Progress as Disentanglement

6.1 Accounts of progress

The question whether philosophy makes progress cannot be answered without a view about what progress is, and the philosophy of science offers four. On the epistemic account (Bird 2007), a field progresses when it accumulates knowledge. On the semantic account (Niiniluoto 2014), it progresses when its theories increase in truthlikeness. On the functional or problem-solving account (Kuhn 1962; Laudan 1977), it progresses when it solves more problems, or more important ones, than it did. On the noetic account (Dellsén 2016), it progresses when it increases understanding, where understanding is a grasp of dependence relations that need not amount to knowledge and need not be accompanied by convergence.

Chalmers (2015) applies to philosophy a criterion that combines the epistemic and the convergent: progress on the big questions would be "large collective convergence" on the truth about them. Stoljar (2017) argues that the big questions are not the ones philosophers actually work on, and that on the "topic" and "boundary" questions they do work on there has been a great deal of progress by any reasonable criterion. Dellsén, Lawler, and Norton (2022) argue that the noetic account, applied to philosophy, yields the verdict that it progresses, and that this is not a redefinition but the application of the account of progress that fits science best.

I take the noetic account to be closest to the truth for philosophy, and I take the present account to supply what it leaves open: a specification of what it is that philosophy comes to understand. On the noetic account, understanding is a grasp of dependence relations. The dependence relations that philosophy comes to grasp, on my account, are those among the functions of our load-bearing concepts and the practices they serve: how the concepts hang together, in Sellars's (1962) phrase, and where they come apart. That is a particular kind of understanding, and this chapter argues that it is cumulative, that it is agreed across doctrinal divides, and that it is what the history of a persistent problem has in fact produced.

6.2 Disentanglement

T4 says that progress on a persistent aporia consists in four things. I set them out and argue that each is a cognitive achievement with truth conditions.

(a) Location. A divergence case is constructed: a possible situation in which the application conditions of two or more functions of the concept deliver different verdicts. That a given case does this is a matter of fact, settled by the case-partition criterion of §3.3. Gettier located a region of the divergence space of "knows" that no one had visited; Frankfurt located one for "free"; Parfit for "same person"; Hall for "cause". Locating a divergence region is the discovery of a modal fact about the concept, that its functions can come apart in such-and-such a way, and it is knowledge.

(b) Articulation. The functions whose divergence the case reveals are named and their application conditions stated. That a given thesis of the aporetic cluster articulates a given function is again a matter of fact, and one that can be got wrong: it took the literature some years to see that Frankfurt's case separated leeway from reasons-responsiveness rather than, as was first thought, simply refuting the principle of alternate possibilities. Articulation is the achievement of a description of the concept's functional structure, and it is knowledge.

(c) Disentanglement proper. Distinct concepts are introduced for the distinct functions, by coining terms ("guidance control"), by subscripting ("knowledge₁" and "knowledge₂", "animal" and "reflective" knowledge), or by explicit disambiguation ("in the sense of production" and "in the sense of dependence"). This is a change in the community's conceptual repertoire, and whether it has been made is a matter of fact about the community. Whether it should be made is a further question, but that the resources for making it are now available is an achievement, and one that cannot be undone: by the case-partition criterion, two functions that a case has separated cannot be re-merged by any further case. Disentanglement is cumulative in a way that the position on the headline question is not.

(d) Relocation. The residual question, which function the concept should serve in the divergence region, is recognised as normative and handed to the ethics of the relevant practice. That the residue is normative is a claim with truth conditions, defended in Chapter 5; and recognising it is understanding of the structure of one's own inquiry, of the kind that Dellsén's account counts as progress and that any account should.

Each of the four is agreed across doctrinal divides. Compatibilists and incompatibilists agree that Frankfurt's case exists, that it separates leeway from guidance control, that the two can be given distinct names, and, increasingly, that what divides them is a question about the ethics of blame. What they do not agree on is the headline. The account explains why the four are agreed and the headline is not: the four are matters of fact and the headline is a weighting.

Progress by disentanglement changes the question. "Is free will compatible with determinism?" is not answered; it is replaced by a set of questions, most of which are answered. Is leeway in the unconditional sense compatible with determinism? No, by the Consequence Argument, which is widely though not universally accepted (Lewis 1981 is the principal dissent, and it turns on a further disentanglement, of two senses of "able to"). Is reasons-responsiveness compatible? Yes, and this is not contested. Is ultimate sourcehood compatible? No, and Strawson (1994) argues it is not satisfiable at all. Which of these does the practice of holding responsible require, or should it require? That is the residue, and it belongs to ethics. To answer a question by decomposing it is a form of answering, and for a question posed with a polyfunctional concept it is the only form available. Biology answered "is a virus alive?" in the same way, by distinguishing metabolism from replication from evolvability and noticing that viruses have some and not others; no one regards that as a failure to answer.

