occupy an epistemic position analogous to that of designers and from this position “design” uses for the artifact. Artifacts are characterized by the social division of epistemic labor that is thus achieved (Houkes 2006), whereby “passive users” (Houkes and Vermaas 2010) can make use of artifacts without understanding the mechanisms by which artifacts work. If this is meant to expand to other tokens of the artifact, though, these novel use intentions need to be communicated to others—hence the need for the communication requirement.

In short: on the predominant theory of artifact functions—the intentionalist one—an artifact type A has function F if the designers of the artifact (which may include its users) (i) intend to A to do F, (ii) believe A can do F, and (iii) communicate the intention for A to do F to users. While this theory has many compelling features, it also faces two major challenges.

First, designers’ intentions are often indeterminate. People can and do simply intend to build or fashion something that seems useful for a lot of different things—some of which they may not even be aware of yet. I can pick up a stick and sharpen one end, simply because I think it might come in handy for a variety of uses. I don’t intend it for any specific use, but I think it might turn out to be useful as a spear, or a hammer, or a walking stick. The same is true for contemporary artifacts like smartwatches, which also were designed with a variety of causal powers with little inherent communalities—tracking health and fitness data (heart rates, etc.), relaying messages, playing music, telling the time. No specific use-cases need to have been in the minds of the designers when they designed these watches; their intention may just have been to build an electronic device that can be worn on the wrist and that has a variety of causal powers.

If so, though, then the intentionalist-account would have to say that these artifacts do not have a determinate function. This is not obviously the right answer, though: for example, it may turn out that most people end up using smartwatches as fitness trackers. As noted in the discussion of the function of corporations, this might then make it social scientifically reasonable to see fitness tracking as the function of these watches.Footnote 3 We return to this issue in more detail in the next section, but for now, the key point to note is just that the fact that a designer’s intentions are often unclear makes them a weak basis on which to build an account of artifact functions.

The second problem with the intentionalist account of artifact functions is the fact that patterns of use at least sometimes seem to be able to override intentions (see also Preston 1998, 2009, 2013; Scheele 2006). This is again related to some of the arguments made in the chaps. 4 and 5, and will be made clearer below. However, for now, it is sufficient to note that a maker may design an artifact A to do F, but people mostly or always use it do G. This could be because people’s values or intentions diverge from those of the designers, or because they have discovered uses of the artifact not recognized by the designers.Footnote 4 In that case, though, it seems that the role that functional ascription here plays, social scientifically, would seem to be focus on G. For example, frisbees were designed as pie containers, but came to be mostly used as toys, WD-40 was designed as a rust-preventer, but it is mostly used as a mechanical lubricant. Saying that frisbees are being mis-used as toys or WD-40 is being mis-used as a lubricant would run counter to these established practices and is unlikely to be socially scientifically compelling.

In the background of this objection is the need for an account of what it is to design or create an artifact. Is picking up a stick creating a hammer? If so, what if I start using the stick as a fishing spear? Is that equivalent to creating a new artifact? Houkes and Vermaas (2010) have an account of function change that occurs through use. For them, functional ascriptions must be traceable back to designers or “justifiers”: creative users who understand the properties of artifacts well enough to support novel uses. For Houkes and Vermaas, functions change when justifiers participate in the “design” of a new kind of artifact; for example, those who used Frisbie pie containers as flying discs thereby designed a toy that used the pie container as material. Once these designers could support their use plan with an understanding of the artifacts and communicated this use plan to others, the function changed. Below, we will argue that our account provides a better explanation of function change than this, at least for our purposes. For now, it suffices to say that the emphasis on the intentions of designers, or indeed on those of any privileged epistemic agents, sits oddly with the flexibility of artifact use.

Indeed, this is related to the point that was made in the previous chapters, and which will also become important again below: a key role that functional ascription plays in the social sciences is to help us determine which parts of science are stable, long-terms features of that society, and which merely transient phenomena. In this context, if an artifact is a stable feature of society—as underwritten by its pattern of use centering on G—then focusing on the designer’s intentions for A to do F would miss this. Again, this will be made clearer in the next section below, but for now, the key idea is just that the intentions of a designer, even where they are determinate, do not clearly seem to be a compelling basis for a functional ascription in cases where patterns of use diverge from these intentions.

