Disagreement Reconsidered

5.1 Three layers

When a compatibilist and an incompatibilist disagree about whether Jones, in Frankfurt's case, acts freely, what exactly do they disagree about? The question sounds naive, but the persistent aporiai are, I shall argue, precisely the disputes in which it has no naive answer, and the account of Chapter 4 makes it possible to give a careful one.

Any dispute conducted with a polyfunctional concept in its divergence region has three layers.

The factual layer comprises the questions about the case whose answers do not depend on which function of the concept one privileges. Is Jones's action produced by a reasons-responsive mechanism? Could Jones have done otherwise, given Black's presence? Was the choice determined? Is Henry's belief safe? Did his perceptual faculty function normally? Is the bottle's shattering counterfactually dependent on Suzy's throw? Is there a process connecting them? These are questions with answers, and in a well-constructed case the answers are stipulated. The parties do not disagree about them. Where a dispute is about them, as when compatibilists and incompatibilists disagree about whether determinism is true, it is an ordinary factual dispute and the epistemology of ordinary factual disputes applies.

The verbal layer comprises the question whether the concept's term, "free", "knows", "cause", "the same person", is to be applied to the case. Once the factual layer is settled and the functions are articulated, this question can be seen to have no answer independent of a decision about which function's application condition is to govern. The parties disagree here, but their disagreement is, at this layer, a disagreement about the use of a word, and it is at this layer that the deflationist's diagnosis is correct.

The residual layer comprises what is left when the factual questions have been answered and the verbal question has been identified as such. I shall argue that something is left, that it is a normative question about which function the shared concept should serve, that it is substantive, and that it is the actual content of the persistent dispute.

The three-layer structure explains E7 of §1.4. The parties are right that they disagree about something substantive, because the residual layer is substantive. The observer is right that they talk past each other, because at the verbal layer they are applying different application conditions to a case about whose facts they agree. Both are right about different layers, and the account of this chapter is an account of the relation between them.

5.2 Against the verbal diagnosis

Chalmers (2011) offers a method for detecting verbal disputes. Given a dispute in which a term T figures, bar T and its cognates from the vocabulary and ask whether the parties still disagree about anything they can state. If not, the dispute was verbal; if so, the residual disagreement is the substantive content, and one may proceed to argue about that. Chalmers argues that many philosophical disputes are at least partly verbal in this sense, and he expresses the hope that the method of elimination can clarify them and, in some cases, make them go away.

Let us apply the method to the fake-barn case. Bar "knows" and its cognates. Is Henry's belief safe? The parties agree: no; he could easily have been wrong. Did Henry exercise a reliable perceptual competence? They agree: yes. Would you rely on Henry as a source of information about barns? They agree: yes, in general, though not in this county. Is Henry's belief true and well-founded by ordinary standards? They agree: yes. At the first-order level, nothing remains. The dispute about whether Henry knows has dissolved into agreement about everything that can be said without the word.

But now ask: may Henry assert "that is a barn"? Should Henry act on his belief? These questions are not about the word "knows"; they are about the practices of assertion and action. And here the parties may well disagree, because the norms of assertion and action are not stipulated by the case, and because which epistemic condition should govern them is not settled by the facts about Henry. The anti-luck theorist who denies knowledge will typically hold that Henry's assertion is defective, since assertion requires a position from which one could not easily have been wrong. The virtue theorist who grants knowledge will hold that it is not, since Henry did everything an asserter can do. The method of elimination has not dissolved the dispute. It has relocated it: from a dispute about the application of "knows" to a dispute about what should govern a practice that "knows" is used to regulate.

Chalmers anticipates this. He allows that disputes over the use of a term may be worth having when the term is tied to practices with consequences, so that how the term is used affects what people do; a dispute over whether a certain arrangement is a "marriage" is verbal at the first-order level and substantive at the level of legal and social consequences. What he does not do, and what I am doing, is to argue that all of philosophy's persistent aporiai are disputes of this kind, and to explain why. They are disputes of this kind because the concepts at their centre are load-bearing for several practices; and the method of elimination, applied to them, always relocates rather than dissolves, because barring the term leaves the practices in place with their conflicting demands.

