Likewise, just because EC is a thesis about knowledge-sentences’ truth conditions—namely, that they are context-variable—it is not a thesis about knowledge itself. … So it is misleading too when EC … is described, as it sometimes is, as the view that whether one knows depends upon context …

EC, then, is an epistemological theory because, but only because, it concerns sentences used in attributing (/denying) “knowledge”, as opposed to those employing some non-epistemological term(s); it is not a theory about any such epistemic property/relation itself. (Rysiew, ‘Epistemic Contextualism’, §2)

There is a sense in which this is obviously true. Since it is only natural language expressions that are context-sensitive, it will be false (in a category mistake kind of way) that knowledge is context-sensitive (since knowledge is a relation between subjects and propositions, so ipso facto not an expression). There is another (slightly deeper) sense in which it is true: the proposition expressed by ‘A knows that P’ in a context is absolutely true or false. Contextualism is therefore not the view that the proposition K expressed by ‘A knows that P’ is context-sensitive; for all contextualists say, that proposition could be modelled as a function from worlds to truth values. Such a function certainly doesn’t vary from context to context within a world.

But epistemic contextualism is the view that whether a sentence of the form ‘A knows P’ expresses a truth depends on context. Since the T-schema itself is not context sensitive, we know that, when S means P, that S expresses a truth iff P. And we also know that we can give homophonic instances of the T-schema, so it will be true in every context that ‘P’ expresses a truth iff P. One such instance is this:

‘A knows P’ expresses a truth iff A knows P.

Since the left hand side of this invariantly true biconditional depends on context, according to the contextualism, so too the right hand side will. And this dependence can surely be expressed by the words, ‘whether A knows P depends on context’. What I find a bit puzzling is that this is precisely what Rysiew claims is ‘misleading’.

The key might be in what he says later, that contextualism ‘is not a theory about any such epistemic property/relation itself’. If a presupposition of that quoted claim is that there is a single epistemic property/relation which in some privileged way is the knowledge relation, then one of the consequences of contextualism is that there is no such single privileged knowledge relation. In different contexts of utterance, different epistemic relations end up being the semantic value of ‘knows’. And, just like other context-sensitive expressions, e.g., ‘I’, no relation which is the meaning of ‘knows’ in some context is really the knowledge relation in contrast with any of the other relations which are the meaning of ‘knows’ in other contexts. But the claim that there is no such thing as the knowledge relation is surely a theory of knowledge in some sense; since contextualism entails this claim, it does lead to some sort of theory of the putative epistemic relation of knowledge: namely, that there is no such relation. Likewise, that there is no one person who is I, and that there is no one place which is here, are both theoretical consequences of the correct understanding of indexicals; and that there is no height such that all and only people above that height have the putative property tall.

Or—to take a more philosophically interesting example—some compatibilist accounts of free ability have the theoretical consequence that there is no single relation expressed in all contexts by ‘is able to’, and hence that there is no one thing that having an ability consists in. It seems odd indeed to say that such versions of compatibilism are not in fact theories about our ability to do otherwise in some important sense. That is because such a view about ‘is able to’-ascriptions entails that there are systematic possibilities of error in philosophical argument involving such ascriptions, and that therefore the conclusions of various other theories (such as incompatibilism) are unwarranted. That can certainly serve as a reason for rejecting incompatibilism, which seems to be a first-order result about ability.

Similarly, the contextualist thinks that the behaviour of ‘knows’-ascriptions is subject to similarly systematic patterns of error caused by neglect of context-sensitivity, and that instances of sceptical arguments typically exemplify such neglect. This forms the basis for an argument that the sceptical conclusion—that we almost entirely lack empirical knowledge—is unwarranted. And this is, again, a first-order result about knowledge.

Plausibly, as Bach claims in ‘The Emperor’s New ‘Knows’’, contextualists haven’t been clear enough on this; perhaps some really have ‘misleadingly suggest[ed] that the truth value of a knowledge attribution can somehow vary with context while its content remains fixed’ (footnote 12). Given the trickiness of the details here, I think that possibility is both probable and worth drawing attention to, as Bach and Rysiew both do. But it is another thing entirely to conclude on the basis of contextualist sloppiness about semantic ascent that there aren’t powerful conclusions about knowledge that can be reached by means of consideration of contextualism.