A "solution" to a persistent aporia that answered the headline question without decomposition would, on the account, have to be one of two things. It would be an act of conceptual legislation, a decision that one function is to govern, presented as a discovery; or it would be a mistake, an argument that trades on the concept's manifest unity to move from a claim true of one function to a conclusion about another. The history of the persistent problems contains many arguments of the second kind, and the exposure of them is a good part of what the disentanglements have consisted in.

6.3 The record

If T4 is right, the history of each persistent problem should be a sequence of disentanglements, agreed across doctrinal divides and now part of the equipment of every competent practitioner. The following table records the principal ones for the four case studies and for the further cases of §4.6. I include the source in which each was made or made explicit; several were prefigured earlier.

Several observations about this record.

It is long, and it is not exhaustive. Every entry is a distinction that a graduate student in the relevant field is expected to know and that a referee would expect to see respected. The record is, in effect, the content of professional competence in these fields, and it is what distinguishes a competent contemporary philosopher from an intelligent beginner. If that is not progress, it is hard to know what the word could mean.

It is agreed across the doctrinal divides. Process theorists and counterfactual theorists of causation use Hall's distinction; animalists and psychological theorists use Schechtman's; deflationists and correspondence theorists use Tarski's. The distinctions are not the property of a school.

It is cumulative in the strong sense noted in §6.2: no entry has been reversed. Functions once separated stay separated. This gives philosophy an arrow of progress that does not depend on convergence about anything but the distinctions themselves.

And it is understanding in the noetic sense. Each entry records a dependence relation: that the verdict of the concept in such-and-such cases depends on which of two things one takes it to be for. To grasp the record is to understand how the concepts of knowledge, freedom, personhood, and causation hang together with the practices of inquiry, blame, self-concern, and explanation, and where they come apart. That is what Sellars said philosophy was for.

Two entries deserve comment because they show that the record extends into what is usually regarded as philosophy's most settled territory. Frege's distinction between sense and reference was forced by a divergence case: the identity statement. "The morning star is the evening star" and "the morning star is the morning star" have the same reference-determining content and different cognitive value, and the concept of meaning that had served both functions without strain in every other kind of sentence came apart on this one. Kripke's separation of the reference-fixing role of a description from its meaning-giving role was forced by another: the standard metre, whose length is fixed by a description ("the length of stick S at t₀") that does not give the meaning of "one metre", since the stick might have been longer. These are the paradigm results of analytic philosophy, and they have the form the account predicts.

6.4 Re-unification

Disentanglement is not always the terminus. In several cases the disentangled functions have subsequently been re-unified by a higher-order schema in which the choice of function appears as an explicit parameter.

MacCallum (1967) argued that Berlin's negative and positive liberty were not two concepts but one triadic concept, "x is free from y to do z", in which negative and positive accounts differ in what they admit as values of y and z. Lynch (2009) argued that the several theories of truth, correspondence, coherence, superassertibility, were not rivals but accounts of the different properties that realise a single functional role in different domains of discourse. Schaffer (2005) argued that the selection function of "cause" is not a distortion of the concept but part of its logical form, so that causal claims are always implicitly contrastive and the contrast class is a parameter. Contextualism about knowledge (Cohen 1988; DeRose 1992) makes the stringency of the epistemic standard a parameter set by conversational context. And Fischer and Ravizza (1998) retain the single word "responsibility" while making explicit that it is guidance control, not regulative control, that the word tracks.

Re-unification is progress when it makes the choice of function explicit and parametrises it, so that a user of the concept must say, or can be asked, which value of the parameter is in play. The parameter is exactly the function-selection made visible: the y and z in MacCallum's schema encode which impediments and which achievements the concept is to track; the contrast class in Schaffer's encodes which condition is to be selected; the context in the contextualist's encodes which practice's standard applies. A re-unified concept of this kind does not hide the divergence; it displays it, and it reminds every user that a verdict in the divergence region is a verdict relative to a setting.

Re-unification is not progress when it papers over the divergence. Defining knowledge as "justified true belief that is also safe and also creditable", without saying what the concept does when safety and credit conflict, is a conjunction and not a resolution; it inherits the divergence region of each conjunct. Pritchard's (2012) anti-luck virtue epistemology is more careful than this, since it argues that both conditions are needed and says what work each does, but it does not, on my reading, say what to do about Henry, and the account predicts that it cannot without a decision. The test for a good re-unification is whether the divergence case has an answer under it; the test for a bad one is whether the divergence case has simply been forgotten.

The pattern of disentanglement followed by parametrised re-unification is, I suggest, the characteristic shape of a mature philosophical treatment of a polyfunctional concept. It is also, notably, the shape of the treatment that mathematics gave to "number" (natural, integer, rational, real, complex, each with its own function, unified by the algebraic structures they instantiate) and that physics gave to "mass" (rest mass and relativistic mass, separated by special relativity and re-unified by the energy–momentum four-vector, whose time component fixes the relativistic mass and whose invariant length is the rest mass; see Field 1973 for the argument that Newtonian "mass" partially denoted both).