Plunkett and Sundell (2013; see also Sundell 2011) give the relocated dispute a name and an analysis. A metalinguistic negotiation is a dispute in which the parties, by making first-order claims that appear to conflict ("waterboarding is torture"; "waterboarding is not torture"), are in fact negotiating how a word should be used, and are doing so because the word's use has consequences that both parties care about. The dispute is not about the word's current meaning, which may be indeterminate over the case; it is about what its meaning should be. Plunkett and Sundell argue that such disputes are genuine disagreements, expressing conflicting attitudes about a matter both parties regard as important, and that they are substantive whenever the word matters.

The persistent aporiai are metalinguistic negotiations in this sense. "Jones is free" and "Jones is not free", said of the Frankfurt case, are moves in a negotiation over how "free", and with it the practice of holding responsible, should treat agents who act on their own reasons without alternatives. The negotiation is genuine because the practice matters. It is stable because, as §5.4 argues, nothing available to the parties settles it. And it is conducted in the first-order idiom of "analysis", as though it were a dispute about a fact, for a reason the next section gives.

Why do the parties not see their dispute as a negotiation? Two reasons. First, the concept presents itself, in Haslanger's (2012) terms, as a manifest concept with a single content; that is how it was learned and how it functions across the coincidence range. The idiom of analysis, of asking what the concept "really" applies to, is the appropriate idiom in the coincidence range, where there is a determinate answer, and the parties carry it into the divergence region, where there is not. Second, the practices at stake are ones the parties are deeply invested in, and a normative disagreement about a practice one is invested in has the phenomenology of a factual disagreement: it feels as though something is true about whether Jones is responsible, and not merely to be decided. That phenomenology is, I shall argue in §5.7, not wholly misleading, because normative questions may have answers. But it misleads as to the kind of answer.

5.3 The residue is normative

The residual question of a persistent aporia has the following form.

(5.1) Given a divergence case in which the application conditions Aᵢ and Aⱼ of concept C conflict, which of them should govern the application of C in cases of this kind, given the practices that C serves and the ends those practices have?

This is a normative question. It asks what should be done with a shared instrument. It is not answered by the facts of the case, which are agreed. It is not answered by the concept, which, having been learned in the coincidence range, contains no instruction for the divergence region; the manifest unity of the concept is silent exactly where the question arises. It is answered, if at all, by considerations about the ends the practices serve, the costs of serving one at the expense of another, the feasibility of institutional arrangements, fairness to those affected, and coherence with the rest of the practice. These are the considerations of conceptual ethics (Burgess and Plunkett 2013), and the residual question of a persistent aporia is a question in conceptual ethics.

Two versions of the question should be distinguished. The internal version takes the practices as they are and asks which of their ends most needs the concept's service in the region at issue. The ameliorative version (Haslanger 2012) asks how the practices themselves should be arranged, and what concept would serve the practices we ought to have. The internal question is the one the parties to a persistent aporia are usually, implicitly, arguing about. The ameliorative question is deeper, and answering the internal question often reveals that it is the ameliorative one that matters.

The clearest confirmation that the residue is normative is that the debates have, in their most self-aware recent phases, turned into ethics. Consider free will. The decomposition of §4.3 leaves the factual questions largely agreed: leeway in the unconditional sense is incompatible with determinism, reasons-responsiveness is compatible, ultimate sourcehood is incompatible and perhaps unsatisfiable. What remains is the question which of these the practice of holding responsible requires, and that is a question about what blame is for. It is answered differently by those who take blame to express basic desert (Pereboom 2014, in the negative), by those who take it to be a form of moral communication or address (McKenna 2012; Fricker 2016), by those who take it to express the significance of relationships (Scanlon 2008), and by those who take it to be a forward-looking instrument for shaping conduct (Vargas 2013; Caruso 2021). The compatibilism debate, decomposed, is the ethics of blame, and the recent literature has increasingly conducted it as such.