For example, Bach (following Heller in ‘The Proper Role for Contextualism in an Anti-Luck Epistemology’) takes issue with Lewis’ claim in ‘Elusive Knowledge’ that ‘when we do epistemology, we make knowledge vanish’. But, says Heller, doing epistemology needn’t alter any of the epistemic relations that a subject stands in—if ‘knows’ prior to doing epistemology designated the relation ‘A has good reasons for their true belief that P’, then A can continue to stand in that relation even when sceptical possibilities are made salient. So, in a perfectly good sense, the knowledge that the subject had—those propositions that the subject stood in the knowing relation to—remain exactly the same after the sceptical possibility is raised. But in another perfectly good sense, knowledge does vanish—because ‘knows’ ascriptions that were true, are now false. So one cannot continue to truly self-ascribe, using the kinds of sentences that we must use to attribute knowledge, knowledge of propositions that one had prior to the raising of the sceptical scenarios. Heller says ’there is no property that [the subject] loses’—while true, this appears to be beside the point, since the contextualist denies that there is any single property of knowing-that-P, the loss of which involves the vanishing of knowledge that P.

❧

I think we can say something more positive about the subject matter of epistemology according to the contextualist. So far I’ve conceded to Bach, Rysiew, and Heller the claim that in a certain sense there is no such thing as knowledge to be the subject matter of epistemology, and as such that contextualism can’t tell us anything about knowledge as opposed to ‘knows’-ascriptions. In this last paragraph, I’d like to push back against that. Take Lewis’ version of contextualism as the starting point. His account of knowledge is that

A knows that P iff A’s evidence eliminates every possibility in which not-P.

In Lewis’ official formulation, he makes explicit a sotto voce provision—‘except for those possibilities that we are properly ignoring’. But we don’t need the provision, provided that context selects a domain of quantification for ‘every possibility’ in the official formulation that consists of the not-properly-ignored possibilities. The thought would be that it is part of the context to supply such a set of possibilities—we might call them the genuine or relevant possibilities—and that interpretation of a quantifier over epistemic possibilities, like ‘every possibility’ in the account of knowledge, is sensitive to that contextually-supplied parameter. It’s a substantive claim, of course, that context supplies any such thing; Lewis gives a long story about which possibilities are those that are not properly ignored, and it may or may not strike anyone as plausible that those conditions could be satisfied by a contextually-supplied implicit parameter. But set that worry aside. The form of Lewis’ account is impeccable: it supplies a context-dependent left hand side of the biconditional to match the context-dependent right hand side; it identifies a plausible mechanism of context-dependence (quantifier domain restriction, which we know antecedently to exist, and seems to make more sense of ‘knows’ that pure indexical or gradable adjective treatments of ‘knows’); and it is true in every context, which makes it at least a viable candidate for an a priori or conceptual truth about knowledge, if you like that sort of thing. Lewis’s proposal specifies something like the Kaplanian character for ‘knows’; and if there is such a character-like thing in the vicinity, then despite having inconstant content, it seems to me that there is a genuine subject matter for epistemology—namely, the concept of knowledge—even though that concept turns out to have inconstant content.

Oren Ambarchi & Joe Talia – Hit & Run (Touch)

Unexpectedly rhythmic outing for Ambarchi when joined by Talia, lots of detail and interest. (Sounded a lot more ‘rock’ actually at Cafe Oto.)

Big Troubles – Romantic Comedy (Slumberland)

Jangly catchy noise pop, just like they used to make it.

Earth – Angels Of Darkness, Demons Of Light I (Southern Lord)

Latest missive from Earth’s exploration of outer limits scorched desert drone.

Isidore Ducasse – Isidore Ducasse (Blackest Rainbow)

Elegant and sparse—supposedly a soundtrack but stands fine alone.

Glenn Jones – The Wanting (Thrill Jockey)

Comprehensive statement by an elder statesman of the acoustic guitar (insightful liner notes too).

Koboku Senjû – Selektiv Hogst (Sofa) 

Noregian/Japanese EAI—warm and complex. Also probably the best show I saw this year.