Consider next the other major account of the artifact functions: the historical, reproduction-based account. The structural-functionalist and virtual selectionist accounts are not central in this discussion, and since they were discussed at length already, will not be re-considered here. In fact, even when it comes to the historical account, the discussion here can be quite brief, as several of the main points have already been made in the previous chapters. Still, a few remarks specifically in the context of functional ascription to artifacts are important to make.

Numerous authors have compared the development of technologies to evolution by natural selection (Ziman 2000). Biological functions are valuable because they are explanatorily deep: they give insightful explanations for the existence of characters to which they are ascribed (Garson 2019; see also Wright 1976). Similar explanations are possible for many cultural entities, of which artifacts are just one type (Boyd and Richerson 1985). A theory of artifact functions that takes this approach can be called a reproduction theory. The most developed reproduction-based view is given by Preston (1998, 2009, 2013), who draws primarily from Millikan’s (1984) theory of proper functions in biology. Preston (2013) gives the following definition of artifact functions:

A current token of an item of material culture has the proper function of producing an effect of a given type just in case producing this effect (whether it actually does so or not) contributes to the best explanation of the patterns of use to which past tokens of this type of item have been put, and which in turn have contributed to the reproduction of such items. (Preston 2013, p. 187)

As in the biological case, functions are meant to explain differential reproduction resulting in currently existing tokens. In particular, Preston appeals to social processes to explain patterns of use, and these same patterns of use can help to explain the intentions. A watch found on a beach, for example, indicates not just a designer who created it to satisfy a specific intention, but almost certainly an entire cultural history that itself explains why a designer would intend to keep time in the first place (Preston 2013, pp. 157–158). Designers’ intentions are just one part of a larger explanatory picture. Even when they play a prominent role in the explanations of the patterns of use of artifacts, however, artifacts do not have their functions in virtue of being designed with specific intentions, but in virtue of being explicable by a kind of selection.

Incorporating non-intentional factors into the functions of technologies may also be fruitful because non-intentional aspects of technologies have arguably been those of most interest to philosophers of technology for decades. Winner (1980), in his seminal essay on the politics of artifacts, compares technologies, in their power to restrict human freedom and regulate behavior, to “legislative acts or political foundings that establish a framework for public order that will endure over many generations” (p. 128). He uses nuclear power to illustrate this restrictive aspect of technology (see also Mumford 1964). Nuclear power, Winner (1980) argues, is predicated on institutions that make it safe to use; once the technology is in use, dissolving the institutions that oversee production and handle radioactive byproducts becomes a practical impossibility. Regardless of whether this assessment of nuclear power is correct, the point is that a critical part of what technologies do to individuals and societies is independent of what designers and users intend to do with them. Technologies are more than instruments for our use. A non-intentionalist account like Preston’s opens up the possibility of incorporating these political footprints into the functions of technologies.

However, there are also several problems for this way of ascribing functions to artifacts. On the one hand, as before, many technologies seem to lack the necessary history to make this kind of functional ascription possible. For example, it may be social scientifically useful to ascribe functions to Facebook, even before there was a history of social media platforms to fall back on. These issues are familiar from the rest of this book, and so do not need to be rehearsed here. However, there is also a more specific second problem for the account that needs to be discussed here.

This second problem concerns the fact that the notion of selection at the heart of the account is not fully clear. As formulated by Preston (1998), functional ascription to an artifact depends on whether appealing to effect E contributes to the best explanation of the patterns of use to which past tokens of this type of item have been put, and which in turn have contributed to the reproduction of such items. However, there are a few underspecified aspects of this characterization. It is not entirely clear how much of a contribution to the best explanation the effect has to have—or exactly what this best explanation consists in. Nor is it clear what contribution to the reproduction of the artifact the relevant patterns of use are meant to make. For example, it is not clear whether it matters that the patterns of use incidentally increase (or decrease?) the reproduction of the artifact because they are correlated with something else that affects this reproduction. In general, it is not clear exactly what is meant by the “reproduction” of these artifacts. In short, the relationship between an artifact, its features, its patterns of usage, and its “reproduction” is not fully clear on this account.Footnote 5

