In what follows, I will show that Deutsch’s dilemma for conceptual engineering does not arise. In doing so, I will mainly focus on conceptual re-engineering, for two reasons. First, this is the kind of conceptual engineering that most of the contemporary discussion focusses on. Second, I partly agree with Deutsch that conceptual construction is less new and exciting than conceptual re-engineering. Conceptual engineering would indeed lose much of its appeal if it were nothing but the recommendation that philosophers introduce new technical vocabulary where necessary. I suspect that most conceptual engineers would agree. So, my strategy will be to grant Deutsch the second horn of the dilemma but to reject the first.Footnote 3
Recall that the first horn of Deutsch’s dilemma is a dilemma in itself. The charge is that conceptual re-engineering is either infeasible or trivial, depending on whether it is construed in terms of semantic meaning or speaker-meaning.
My first objection to Deutsch is that this construal of the options is a false dichotomy. It’s not that we can either aim to change semantic meanings, or else settle for individual-level speaker-meaning. There is room for a plausible, implementable and non-trivial project that lies between the two, namely, to convince other members of one’s group to use an existing term in a new way. In many areas of philosophy, and in academic disciplines in general, it is important that the participants in a debate ‘speak the same language’. The fact that their terms have similar or identical semantic meanings is not enough. It is also important that their communicative intentions are aligned in the right way. In other words, it is important for the participants in a debate to understand what others intend to communicate when they use certain expressions. The process of mutual understanding is facilitated enormously if the participants share their communicative intentions. Since certain speaker-meanings are less conducive to successful philosophical or academic work than others—an assumption that conceptual engineers typically make, and which will be scrutinized below—the project of improving what a group of researchers speaker-means by its key terms becomes both salient and important. For this reason, the speaker-meaning construal of conceptual re-engineering need not have the trivializing tendencies alleged by Deutsch.Footnote 4
If this is right, then conceptually engineering group-level speaker-meanings is an activity that is both important and implementable. Now, perhaps Deutsch would grant this much, but object that this activity is simply not exciting and new enough to be called ‘conceptual engineering’ or to deserve its current attention.Footnote 5 Deutsch raises this concern when he discusses the hypothesis that conceptual engineering amounts to ‘stipulative additions’, by which he means the addition of new meanings to terms already in use. Stipulative additions are very close to the proposal currently on offer. And while Deutsch concedes that these do play a legitimate role in philosophy and elsewhere, he argues that if conceptual engineering were construed as the attempt to stipulate additional meanings, then there would be “nothing particularly new or neglected about [it]” (p. 14).Footnote 6
How new and neglected does conceptual engineering have to be for the hype to be justified? Many conceptual engineers happily concede that conceptual engineering is and has long been practiced throughout philosophy (cf. Cappelen 2018, ch. 2; Plunkett 2015; Sawyer 2020; Thomasson 2020). According to them, what is new is not that it is being done, but that it is being recognized and theorized. Moreover, scrutinizing the meanings of one’s terms and engineering new group-level speaker-meanings in the way suggested above has arguably not taken center stage in mainstream analytical philosophy. So while conceptual engineering, in the sense currently at issue, might not be entirely new or altogether neglected, one might still hold that it deserves more attention than it has received thus far.
Putting this aside, I agree with Deutsch that conceptual engineers sometimes have higher ambitions, i.e. they want to change semantic meanings rather than speaker-meanings. I also agree that changing semantic meaning is difficult, and that it cannot be achieved by mere stipulation. The crucial question is whether it is so difficult that conceptual engineers should not even aim to do it—and here I do not share Deutsch’s pessimism.
As outlined in the previous section, Deutsch’s implementation challenge has two components. The first is that the obvious tools of conceptual engineers are ineffective when it comes to semantic meaning change. The second is that we are not in a position to know what else is necessary to bring about semantic meaning change. Deutsch argues that conceptual engineers carry the burden of proof: those who aim to revise semantic meanings “owe us an account of how they can actually succeed in doing so” (p. 7). In what follows, I will sketch one such account, defend it against Deutsch’s objections, and show that it renders conceptual engineering feasible.
Since the first part of Deutsch’s skepticism questions whether it is within our control to bring about meaning change, it is worth reflecting on different notions of control. In Koch (2018), I make two distinctions concerning control that are relevant in this context. The first concerns who we ascribe control to, i.e. individuals or collectives. The second is between immediate and long-range control. Unlike immediate control, exercising long-range control requires engaging in long-term action. Many if not most of our dearest goals, both as individuals and as collectives, are only within our long-range control. Reducing world poverty, increasing social justice, or lowering CO2 emissions, for example, are merely within our (collective) long-range control; but this does not undermine our commitment or even our obligation to pursue them.