BJ Nilsen & Stilluppsteypa – Big Shadow Montana (Helen Scarsdale Agency)

Icy drones and hazes, but with a comic undertone; nice follow-up to their earlier Space Finale (Editions Mego) which I also listened to a lot this year.

Peaking Lights – 936 (Not Not Fun Records)

Dubby and detached, a real grower for me this year.

Times New Viking – Dancer Equired! (Merge)

Widely accepted as a step down, on the whole, but I still listened a lot and found much to enjoy.

Mika Vainio – Life (… It Eats You Up) (Editions Mego)

Doesn’t keep up the intensity of its terrifying opening track, but a consistently dark and moody vision.

Veronica Falls – Veronica Falls (Slumberland/Bella Union)

Minimal-ish surf garage with catchy harmonies from an apparent ‘it’-band of the moment.

White Hills – H-p1 (Thrill Jockey)

Some heavy riffage and some spaced noise, soundtracked my driving around for quite a while in autumn.

*

Finally, a special mention for my most regretted purchase of 2011: Fucked Up’s David Comes To Life (Matador). Bloated pseudo-hardcore, like a prog rock Blink 182. I guess I just didn’t get it.

A rational credence function is a probability function. (If you don’t like Dutch book arguments, or representation theorem arguments, that aim to prove this claim, then I will happily take this as an empirical claim—the hypothesis that rational agents have probabilistic credences is acceptable because of its explanatory and theoretical virtues.) Like all probability functions, it’s a function from propositions to numbers. And I follow a standard proposal and say that propositions are sets of possibilities.

But there are lots of different kinds of possibilities. Of interest to us are the epistemic possibilities for the agent with the credences. These are, roughly, all those possibilities not ruled out by what that person knows or could easily find out. And there will be a variety of proposition, epistemic propositions, that are sets of epistemic possibilities. On many views of epistemic possibility, it can be epistemically possible for someone that Hesperus isn’t Phosphorus without it being epistemically possible for that person that Hesperus isn’t Hesperus. So the epistemic proposition that Hesperus is Hesperus can’t be the epistemic proposition that Hesperus is Phosphorus, even if the set of metaphysical possibilities in which ‘Hesperus is Hesperus’ is true is exactly the set of metaphysical possibilities in which ‘Hesperus is Phosphorus’ is true. (To do justice to this sort of commonsensical thought probably involves going two-dimensional; one attractive proposal might be to identify what I’ve been calling the ‘epistemic proposition’ with Stalnaker’s ‘diagonal proposition’, for example.)

It’s been thought to be a problem for the hypothesis that people have credences in propositions if it turns out that propositions true in exactly the same metaphysical possibilities have to get the same credence. So we shouldn’t take metaphysical propositions to be the objects of credence. Credence functions should be defined over the right sort of propositions; it seems very plausible to take them as defined over the epistemic propositions.

Suppose an agent has non-zero credence in ¬p. Then ¬p is an epistemic possibility that has not been ruled out due to inconsistency with what the agent knows (their evidence, if Williamson is right). A non-zero credence in ¬p thus commits the agent to the epistemic possibility of ¬p. Hence it is the case, for all they know, that ¬p; so they don’t know p. By contraposition, knowledge entails certainty.

This conclusion is clearly a species of infallibilism about knowledge. But it is consistent with contextualist varieties of infallibilism, like Lewis’ in ‘Elusive Knowledge’. It therefore has the consequence that one’s credences might vary from context to context; if different epistemic possibilities are those properly considered in different contexts, then one’s credences, defined over the properly considered epistemic propositions, will similarly exhibit contextual variability. (That is: one’s credence function in a context is the one picked out by the context-sensitive expression ‘the credence function which is defined over all the epistemic possibilities’; if quantifier domain restriction alters the extension of ‘all’, it’ll alter the referent of this description.)