All in all therefore, the intentional account has the benefit of avoiding an appeal to the past history of the artifact or unclear statements about its patterns of use and reproduction. However, it is problematic for focusing on the designers’ mental states—which are also often highly unclear—and for failing to pay attention to patterns of use, which do seem to be very important for functional ascriptions to artifacts. By contrast, the historical, reproduction-focused account is good for focusing precisely on these patterns of use and avoiding appeal to the designers’ mental states. However, it is burdened with a historical focus and an unclear account of the role of patterns of use on the reproduction of the artifact. By applying presentist social functionalism to this case, we can get the best of both of these explanations—while still avoiding their pitfalls. The next section makes this clearer.

6.3 The Presentist Social Functionalist Account

To develop a new, presentist social functionalist treatment of artifact functions, the first thing to do is to build on the insight of the historical, reproduction-based account that the best way to analyze artifact functions is by placing them into the context of their pattern of use—that is, to see artifacts and social institutions as closely intertwined.

Technologies and institutions are continuous. As noted in the discussion of Winner (1980) in the previous section, artifacts can affect the world and shape institutions in ways that do not depend on the intentions of users or designers. This continuity between artifacts and institutions at the level of the power they exercise makes it reasonable to try to understand their functions under a common framework. Latour (1990) also argued that technology is “society made durable,” by which he meant that the social world cannot be stable without the presence of non-human actors; society necessarily involves the interplay of both human agents and material things. Technology is in part constitutive of the social, and the social is also in part constitutive of technology (Naoe 2008). Social processes play a part in determining where (and whether) boundaries exist between artifact kinds (Feenberg 1999, chap. 4; Bijker and Pinch 1987).

Most importantly for the present chapter, artifacts and institutions also seem to be related in the sense that we want their functions to perform some of the same discursive work. As noted throughout the book, it is important for social scientists to tell us where a society is going (Bigelow 1998)—in part for its own sake, but also so that we can intervene to affect the future of that society. A principled way of identifying institutional corruption is useful, at least in part, because undesired effects resulting from corruption seem to demand different interventions than undesired effects resulting from an institution that is functioning well. For example, debates about whether the unjust effects of policing demand reform or abolition might be read as hinging on how we identify the function of policing. If presentist social functionalism yields a useful notion of institutional failure—as argued in the previous chapter—then this provides some reason to think that an analogous view will yield a useful notion of technological (mal-)function.

With this in mind, consider how to apply presentist social functionalism to the case of artifacts. Consider an artifact A that is being used in a variety of ways F1 to Fn. For example, this could be a “hammer” that is used to insert nails into walls, to hold books in place, measure the straightness of a board, smash rocks, accessorize clothing, etc. Then, we propose that:

Artifact type A has function Fi (out of F1 to Fn) if usage pattern Fi of tokens of A increases the expected reproductive or survival success of the type A in the current social environment.

To understand this better, several points need to be made.

First, as before, this account is presentist, actualist, and non-arbitrary, and thus avoids the problems of the historical or structural functionalist accounts. This is in much the same way as was the case for the other applications noted in the book, and so there is no need to discuss this further here.

Second, though, the account is also non-intentionalist. In particular, it avoids the problem of the intentionalist account for being focused on the—often unclear—intentions of an artifact’s designers. Indeed, it is also focused on the patterns of use of the artifact, rather than the artifact creation. To understand this better, it is important to note the role that the type/token distinction plays in the above account. The goal of the present account is to assign functions to artifact types (Franssen 2006; Franssen et al. 2014; Evnine 2016) by considering the effects of patterns of usage of their tokens.Footnote 6 In this way, artifact tokens can be seen to get their functions as a result of their membership in kinds.