What does it take for a term to change reference? Although the early externalists primarily aimed to answer an opposing question, namely, how a term can have stable reference across different times and speakers, this question received some attention in the seventies and early eighties. Most notably, Gareth Evans and Michael Devitt took it to be a key desideratum of causal theories of reference that they also explain and account for the possibility of reference change.Footnote 7 The accounts that both of these authors ended up endorsing go roughly as follows: The reference of a (kind) term t is the (kind of) entity that causally grounds those of a speaker’s beliefs that dispose her to use t (Evans 1973; Devitt 1981).Footnote 8 Thus, if it is experiences with instances of the kind dog that causally ground those beliefs of mine that dispose me to utter ‘dog’, then ‘dog’ refers to dogs. Evans and Devitt note that this version of the causal theory of reference, here only roughly sketched, nicely accommodates cases of reference change. When Marco Polo started applying ‘Madagascar’ to the East African island rather than to Mogadishu (a part of the African mainland), he initially (unknowingly) referred to Mogadishu, because most of the beliefs that disposed him to utter ‘Madagascar’ were causally grounded in Mogadishu. But this changed over time. New beliefs that were grounded in Madagascar, the island, were acquired, old beliefs faded, and preserved beliefs acquired new sustaining grounds. Today, long after Marco Polo’s endeavors, ‘Madagascar’ undoubtedly refers to the island.
This Evans-Devitt-style metasemantics allows for collective long-range control over reference change (Koch 2018). As people apply an existing term to a new kind of entity, they gather beliefs that have the new kind as their causal ground. If, for instance, I tell you about an observation I made about (what are now called) dogs, but I use the term ‘cat’ instead of ‘dog’ in my testimony, then a belief about dogs will enter your dispositional profile of ‘cats’. If this happens again and again, there will eventually come a time when your dispositional profile of ‘cat’ consists mainly of beliefs about dogs. When that point is reached, what you refer to by ‘cat’ will switch from cats to dogs. There is a range of possible views about what happens in the meantime. According to Devitt (1981), ‘cat’, in your mouth, will refer partially to cats and partially to dogs, depending on the overall proportion of cats and dogs in your dispositional profile (p. 193). Evans (1973), in contrast, seems more inclined to think that there is a threshold concerning which causal source is ‘dominant’ (pp. 200–201) such that, once it is crossed, reference switches from one kind to another. Regardless of these details, both views entail that larger collectives, e.g. linguistic communities, can exercise collective long-range control in order to bring about reference change.
Here we have, at least in rough sketch form, an externalist account of how conceptual engineers can pursue their goal of changing semantic meaning. In short: They can use the term in question as if it already had the new meaning, and try to convince as many other people as possible to follow their lead. If they reach enough people and remain consistent, then this will eventually change the term’s semantic meaning—either bit by bit, as suggested by Devitt, or once a certain threshold is crossed, as suggested by Evans.
I anticipate two kinds of objections. You might think that Evans and Devitt’s underlying view is false. Like any other metasemantic view, the Evans-Devitt view is not uncontroversial, and the general idea sketched above undoubtedly needs refinement. It is beyond the scope of this paper to properly defend this view. But notice that all that is really needed to establish the feasibility of intentional meaning change is the idea that if terms are used in new ways by many speakers over long periods of time, then their semantic meanings will eventually change. Evans and Devitt offer specific accounts of why and how this happens that you might object to for one reason or another. But the general point here should be rather uncontroversial, even for externalists. If one denies that even radical changes of use—by however large a group of people, over however long a time period—will change semantic meanings, one is in effect denying that there is any connection between use and meaning. Few philosophers, even staunch externalists, will be happy to endorse this conclusion.
You might also think that the Evans-Devitt view is fine in itself, while maintaining that the power it gives us over semantic meanings is insufficient to block Deutsch’s implementation challenge. This line is taken by Deutsch, who criticizes my view as follows:
[T]he problem with Koch’s picture, even granting an Evans-style metasemantics, is that there is no telling, in advance, that a plan to get large numbers of speakers to use t as if it refers to x will actually turn x into the dominant causal source of the attitudes speakers express when using t. People can make widespread mistakes, for example, using t even in relation to things that are not x’s, despite the intention to use t only in relation to x’s (Deutsch 2020, p. 20).
Deutsch is surely right that given the Evans-Devitt-style metasemantics sketched above, there is no guarantee that a conceptual engineering project will be successful. It is far from trivial to reach or convince the required number of people. Moreover, even if we did convince the required number of people, other things can go wrong. We might, for example, be in massive error about what counts as an x, and thus unintentionally end up applying t to things other than xs.
However, I am not convinced that either of these obstacles gives rise to an implementation challenge that is specific to conceptual engineering. By contrast, I take it that almost any kind of worthwhile long-term project faces these obstacles (and potentially others), but that this does not and should not stop us from pursuing them. Individuals engage in all sorts of long-term activities without having any guarantee of success or antecedent knowledge about their outcomes. Think of activities like obtaining a university degree, pursuing a career, or getting married. As collectives, we negotiate peace treaties with other nations, make plans to reduce CO2 emissions, and to improve our education systems. Arguably, we are morally obliged to engage in some of the activities on this list—despite the fact that our only hope of achieving them is by exercising collective long-range control, despite success not being guaranteed, and despite our lack of antecedent knowledge of whether we will succeed. The problem with Deutsch’s objection, then, is that it runs the risk of overgeneralizing to many human activities and projects that are uncontroversially worthwhile.
In light of this observation, the challenge for Deutsch is to specify some relevant difference between conceptual engineering and other projects such as obtaining a university degree, reducing CO2 emissions, or improving education systems—a difference that explains why the obstacles we face in achieving the latter do not undermine our commitments or even our obligations to pursue them, whereas similar obstacles do undermine any moral or methodological commitment to conceptual engineering. As I’ve argued above, systematic differences concerning their feasibility cannot serve as the explanans that is needed here.