❦

One apparent difficulty here is that revision of knowledge will thus involve a non-conditionalising update of a coherent credence. If I forget that p, or my knowledge that p comes to be defeated, that must involve revising my credence in p downwards from 1 to some lower value. And it is familiar that if I update credences by conditionalising, any proposition which has credence 1 in my prior credence Cr_old will have credence 1 in my posterior credence Cr_new. (If we update by conditionalisation, then for some φ, for all ψ, Cr_new(ψ) = Cr_old(ψ|φ) = Cr_old(ψ∧φ)/Cr_old(φ). So if Cr_old(χ) = 1, Cr_old(χ∧φ) = Cr_old(φ), and so Cr_new(χ) = 1 also.) So forgetting or having knowledge defeated will require a non-conditionalising update; and this will conflict with the standard Bayesian view that the only rational way of updating coherent credences is conditionalizing on new evidence. So forgetting or having my knowledge that p defeated involves an irrational update, if this argument succeeds, and that has seemed implausible. Williamson says, ‘Forgetting is not irrational; it is just unfortunate’ (KAIL, §10.2); and if forgetting isn’t irrational, then changing one’s mind in response to evidential defeat isn’t irrational either.

I guess I don’t think it’s that implausible. These revisions of knowledge, on a standard view that does not invoke credences, requires an expansion of a consistent set of epistemic possibilities to include some previously eliminated epistemic possibilities. Even having one’s knowledge defeated involves this; for it involves coming to be convinced that certain eliminated possibilities must in fact be uneliminated. It is not rationally possible, by one’s own lights at t to begin to regard some things which are not epistemically possible at t as being epistemically possible. Forgetting and epistemic defeat, even if they are rational updates, are not rational by the agent’s own lights prior to forgetting and epistemic defeat.

The way I’m thinking about Bayesian conditionalisation is this: conditionalisation is the only coherent rule for updating coherent beliefs on receipt of evidence which one’s prior epistemic state permits a rational response to. One might well agree that there could be cases like forgetting and epistemic defeaters; and that if one has to respond to them, one can’t conditionalise. But the orthodox view would go further, and say that no other rule can be implemented in such cases either; and in fact no rational response to such cases is even possible. For a response to be rational, they might say, it has to be reasonable by your own lights; and since it can’t be reasonable to think that something epistemically impossible has occurred, one can’t reasonably respond to conclusive evidence that something epistemically impossible has occurred. So we might say: if one wants to update rationally, it better be that you use conditionalisation. In cases of forgetting and defeat, one can’t conditionalise; but that doesn’t mean that whatever you happen to do in adopting a new credence function that does accommodate forgetting or defeat is rational. Rather, one can’t rationally respond to a genuine case of (1) or (2). So no rational way to update coherent credences exists other than conditionalisation.

This is not to say that forgetting or defeat is irrational, if that term denotes some kind of epistemic censure as appropriate for someone who updates in that way. Of course, to adopt forgetting as a rule for updating would be irrational. But to forget is, as Williamson says, unfortunate; it can’t be rationally condoned by one’s own lights, but undergoing it is not an irrational update, but a non-rational update.

It’s also true that, having undergone a non-rational update, and being in a new coherent credential state, those non-rational updates don’t look so bad. If at t′ one no longer knows that p, then assuming that all the ¬p possibilities are ruled out is irrational; one’s prior to forgetting state, when one knew that p, looks irrational by one’s current lights at t′, where one’s evidence doesn’t suffice to rule out all the alternatives to p. And this is even more clear in the case of epistemic defeat, where the defeater will right now be sufficent to undermine knowledge that p, and make past knowledge that p look as if it were merely apparent knowledge. That appearance is of course misleading, just as present ignorance is not grounds to undermine future knowledge that p.

This line of argument rests on the idea that assessments of rationality are assessments from the perspective of a given epistemic state; this proposal is then the claim that the only rationally self-vindicating rule (a rule that looks reasonable by one’s present lights) is conditionalisation. This is a pretty weak claim. It’s certainly not the claim that updating in another way leads inevitably to rational regret; non-systematic forgetting leads to a state in which one is perfectly reasonable, and though one may wish one hadn’t forgotten, that’s not because one thinks that suffering that misfortune is also ground for epistemic censure. In Williamson’s terminology (KAIL, §10.2), MONOTONICITY is false—it’s not true that ‘once a proposition has … probability 1, it keeps it thereafter’. But MONOTONICITY is a principle that has the virtue that if one updates is a way one currently regards as rational, one will implement an update that does satisfy MONOTONICITY.