This distinction matters, as the persistence or reproduction of a type does not always entail the persistence of a token, and vice-versa. Using a hammer as a fashion accessory may increase the survival chance of individual token hammers, but not (it may be presumed) that of hammer types: the type hammer is not likely not stay in existence because of that fact that individual hammers are being used as fashion accessories. No new hammers are produced, and old hammers will increasingly break or get lost. By contrast, using hammers to drive nails into walls is likely to lead to the production of more hammer types, even if it leads to the destruction of individual hammer tokens. For a more extreme version of this sort of case, consider batteries. The use of a specific token battery does not promote the continued use of that token battery beyond the short term, since it is used up in the process. However, it may promote the continued use of batteries in general: if they prove useful for a given purpose, thereby satisfying a user’s intention, they are more likely to be used for that purpose in the future. They may also promote the persistence of their kind by non-intentional means by promoting the use of devices that will require more batteries in the future. Thus, by performing its function, the token contributes to the stability of its type.

This is not to say that artifact types and tokens never line up in their persistence conditions. If my car gets me to work, I’m more likely to keep using it than I would be if it failed to do so. If my car is a very good one, I may want to use it as long as possible; in this sense, the car’s functioning may contribute to its own persistence. This is one way of contributing to the persistence of its type; continuing to use my car is sufficient for continuing to use cars. In many paradigmatic cases, however, artifacts contribute to the stability of their types in broader ways that extend beyond the continued use of a given token. My car affords uses that cause me to depend on cars in general; it allows me to live far from where I work and even in the absence of public transportation, such that I depend on having a car to survive. As a result, I may be more disposed to buy a new car after my current one stops working than I would be if my way of life had not been so shaped by driving. The point is that even when artifact tokens contribute to their own persistence, by shaping users and societies, they typically contribute to their persistence of their types more broadly, as well; however, even contributing to the persistence of a single token is sufficient for contributing to the persistence of a type. In sum: the explanandum of artifact functions on the persistence theory is the stability of artifact types.

This also matters—and this is the third key point to note here—in that it helps to make the relevant notion of “reproduction” clearer. As noted earlier, this was left unclear in the key historical, reproduction-based account. By contrast, the idea here is clearer, and avoids appeal to “best explanations” or vague notions of “contribution.” Rather, the claim is just this: if using a hammer in a certain way makes it more likely for the type hammer to stay in existence or to lead to the reproduction of new, offspring types of hammers, then that way of using hammers is part of the function of hammers. In this way, the account allows for both reproduction and persistence, and as made clearer in chaps. 2 and 3, fills a lacuna left by the historical account.

The fourth point to note here is that, in the background of this account of artifact functions is some way of classifying artifacts into types. This is parallel to the issue of the individuation of institutions and their features noted in chap. 3, though, and thus does not need to be rehearsed here. Also, as was the case there, this again allows for a bootstrapping treatment: an individuation schema for artifact types can, at least in part, be justified though its usefulness in a functionalist treatment of artifacts.

In short, on the present treatment, the function of an artifact is not readable off of the artifact itself, nor off of its design, but depends ineliminably on the social context in which the artifact is used. Artifact types acquire functions through those patterns of use of their tokens that increase their (i.e. the types’) expected reproductive or persistence success. Importantly, this account can also account for the six hallmark features of artifacts that Preston (2009) suggests any adequate account of artifact functions should be able to address: multiple realizability, multiple utilizability, recycling, reproduction with variation, malfunction, and phantom functions. Considering these features is useful for making the nature of the present account clearer.

6.3.1 Multiple Realizability

Artifact functions, in general, are multiply realizable. This means that objects with different physical properties can perform the same function, at least in principle. Corkscrews can be made out of a variety of materials, lightbulbs can use LEDS or wires, etc.—and such differences need not imply different functions.Footnote 7 At the root of this is the fact that, typically, not everything an artifact does is its function, so its non-functional properties can be changed without changing the function. However, it is not always clear which features are functional and which are not, so it is important that a theory of artifact functions can help draw the distinction. The present account of artifact functions does this, and thus allows for, and indeed explains, multiple realizability.

To see this, consider any of the above examples. So, different kinds of corkscrews and different kinds of lightbulbs seem to persist for largely the same reasons (respectively), even when the relevant kinds are made in different ways.Footnote 8 While some things that an artifact does promote the persistence of its kind, other things it does will be irrelevant. The latter features can vary across artifacts that nevertheless have the same function.