6.5 Anticipation

Philosophy's divergence cases are exotic, and this has always been held against them. What is the use of knowing what we would say about a person who divides, when no one does?

The answer is that the world produces divergence cases, and when it does the community needs a map of the concept's functions and of the costs of privileging each, and needs it quickly.

The ventilator produced one for "death" in the 1960s: a human organism could now continue with all brain function irreversibly gone. The Harvard committee (1968) and the President's Commission (1981) had to decide which function of "death" was to govern, the cessation of integrated organismic function, the cessation of the capacity for consciousness, or the cessation of heartbeat and breathing, and their reports drew on philosophical work about persons and their persistence (Green and Wikler 1980). Assisted reproduction produced several for "mother", "father", and "parent" in the 1980s, separating genetic, gestational, and social parenthood that had coincided in every previous human birth; the Warnock Report (1984) in the United Kingdom, chaired by a philosopher, was the conceptual legislation. The discovery of Eris in 2005, taken at the time to be larger than Pluto and since measured as slightly smaller in diameter and appreciably greater in mass, produced one for "planet", and the International Astronomical Union (2006) legislated three criteria and a new term. The joint-stock company produced one for "person" that the law has been legislating for two centuries. Dementia produces one for "same person" in every case of an advance directive (Dworkin 1993; Dresser 1995). Machine learning is producing them now for "agent", "author", "decision", "responsible", and "understand".

In each of these cases the concept's functions had come apart in reality, a decision was required, and the quality of the decision depended on whether the functions had been articulated and the costs of privileging each understood. Where philosophy had done the work in advance, as with death and personhood, the legislation could draw on it. Where it had not, the legislation was made in haste and has been contested since. Philosophy's construction of exotic cases is, on this account, anticipatory: it maps the divergence regions of load-bearing concepts before the world sends us into them. The person who divides is not going to exist. But the patient whose cortex is gone while her brainstem functions does exist, and the concept that must be applied to her is the one whose divergence region Parfit's cases charted.

This gives philosophy a practical value that is independent of whether it converges on anything. A map of a region no one has entered is not made useless by the absence of a consensus about where in it to build. It is made useful by the arrival of the settlers.

6.6 Reconciling the parties

I return to the three responses of §1.2.

The pessimist is right that the headline questions do not converge and that they will not. Dietrich's (2011) claim that philosophy's problems are "not the kind of thing that can be solved" is, on the present account, correct: they are the kind of thing that can be decomposed and, in the residue, decided, but not solved, because a question posed with a polyfunctional concept about its divergence region has no answer that the concept supplies. Where the pessimist goes wrong is in the inference from non-convergence to non-progress. That inference requires that progress be measured on the headline question, and the account shows that the headline question is the one place where, given the structure of the problem, progress could not show up.

The optimist is right that philosophy progresses, and the account gives the optimism specific content. Stoljar's boundary and topic questions are, in many cases, the decomposed questions of §6.2, and the account explains why they are answerable when the big question is not. Dellsén, Lawler, and Norton's understanding is, in the persistent aporiai, understanding of the functional structure of concepts and its relation to practice. Chalmers's own observation that philosophy's agreed results are negative and conditional is the observation that components (a) to (c) of disentanglement are agreed. The optimist has not needed to redefine progress; the optimist has needed to say what progress on a polyfunctional concept could consist in, and the account says.

The deflationist is right that the first-order dispute has a verbal layer and that the method of elimination exposes it. Where the deflationist goes wrong is in supposing that the residue is empty, that once the verbal layer is removed there is nothing to argue about and the fly may leave the bottle. There is something to argue about, it is normative, and it has consequences; the Wittgensteinian counsel to stop is good counsel for the squirrel and bad counsel for the patient in the bed. Carnap was closer: the residual question is, as he said of external questions, practical. But "practical" does not mean "arbitrary", and the choice of a framework for the concept of death is not like the choice of a coordinate system.

There is a residual worry, which should be met directly. Is progress by disentanglement really progress on the problem, or is it a change of subject? Someone who asked whether we have free will and is told that the question decomposes into three answered questions and one ethical one may feel that her question has been evaded rather than answered.

Three things can be said. First, for a question posed with a polyfunctional concept, decomposition is not an evasion but the only available form of answer; the alternative is legislation disguised as discovery. Second, the questioner's sense that something has been evaded is the sense that the residue has not been answered, and it has not; but the residue is a question in ethics, and the questioner is now in a position to ask it as such, which she was not before. Third, the disentanglements are what the practitioners of these fields themselves regard as progress, as any survey of the literature's self-understanding shows; the account does not impose a standard on the field but articulates the one the field already uses.

The scandal, then, is not a scandal. Philosophy's persistent problems persist because they are questions about the divergence regions of concepts that several practices load at once, and no practice has authority to legislate for the rest. What progress on such questions could consist in, philosophy has made, and made cumulatively. What it could not consist in, convergence on the headline, it has not made, and the expectation that it should was a mistake about the kind of question being asked.