Part of this post was prompted by some discussion with my student Katherine Munn.

Students and others sometimes worry that scepticism is ‘self-defeating’ in some sense. The thought seems to be something like this:

the sceptic couldn’t consistently be a sceptic if their argument is a good one, since the thesis of scepticism will undermine all knowledge of the external world. But then this will undermine knowledge of the premises of the sceptical argument; and an argument with unknown premises couldn’t be a persuasive argument.

Michael Huemer in his book Skepticism and the Veil of Perception discusses something very like this worry (though he certainly doesn’t think it’s a good objection, and he has a useful discussion of why it’s a bad argument against the external world sceptic):

Most people, upon hearing the thesis of philosophical skepticism, immediately come up with the following obvious query for the skeptic: How do you know that skepticism is true? The skeptic is, after all, a person who believes skepticism. If skepticism is true, it follows (right?) that all beliefs are unjustified; therefore, the belief in skepticism itself is unjustified. … If radical skepticism is a challenge to our common sense beliefs, it is, in exactly the same way, a challenge to itself. This being the case, it cannot rationally be accepted at the same time as it itself is accepted; that is, it cannot rationally be accepted. (p. 27)

Some sorts of bad sceptical argument do seem to be subject to this kind of worry. Consider the sceptical hypothesis that all my experiences are being fed to me by a computer. Suppose we ran this argument as an attempt to establish scepticism (here, and throughout, I use ‘I know’ as shorthand for something like ‘I know or am in a position to know’):

1. Consistently with all I know, my experience as of my hands is delusive.
2. If I know I have hands, then I know my experience as of my hands isn’t delusive.
3. Therefore, I don’t know I have hands. (1, 2, modus tollens)

This argument is self-undermining is a some way; for an argument with exactly the same form should be able undermine my other empirical knowledge (not just my knowledge of my hands), and the second premise of this sceptical argument is an empirical a posteriori one. For the conditional premise is a contingent one; we can certainly imagine many circumstances in which the premise is true but the conclusion false. For example, suppose someone knows they have hands, but has never considered the possibility that their hand experience is delusive—a position many of us were in before the sceptical argument was presented to us. (Maybe they can’t even consider that possibility because of some cognitive difficulty or a brain lesion.) But since they’ve never considered that proposition, it seems prima facie that they don’t believe it, and hence cannot know it.

Suppose we try and remedy this defect; rather than considering a conditional premise and using modus tollens, we might consider a premise about knowing a conditional, and use a rule of closure under known implication:

4. Consistently with all I know, my experience as of my hands is delusive.
5. I know that (if I have hands, then my experience as of my hands isn’t delusive).
6. Assume for reductio that I know I have hands.
7. I know that my experience as of hands isn’t delusive (5, 6, closure)
8. Therefore, I don’t know I have hands. (4, 7, reductio)

But this argument too has an empirical premise, since the sceptical hypothesis under consideration isn’t inconsistent with the ordinary knowledge. I may well have hands even if my experience as of my hands is computer generated—this sceptical hypotheses leaves open the possibility of a veridical delusion. But then premise (5) is a posteriori, because there are possibilities in which I know I have hands and the sceptical hypothesis is true. So the conditional ‘if I have hands, then my experience as of my hands isn’t delusive’ is an empirical one, because knowing its truth involves ruling out those genuine possibilities of veridical delusion, and knowledge that one knows something empirical must be a posteriori knowledge.

So we might opt for a sceptical hypothesis that is genuinely inconsistent with the piece of ordinary knowledge—say, the possibility that I am a hand-less Brain-In-a-Vat. Then we run this argument:

9. Consistently with all I know, I am a hand-less BIV.
10. I know that (if I have hands, then I am not a hand-less BIV).
11. Assume for reductio that I know I have hands.
12. I know that I am not a hand-less BIV. (10, 11, closure)
13. Therefore, I don’t know I have hands. (9, 12, reductio)

The argument uses closure of knowledge under known implication to get the sceptical conclusion that I don’t know I have hands. But the premises of this argument are both plausibly a priori, and not empirical at all—and so not subject to the self-undermining worry. Premise (9) is a claim about certain epistemic possibilities, a claim which is established by a priori reflection on which possibilities my evidence conclusively rules out. (We need some sort of principle like ‘knowledge that p involves ruling out all alternate not-p possibilities’ to be true to get any sort of sceptical argument going at all, including this one, but such a principle is very plausible and has the status of a non-a posteriori truth about knowledge.)