The same is true elsewhere. The knowledge debate, decomposed, becomes a set of questions about the norms of assertion, action, and inquiry: what epistemic position should one be in to tell others things, to rely on a belief when much is at stake, to stop looking? These are questions in the ethics of epistemic practice, and the work of Williamson (2000), Hawthorne and Stanley (2008), and Fantl and McGrath (2009) is, whatever its authors' self-understanding, a contribution to them. The identity debate, decomposed, becomes a set of questions in medical and legal ethics: whose interests govern the treatment of the demented patient, whose consent binds, when a life has ended for the purposes of organ retrieval. The causation debate, decomposed, becomes a set of questions about the aims of the law and of the several sciences: what relation liability should track, what relation an explanation in the special sciences should cite, whether fundamental physics needs any such relation at all. Woodward (2003) is explicit that the interventionist account is designed for a purpose and is to be judged by how well it serves it.

I do not claim that the parties to these debates have all recognised that this is what they are doing. I claim that it is what they are doing, that the account predicts they would end up doing it, and that the trajectory of the debates confirms the prediction.

5.4 Why it is stable

Why does the residual disagreement not get settled? Five features of the situation jointly explain its stability.

Underdetermination by the concept. Nothing in the concept, as learned from cases in the coincidence range, fixes the relative weights of its functions. The parties cannot appeal to "what the concept really means" because in the divergence region it does not really mean any one thing; the appeal is to a manifest unity that the case has dissolved. Arguments from further cases can locate the divergence more precisely, but they cannot produce a weighting, because the cases are the very thing the weighting is needed to adjudicate.

Plurality of ends. The functions serve ends that are genuinely plural and not reducible to a common measure. The practice of holding responsible serves fairness to the agent, protection of others, communication of norms, and the expression of attitudes that are constitutive of certain relationships, and these do not reduce to one another. This is a special case of the plurality of value that Berlin (1958), Nagel (1979), and Williams (1985) have described, and Rawls (1993, Lecture II) listed among the "burdens of judgement" that make reasonable disagreement about fundamental matters the expected condition of free reflection: the difficulty of weighing considerations of different kinds, and the fact that "there are different kinds of normative considerations of different force on both sides of an issue". Where ends are plural, reasonable people weigh them differently, and the difference is not a failure of rationality but its expected output in the absence of a common measure.

Absence of a legislating authority. In the domains where divergence cases get settled, one practice has authority over the concept. The International Astronomical Union can legislate what "planet" means for astronomy because astronomy is the practice that uses the word for anything that matters; when it did so in 2006, adopting three criteria and introducing "dwarf planet" for bodies that met two, the dispute ended, not because anyone discovered what Pluto is but because the body with authority decided. The Supreme Court of the United States could rule in Nix v. Hedden (1893) that a tomato is a vegetable for tariff purposes, prioritising the culinary function of the word over the botanical for the practice of customs, because the practice of customs has a court. Philosophy's concepts are load-bearing for several practices at once, law, science, morality, medicine, and ordinary life, none of which has authority over the others. No one can legislate what "free" means for all the practices in which responsibility figures, and so no one does.

Disciplinary home. The parties characteristically approach the concept from a home practice that they implicitly privilege. Epistemologists trained in the philosophy of science approach "knowledge" as a concept for inquiry and weight the inquiry-closing and informant functions; those trained in ethics approach it as a concept for assertion and action and weight the licensing functions. Free will theorists who are metaphysicians weight leeway and sourcehood; those who are moral psychologists weight reasons-responsiveness and reactive aptness. Training entrenches the weighting and supplies the cases that display it to advantage. The positions are, in this respect, sociologically as well as normatively stable.

Self-selection. Finally, philosophers do not arrive at positions on the aporiai by weighing the arguments from a neutral standpoint; they arrive with weightings and find the arguments that express them. This is not a criticism. It is the natural consequence of the fact that the residual question is normative and that normative outlooks precede philosophical training. But it means that the accumulation of arguments about cases cannot be expected to change the distribution of opinion, because arguments about cases sharpen the map and do not change the weights; and arguments about weights, which are arguments in ethics, move slowly and are themselves subject to reasonable disagreement.