6.3.2 Multiple Utilizability

Artifacts are also multiply utilizable, in that they can be put to different uses. Because not all of an artifact’s properties are relevant to its function, any given artifact could likely serve a variety of purposes, including ones that are not its proper function. Thus a spoon can be used like a shovel to dig holes. Digging does not thereby become its function, since it presumably plays no part, or at least a marginal part, in explaining why spoons persist. If, however, agents began to routinely use spoons to dig holes, such that digging became a significant part of the explanation for why spoons persist—in other words, a significant part of what spoons are depended on to do—then the function of spoons would change.

In other cases, materially identical artifacts may have different functions because they are members of artifact kinds that are stable for different reasons. To build on the above example, suppose the type of spoon used in one culture is materially identical to a type of small shovel used in another culture. In the two cultures, the artifacts promote their persistence by producing different effects, and in the face of different selection pressures. Therefore, they form distinct functional types. This is possible because functions do not supervene on physicochemical properties. Identical objects can therefore count as artifacts of different kinds. We need not invoke the intentions of users or designers to explain this, however. Rather, we can focus on the ways that artifacts are used and the roles that they play in society, which need not be intended or acknowledged (see also Lachney and Dotson 2018).

6.3.3 Recycling

Recycling is another hallmark feature of artifacts that follows from multiple realizability and multiple utilizability. It results from the facts that any given artifact has properties relevant to a variety of possible functions, and can thus be re-purposed. Now, when Preston (2009, 2013) considers recycling, she assumes that recycled artifacts are modified in some way to serve their new purposes. If an artifact is simply used for a non-standard purpose without modifying it in any way, that purpose is only a system function.

However, this move is untenable, as non-standard uses for artifacts often become standard without any such modification. There is nothing infelicitous about calling a mason jar used exclusively for drinking a drinking glass, or calling a tire a couch. Modification could be a part of the change in an artifact’s proper function, but in many cases it is unnecessary. The above, presentist-social-functionalist-based account can easily handle this fact: all that matters is that the artifact type is used in different ways than it has been before. Moreover, this account also fits to and improves on Scheele’s (2006) account of function ascription to artifacts. Scheele (2006) considers situation of the “Pieterskerk”—a gothic church in Leiden in the Netherlands that has served a variety of purposes over the course of centuries. Originally intended for religious purposes, it has effectively ceased to be used for religious purposes at all and is now used as an event space. Scheele writes:

Although in most cases the proper functions of artefacts are determined by the designer or manufacturer this is not necessarily the case. An artefact that is used by everyone in a way alternative to the intentions of the designer will very soon change its proper function, due to the fact that this new use will become generally accepted. The reason for this is, roughly, that the socially accepted use will have changed, i.e. the relevant social norms will have changed and thus have overruled the original function ascription. (Scheele 2006, p. 30)

Scheele therefore argues that the function of the Pieterskerk has changed because the consensus about it has changed; after secularization had long led people to treat the Pieterskerk as something other than a church, it was finally bought by a private organization for use on a “quasi-commercial basis” (Scheele 2006, p. 29).

The present account shares with that of Scheele the fact that it deemphasizes the epistemic authority of designers. However, unlike Scheele (2006), the present account of artifact functions does not appeal to the social consensus surrounding the artifact’s function—Scheele’s view remains agent-oriented in a way that the present account is not. This is beneficial for two reasons. First, as before, the determination of when a “social consensus” is reached is difficult. Avoiding appeal to the latter thus makes the present account clearer. Second and most importantly, the present account has the benefit—shared with the treatment of institutional corruption of the previous chapter—that it is gradualist. Suppose that at t1, “spoons” are strictly used for eating foods, and using them as such increases their expected reproductive or survival success. Now suppose that at t2, “spoons” are used solely for digging holes, and using them as such increases their expected reproductive or survival success. The function of spoons will have thus changed: they have turned from utensils to gardening tools (say). Importantly, this change may (though need not) have occurred gradually: for much of their history, spoons may have been used both for eating and digging. The present account can easily allow for this: the ability to be used for digging may have increased the reproduction of the type “spoon” initially not all, then somewhat, then fully determined it. This is likely to be true in many cases of recycling, where artifacts lend themselves to different uses that each contribute to the persistence of some artifact kind in different ways. In this way, the present account can account for the complex, dynamic, and gradualist nature of artifact functions, much as was true for the function of corporations as laid out in chap. 4.