Premise (10) is a claim to the effect that I know or am in a position to know a certain logical truth. If I know the logical truth, I know it a priori; and arguably I can know that I know it by a priori reflection. So the second premise is a priori too.

The conclusion (13) of this sceptical argument is to the effect that I do not have a piece of a posteriori knowledge; if generalised, it might undermine all a posteriori knowledge. But the premises of this argument can be known a priori, and so this argument is not self-undermining. The premises of the sceptical argument can thus be known to everyone, sceptic and non-sceptic alike, regardless of the sceptical conclusion, since they are not premises about the ‘external world’ but rather about the internal world.

If the premises are a priori knowable, then how do we escape from this form of scepticism? Here’s a contextualist way out (following Lewis in ‘Elusive Knowledge’). This sort of contextualist notes that the principle that knowledge involves ruling out all alternatives has a context-sensitive expression in it—the word ‘all’ is subject to the context-sensitivity involved in quantifier domain restriction. So if one’s evidence eliminates all such possibilities that lie in the possibilities that are in the contextually-specified domain of possibilities, then it is false that consistently with all I know, I’m a hand-less BIV. Certainly, however, any thought or utterance of ‘For all I know, I’m a hand-less BIV’ makes the possibility of being a hand-less BIV salient, so that premise 1 is still a priori knowable since it is true ‘whenever it is put forward by me or conceived in my mind.’

*

It’s hard to find contemporary philosophers who state the self-refutation charge as crudely as I have above; the kinds of self-refutation that are sometimes levelled at the sceptic nowadays tend to have a more sophisticated character. For instance, Douglas Long considers ‘skepticism in a self-conscious, first person form: “I know that I exist, along with my experiences and thoughts, but I cannot claim to have knowledge of the world around me”’, and then says

I question the coherence of self-conscious skepticism on the grounds that our “introspective” self-knowledge ultimately depends upon our capacities to perceive ourselves as individual agents from an objective point of view shared by others. My thesis is that philosophical doubt undercuts the perceptual access to oneself as an individual subject of mental states that is required both to know of one’s existence as a subject and to be capable of self-reference. Hence, self-conscious skepticism is internally incoherent and self-defeating. (‘The Self-Defeating Character of Skepticism’, p. 68)

The considerations given above don’t engage with Long’s argument, but nor do I immediately see how his thoughts bear on the third version of the sceptical argument I gave.

persisting objects do so by existing entirely at each moment at which they exist. As it is sometimes put, an enduring object is ‘wholly present’ at each moment at which it exists…. The natural way to understand this locution is in terms of [the] notion of wholly located, so that an enduring object is contained within a temporally unextended region at each time at which it exists. An enduring object is such that if it exists at a time, it wholly exists at that time (it is contained in that time). This entails, of course, that enduring objects are temporally multiply located. (‘Location and Perdurance’, OSM 5, p. 57)

(I first thought about endurance in a locational way after hearing a talk by my colleague Oliver Pooley; the articulation I gave it in the paper cited above owed a substantial debt to my soon-to-be colleague Josh Parsons‘ paper ’Theories of Location’, OSM, 3.)