Together these explain E5. The distribution of professional opinion on a persistent aporia is a distribution of weightings; weightings are stable under argument about cases; and there is no authority to impose one. The figures of §1.2 are what one should expect.

5.5 The epistemology of disagreement

A literature of the last two decades has asked what one should do, epistemically, on discovering that an epistemic peer, someone with the same evidence and comparable competence, disagrees with one. Conciliationists (Feldman 2006; Christensen 2007; Elga 2007) argue that one should move one's credence substantially towards the peer's; steadfast theorists (Kelly 2005; van Inwagen 1996) argue that one may retain one's view. Applied to philosophy (Kornblith 2010; Goldberg 2013; Frances 2014; Ballantyne 2019), conciliationism seems to recommend suspending judgement on nearly every contested philosophical question, since on nearly every such question there are peers on every side.

The present account changes the terms of this discussion in three ways.

First, it separates the layers. On the factual layer of a persistent aporia, philosophers largely agree, and where they disagree, as about determinism, the disagreement is an ordinary one to which the ordinary epistemology applies. On the verbal layer, there is nothing to conciliate about, since there is no fact. The residual layer is where the question of peer disagreement bites, and there the account says something specific.

Second, it changes the object. The conciliationist framework presupposes a shared proposition about which the parties have the same evidence. The residual question is not of that kind. It is a question of how a shared instrument should be used, given plural ends. The parties do not have "evidence" for their weightings in the way one has evidence for a factual belief; they have reasons, drawn from the ends they take the practice to serve, and the reasons on the two sides are, by the burdens of judgement, of different kinds. It is not clear what "moving one's credence towards the peer's" would even mean for a weighting; one can split the difference between two probabilities, but a weighting that gives leeway and reasons-responsiveness equal priority is not a compromise between the two positions but a third position with its own costs.

Third, it changes what suspension would cost. On a factual question, suspension is always available and often wise. On the residual question, suspension is frequently not available, because the practices must act. A court must decide whether the drunkard is to be punished; a hospital must decide whether the patient is dead; a community must decide whether to blame. Where a decision is forced, the epistemically appropriate attitude is not suspension but what I shall call commitment with awareness: holding one's weighting, arguing for it in the currency of ends, and recognising the opposing weighting as a weighting rather than as an error. This is close to Beebee's (2018) "equilibrism", on which the aim of philosophy is to achieve and maintain an equilibrium among one's commitments rather than to attain knowledge. The present account explains why there are several equilibria and what distinguishes them: each is a coherent weighting of the functions of a load-bearing concept, and the equilibria differ in which practice they let govern the divergence region.

None of this makes philosophical belief a matter of mere preference. Normative questions may have answers, and a weighting may be better or worse supported by the ends it is meant to serve. But it means that the epistemology of philosophical belief, in the divergence region, is the epistemology of normative belief, not of factual belief, and that the pessimist who applies the latter to it has misidentified the object.

Kornblith (2010) raises a worry that any account of this kind must face. If peer disagreement undermines philosophical belief, it undermines belief in the account itself. The reply is that the account is a claim at the factual and explanatory layer, an abductive hypothesis about the structure of a class of problems, and is answerable to evidence in the ordinary way. Disagreement about it is disagreement in an early-stage inquiry, of the kind that is expected to resolve as evidence accumulates, and the account claims no immunity from it. What the account claims is that the persistent disagreements of philosophy are not of this kind, and it gives a reason.

5.6 Lewis's menu and van Inwagen's insight

Two remarks by two of the most careful philosophers of the last century can now be read as reports of the structure I have described.