As noted above, the intentionalist theory also has an explanation for function change. In order for an improvised use to become a proper function, on the intentionalist theory, it must be communicated by a justifier who can support the new use with beliefs about the physicochemical properties of the artifact relevant to performing the new function. This requirement is appropriate if a goal of a function theory is to explain the rational justification of artifact use. As noted above, the intentionalist theory involves a social division of knowledge that explains why it is rationally justifiable for agents to use artifacts that appear to them as black boxes (Houkes 2006). However, from the perspective of social science, it is sufficient to explain why agents use artifacts in the ways that they do, even if this explanation does not amount to a justification. Thus, for our purposes, it is preferable to relax this epistemic requirement.

6.3.4 Reproduction with Variation

Preston (2009) laments that the literature on artifact functions, specifically on the intentionalist side, has focused heavily on production to the neglect of reproduction. It is clear that this is not an issue that affects the present account. By serving users’ intentions, artifacts encourage their continued use; alternatively, by creating habits or dependencies in users, artifacts can seem to demand their continued use. However, an artifact kind’s conditions of persistence may change over time as its selection pressures change. Both reproduction and variation can be explained by pointing to the ways that artifacts are used and are socially situated, as is done on the presentist social functionalist account defended here.

6.3.5 Malfunction

Another important phenomenon for a theory of artifact functions to explain is malfunction: how is it that artifacts can have functions that they are not performing? An account must do more than simply describe what an entity is doing in order to capture this normative aspect of functions. As noted in the previous chapters, malfunction is an important phenomenon for any functionalist theory to describe. The presentist social functionalist account of artifact functions defended here defines function in terms of the present contribution of tokens to the persistence of their type. An artifact malfunctions, then, when it fails to be useable in ways that would promote the persistence/reproduction of its type, given the socio-cultural selection pressures it currently faces. If my car’s engine breaks down and I am unable to drive it, it is malfunctioning: the ability to drive them is what underlies the spread of cars (it may be assumed). However, if my car’s engine gets muffled, and I am thus unable to make loud noises with it, it is not malfunctioning: making loud noises is not what drives the spread of cars (it may be presumed). This is exactly parallel to the treatment of institutional corruption offered in the previous chapter.

6.3.6 Phantom Functions

Lastly, an account of artifact functions should have something to say about phantom functions (Preston 1998, 2009, 2013; Holm 2017). A phantom function is one that is ascribed to an artifact that it could never perform—such as an amulet meant to ward off ghosts. As noted in note 65, phantom functions resemble genuine functions in the epistemic division of labor that they produce. I may receive an amulet from a designer who purports to have a mechanistic explanation for how it performs its supposed function; I can use the amulet in my practical reasoning without understanding how it supposedly performs its function, simply by following use plans communicated to me by the maker. However, the stability or reproduction of the amulet cannot be attributed to its actually warding off ghosts. The present theory suggests a response similar to Preston’s (1998) response. Because the persistence theory is a property-oriented, causal-role account, it must explain the stability of these kinds in virtue of things that they actually do. For this reason, phantom functions cannot be genuine functions. Nevertheless, artifacts that have phantom functions, such as fengshui mirrors and magical amulets, are often stable parts of cultures. In this way, phantom functions suggest the existence of genuine functions, even though they can never be genuine functions themselves: these artifacts may have the function to provide comfort, say, or increased aesthetic value to a house.

This move helps to assimilate the persistence-based view of functions with the social sciences. As noted above, Merton distinguishes between the manifest functions and latent functions of social entities (see e.g. Merton 1968). While many social practices and institutions are explicitly designed and have transparent meanings for individuals, they also make contributions to the societies in which they exist, which do not always align with what individuals expect. Similarly, we suggest that philosophers should be skeptical of the relationship between human instrumental reason and the functions of artifacts.