So far, so orthodox. Let me introduce now another idea. In the context of developing a nominalistically acceptable version of a physical theory more current than Newtonian mechanics (and thus—finally—beginning to implement as a program the proposal for nominalistic reconstruction of physics inaugurated by Field’s Science Without Numbers), Frank Arntzenius and Cian Dorr (in ‘Calculus as Geometry’, a chapter from Frank’s forthcoming book Space, Time and Stuff) make the following proposal:

One can assume the existence of a ‘mass space’… Each particle is then assumed to ‘Occupy’ a single point in mass space. … Instead of calling the points of mass space ‘points of mass space’ and saying that particles ‘occupy’ them, one could call them ‘mass properties’ and say that particles ‘have’ them. We take it that nothing substantive turns on this choice of terminology. Calling them ‘properties’ might seem to make the positing of them less controversial. There are some views in metaphysics according to which we are obliged to posit a realm of properties as part of our fundamental ontology in any case, no matter how physics turns out. If one subscribed to such a view, one might see a big difference between thinking of some entities as points in an unfamiliar new kind of ‘space’, on the one hand, and thinking of them as belonging to the familiar category of properties, on the other hand. But this is not our attitude. As we have tried to make clear by talking (most of the time) about ‘predicates’ rather than ‘properties’, we think it is an open question, to be settled on physical grounds, whether we should posit any entities that could by any stretch of the imagination deserve the label ‘properties’. And as we will see, it is quite helpful to think of entities like the ones we are currently contemplating as points in spaces with the same kind of geometrical structure as more familiar spaces.
the positing of ordinary space or space-time is essentially the same sort of move as the positing of mass space: the structure of position properties of particles is … best given by assuming the existence of a structured space-time, and then assuming that each particle occupies a particular region in this structured space-time. So why not similarly assume the existence of mass space, when its structure can so simply and nicely explain the usefulness, and the scale arbitrariness, of the canonical numerical representation of the mass properties of objects, and can do so however few distinct mass properties are had by all existing objects? It seems to us that such a posit might not be so hard to justify on the grounds of the theoretical simplicity it yields. (pp. 19–20)

The context matters for the rationale behind this proposal, since (for their nominalistic purposes) it’s important that there be enough unoccupied points of mass space to give a rich enough structure to represent what we might otherwise have been tempted to represent using a real-valued function, ‘the mass of’. But divorced from that context, there seems a good reason to like this elegant proposal, that unifies locational properties and other sorts of properties—mass in this case, but we could extend it to other properties easily enough. (These other properties needn’t have the rich structure of mass; it could be that some properties have a discrete property space.) We can also extend it to relations—for example, two people stand in the sibling of relation iff they are appropriately located with respect to one another in the relatedness space (maybe we can simply take a directed graph of family relatedness and take the nodes as points, let connected nodes count as adjacent, let betweenness amongst nodes count as locational betweenness in this space, etc.—the details matter, but I haven’t taken the time to work them out). Or maybe, a la Humean Supervenience, the non-locational relations turn out to supervene on the pattern of occupied locations in spacetime and in the fundamental property spaces (or more sensibly, to supervene on the pattern of occupied locations in the product space of spacetime and all of the fundamental property spaces).

I should note too my agreement with Frank and Cian that, while this proposal is supposed to be able to make sense of our ordinary property talk, it is not a notational variant of the view that philosophers ordinarily adopt about properties. The postulation of these property spaces is not, for example, a system for representing facts about universals/tropes/etc.—we are to take these spaces to exist, to be genuinely real (as real as spacetime is); and the idea that having a property is being located in a property space is meant literally too—location is not being used in some merely analogical fashion. So the view is quite radical. But it has some very attractive theoretical consequences.

Endurantists of the locational stripe, I think, have especially good reason to like this model. For having permitted multiple location, to account for the phenomenon of change, endurantists have to relativise property possession (in the ordinary instantiating-a-property sense) to locations. This move, though technically unproblematic and easy to implement, is ideologically weird in at least two ways. (These ways aren’t noted very explicitly in the literature, if at all, so its also probably good to record them here as difficulties for other conceptions of endurance.)