Lewis, introducing the first volume of his Philosophical Papers, wrote that the reader in search of knock-down arguments would be disappointed, and that "once the menu of well-worked-out theories is before us, philosophy is a matter of opinion" (1983, x–xi). The remark is usually read as a shrug, or as a confession of the limits of argument. On the present account it is neither; it is accurate. The menu of well-worked-out theories is the map of a concept's functions and of the costs of privileging each; assembling the menu is the work of philosophy that can be done by argument, and it is cumulative and agreed. What remains when the menu is complete is the choice of weighting, and that is a matter of what Lewis called opinion because it is a matter of how one weighs plural ends. Lewis was not saying that philosophy is arbitrary. He was saying that the part of it that argument can settle is the part that assembles the menu.

Van Inwagen, confronting his own disagreement with Lewis about compatibilism, wrote that since he and Lewis had the same evidence and Lewis was at least as able as he, he must suppose that he enjoyed "some sort of incommunicable insight" that Lewis lacked, on pain of admitting that his belief was unjustified (1996). The remark has been much discussed as a statement of the steadfast view. The present account offers a better description of van Inwagen's situation than the one he gave. He and Lewis did not differ in insight. They differed in weighting: van Inwagen gave precedence to leeway, Lewis to the compatibility of ordinary practice with determinism, and each had reasons of the kind appropriate to weightings. The "incommunicability" van Inwagen felt is the mark of a disagreement that is not about a fact accessible to both parties but about the relative importance of ends both recognise. Neither had a defect the other lacked. Both were rational. And the account explains the phenomenology reported in §2.2, of a solved problem the other side refuses to accept: each had solved the problem relative to his weighting, and each experienced the other's weighting as a refusal.

5.7 Relativism

The account will be suspected of relativism. If the residual question is answered by a weighting, and weightings vary, does it not follow that whether Jones is free is relative to a weighting, and that anything goes?

Four replies, in increasing order of importance.

First, most of what we say with these concepts is in the coincidence range, where verdicts are determinate, objective, and shared. The account introduces no relativity there. Henry's neighbour, who looked at a barn in a county without façades, knows it is a barn, on every weighting.

Second, within the divergence region, not every weighting is admissible. Weightings are answerable to the ends the practices serve, and some weightings would defeat those ends. A concept of responsibility that gave no weight to reasons-responsiveness would make blame useless as communication and unfair as a response; a concept of legal causation that gave no weight to selection would make liability unbounded; a concept of knowledge that gave no weight to truth would not flag informants. The space of admissible weightings is constrained by the point of the practices, and arguments about which weightings are admissible are arguments with better and worse answers.

Third, the residual question is normative, and whether normative questions have objective answers is a question in metaethics on which the account is neutral. If normative realism is true, then there is a correct weighting of the ends of the practice of blame, and the compatibilism debate is an objective dispute about it, no less objective for being ethical. If normative realism is false, then philosophy's residual disputes are as objective as ethics is, and no less. Either way the account adds no relativity beyond what one already accepts about ethics. What it denies is that the residual question is a descriptive question about the concept with a fact of the matter that the concept's content supplies. That denial is not relativism; it is the recognition that the concept's content runs out.

Fourth, the account provides a criterion that the relativist cannot, for distinguishing disputes worth having from disputes not worth having. James's squirrel (1907, Lecture II) is the paradigm of a dispute not worth having. A squirrel on a tree trunk keeps the trunk between itself and a man who circles the tree; does the man go round the squirrel? "Going round" has two functions, tracking passage through the compass points and tracking passage around the object's sides, and they coincide for every ordinary object and diverge for one that keeps turning to face you. So far the case is a divergence case in my sense. But no practice with stakes cares which function governs; nothing anyone does depends on the answer; and so the correct response, which James's campers accepted at once, is to distinguish the senses and drop the question. The persistent aporiai are not like this, and the account says why: their concepts are load-bearing for practices whose ends conflict in the divergence region, and something people do depends on the answer. A dispute is worth having, on this account, exactly when the metalinguistic negotiation has stakes. That is a criterion, and it is one that the parties to philosophy's persistent disputes, who have never been tempted to treat them as squirrel cases, have all along been applying.