All in all therefore: applying presentist social functionalism to artifacts suggests that the function of an artifact type consists those of the features of tokens of this type that increase the expected reproductive or survival success of that type. This is a useful account of artifact functions, as it avoids the pitfalls of the alternative accounts while preserving their benefits—and having some novel positive features on top of that. In particular, the account is non-arbitrary, non-historical, non-intentionalist, and actualist. It also can account for the multiple realizability of artifact functions, the multiple usability of artifacts, their ability to be “recycled,” the fact that they can reproduce, malfunction, and the existence of phantom functions. In this way, the account allows us to determine which artifacts are stable parts of society and which merely passing fads: by linking artifacts to patterns of use and social institutions, the study of artifact functions is placed directly at the center of the social science. To make this clearer, the next section briefly considers two concrete cases: off-label drug uses and large language models like ChatGPT.

6.4 Two Case Studies

There are many possible case studies that could be considered to illustrate this theory of artifact functions, but the two chosen here have the benefit of bringing out several further points of interest for the discussion of presentist social functionalism that is central to this book. In particular, they emphasize the interventionist role that the present account can play, as well as making clearer what sort of data we should consider in our social scientific investigations of novel artifacts. There is no question that what follows are not detailed treatments of these very complex issues, but merely brief outlines to bring out the fruitfulness of the theory of artifact functions defended in this chapter.

6.4.1 Pharmaceuticals and off-Label Usage

The pharmaceutical industry is heavily regulated, in the sense that drugs, treatments, or medical devices get approved for a specific set of uses. While medical professionals are free to prescribe them for other uses, such off-label prescriptions may or may not be covered by health insurance, and pharmaceutical companies in the US are not allowed to promote them (Wittich et al. 2012).Footnote 9

For present purposes, there are two key things to note about this practice. First, whatever may be true about the regulatory framework and the research and testing that went into drug approval for purposes P, if a drug becomes sufficiently widely used for a different purpose P′, from the point of view of social science, P′ should be seen to be its function. For a concrete example, consider Ozempic, which was developed and approved to lower blood sugar in patients with type 2 diabetes. However, it has come to be widely used as tool for weight loss (Ozempic for weight loss: Does it work, and what do experts recommend? 2023). If this is true, the social function of this drug should be seen to aid in weight loss. That is, the current theory of artifact function makes clear that, apart from regulatory ways to understand drugs, treatments, and medical devices, there is also a social scientific one that might differ from the former. This is again similar to what was the case in earlier chapters: for example, the legal purposes we have when it comes to classifying something as a “corporation” may differ from our social scientific purposes. The same is true: there may be good medical or public policy-related reasons for seeing the function of a treatment or medical device as whatever it got regulatory approval for; however, there may also be good social scientific reasons to diverge from this, and see this function as lying elsewhere.

Second and relatedly, accepting that Ozempic has the social function of a weight-loss tool, it becomes (pro tanto) reasonable for us to take steps ensuring that this use of the drug is respected and enabled. This may include changing the regulatory frameworks so that this use can be advertised, covered, and perhaps even ensured to be safe and effective; however, it may also include other interventions. Among these is the fact that, if the social function of Ozempic (say) is a tool for weight loss—which potentially has a much higher user-based than patients with type 2-diabetes—steps should be taken that this demand can be met. This may include allowing pharmaceutical companies to produce more doses of the drug than would be expected given the population of type 2-diabtes patients: while this may be inconsistent with the approved label, it is in line with the social function of this artifact. This would do justice to social reality, and also allow type 2-diabetes patients continued access to the medicine.Footnote 10

Overall, the key point to note here is that a presentist social functionalist perspective on artifact functions makes clear that off-label uses of drugs, treatments, and medical devices can become the (social) function of these drugs, treatments, and medical devices. In turn, this may require that we shift our interventions in line with this fact. While this is sometimes implicitly recognized—sufficiently widespread off-label uses can lead to a re-classification of a drug—the treatment defended in this chapter puts this reasoning on a systematic basis.

6.4.2 ChatGPT and Other Large Language Models

ChatGPT and other large language models used for predicting the next set of words in a sequence are currently extremely widely discussed, both in professional and private circumstances. This is not surprising. Technology like this has the potential to disrupt many industries, and indeed many aspects of life as we know it. It becomes possible to generate grammatically correct, semantically sensible (at least generally) text for a variety of purposes in extremely short amounts of time.

However, these models also raise the question of what, exactly, they are for? Are these models chatbots? Research tools for linguists or computer scientists? Aids for internet searches? Answering these questions matters, as doing so has implications for how we should treat this new technology. Should it be banned in schools and universities? Should it be encouraged there? Should it be banned or encouraged among government workers? What about among scientists? Are certain uses of ChatGPT and other LLM’s—as a therapist, say, or medical provider—to be discouraged, and others—say, as an internet search engine—to be encouraged? Without knowing the function of LLM’s like ChatGPT, it is difficult to answer these questions. The presentist social functional perspective can help ground answers to these questions.

To see this, begin by noting that this case clearly illustrates the difficulties of the intentionalist account. While nominally designed as a “chatbot,” it is not clear that the designers behind ChatGPT and other LLM’s had or have a specific use of the app in mind. Rather, they were driven to build large language models for their own sake and because they may be useful for a variety of reasons. This generality makes it difficult to assess which uses are properly functional and which are not. For example, in 2023, two lawyers in New York were fined for using ChatGPT to generate legal briefs, which contained references to non-existent cases (Weiser 2023, June 22). Is this a failure of the algorithm—should it have provided them with accurate case law—or a misuse of the system—should they not have used a chatbot for creating legal briefs? Given the indeterminateness of the intentions behind the “design” of the algorithm, it becomes difficult to answer this question—if intentions are seen to ground the function of the artifact. This again suggests that designers’ intentions are simply a poor foundation for ascribing a function to ChatGPT.

Second, recall that the intentionalist account builds on the commonsense association of functions with design. However, machine learning models like LLM’s are largely not designed by humans but produced through an automated optimization process. ChatGPT is based on GPT-3.5 and GPT-4, which were produced using a deep neural network. Partly as a result, such complex models exhibit a great degree of opacity: often, even the “designers” don’t know (and perhaps can’t know) how these algorithms work (Creel 2020). While the designers understand the training algorithms, they often don’t understand how the resulting model produces the outcomes that it does. Thus, the designers of such models lack the epistemic authority that designers are typically assumed to have. To the extent that this is true, functional ascriptions to such models thus fail to meet Houkes and Vermaas (2010) support requirement—and we might thus need to conclude that LLM’s like ChatGPT have no function.

By contrast, the present account is in a much better position here, and can make clearer the kind of information we need to consider to determine the function of ChatGPT and other LLM’s. This information concerns the patterns of use of these LLM’s. The function of an LLM consists in those of its effects that contribute to its persistent use, given the selection pressures it currently faces. So, if it turns out that people mostly use ChatGPT and other LLM’s to avoid figuring out what to write in response to certain questions—whether these be exams or in work-contexts—then this is its function. Like other tools for cheating, there would then be some justification for banning it in schools—despite the fact that it may have other, more positive uses as well. By contrast, if it were to turn out that people mostly use ChatGPT and other LLM’s as a way to quickly obtain information about a variety of issues that they then use in other own, original writing, this use should be encouraged—and this is so even though it can also be used for cheating purposes.

In this way, the presentist social functionalism perspective of artifact functions makes clear what we need to study, from a social scientific point of view, to understand ChatGPT and other LLM’s better. While it is too early to tell what the results are of this study, it is at least clear that what we need to do here. In this way, the presentist social functionalism account of artifact functions can be shown to be a fruitful tool for the further investigation of issues of contemporary social importance.

6.5 Conclusion

This chapter used the perspective of presentist social functionalism to develop a new account of artifact functions. According to this account, the function of an artifact types consists in those usage pattern of tokens of the artifact which increase the expected reproductive or survival success of the type in the current social environment. This account improves on the mainstream intentionalist accounts of artifact functions, as it avoids appeal to the sometimes unclear mental states of designers and does justice to the social nature of artifact use. It also improves on the historical reproduction-based account, as its core mechanism is spelled out more clearly. This account can be shown to satisfy the key features any compelling account of artifact functions should satisfy: multiple realizability, multiple utilizability, recycling, reproduction with variation, malfunction, and phantom functions. Finally, the account is shown to have fruitful practical implications: for example, it can suggest novel interventions in the context of off-label drug uses and make clearer which data should be collected to establish the function of ChatGPT and other large language models.