• As location shows, there is nothing fundamentally objectionable about an object instantiating a non-relativised property, like being located. So it seems a puzzling brute metaphysical fact that there aren’t other non-relativised properties, like being square, etc. This is because, from a metaphysical point of view, there seems nothing to rule out the existence of such monadic properties modelled on being located. The worry is that even if it turns out that all English monadic predicates express temporally relativised properties, there is no reason why there aren’t monadic properties, and no reason why we couldn’t introduce new predicates to English that express such properties, in which case we’d be able to run a problem of change with respect to those monadic properties. If it’s only a contingent matter that there are no such properties, we would have possible inconsistencies arising due to change, so there must necessarily be no such unrelativised changeable properties. And brute metaphysical necessities are hard to accept.
• Relativising properties to regions means that, in some sense, there is no change in objects over time: an object which stands in the square relation to t, and the round relation to t′, stands in both of those relations at all times. This gives rise to two puzzling features:
  • Why does the object look different at different times, if in fact it has not changed in any of its genuine properties? The explanation on offer—facts about which time is present have changed, leading to different temporally-relativised properties being salient—seems to locate the change in the wrong place, in the passage of time rather than in some change in the object itself.
  • Less gesturally than the previous concern: what an object is like at a region turns out to depend on relations it stands in to other regions; plausible recombination principles generating other possible spacetimes turn out not to hold (one cannot simply adopt something like a patchwork principle, and generate a new possibility by ‘stitching together’ regions from one spacetime with non-overlapping regions from another spacetime to generate a new spacetime). This will be because the ‘stitched together’ spacetime may contain objects with relational properties relativised to regions that don’t exist in the new spacetime—what kind of problem this gives us (truth value gaps maybe?) isn’t clear, but it’s not unproblematic. And yet recombination principles are our best guide to the contents of the space of possibilities. We might resolve this worry by suggesting that how an object is at a place is an intrinsic matter, and intrinsic properties at a region are just the relations the object has to that region. But such a proposal seems worryingly ad hoc: why should a relation to a time correspond to an intrinsic property, and another relation to a different time correspond to an extrinsic property, just because of the time at which the object’s properties are being evaluated, which seems an even more extrinsic and ephemeral fact.

If endurantists adopt the property space conception of properties, these worries all dissolve. It turns out that all property possession fundamentally involves occupying a certain region. Thus we have a principled reason for thinking that the monadic property being located is distinctive; it is a member of a family of properties that concern the basic necessary conditions on having a property at all. Moreover, we now have a reason why relativising properties to locations looks like a legitimate move at all, and not just an instance of what Sider calls the ‘glib response’ which invokes relativisation to time in an apparent mere redescription of the phenomenon to be explained (Four-Dimensionalism, pp. 92–3). It’s a legitimate move because to be an object with properties at all is to be located somewhere in product space, one of the constituent spaces of which is spacetime, so that ‘x is red at R’ is just something like ‘x occupies 〈Red,R,…〉’. (This does raise some apparent puzle, namely why we don’t equally index our predications by other spaces—it should be equally legitimate to say that something is round at red, but that’s not permitted in English. The answer I suspect is that while in the metaphysics room ‘at’ does represent this general locational relation, and while indexing by kinds of regions other than spatiotemporal regions is legitimate, it is nevertheless difficult in English to get ‘at’ to have anything other than a spatiotemporal reading. Same with ’change’—we use it for variation over a temporal dimension, and also [though less frequently] for variation over a spatial dimension, but even though we could use it for variation indexed to other dimensions of property space, we can’t get the English expression to have that more general reading easily. I could speculate about why it’s difficult to get these general readings, but doubt it would be convincing.)

With regard to the second sorts of worries, I think we have an answer too. The fundamental facts about the history and nature of an object are, on the property space conception, facts about trajectories through the most general property space (the product space of all fundamental property spaces). And just as it is trivial to ask when an object has a certain historical trajectory over all time, so it is similarly trivial to ask when an object stands it certain relations to times. We’re already happy with the idea that, in some sense, we shouldn’t expect change over time in the pattern of locations of an object over spacetime. But we’re happy enough to use a trajectory through spacetime to answer certain momentary questions about where an object happens to be located at a time, and to characterise the trajectory of an object as representing the object’s locomotion, change in position over time. So we should similarly be happy on this view to say that, in some atemporal sense there is not much interest to the question of when an object has a certain trajectory through property space, but we can use that trajectory to answer certain momentary questions that arise about where an object happens to be located with respect to some property or other at a time, and to characterise the trajectory through general property space as representing the object’s changing character over time.

With this framework, the question ‘why does the object look different at different places along its trajectory’ is just a generalisation of the question ‘why does the object appear to have a different location at different places along its trajectory’, and gets a similar answer. An enduring object appears to change location (even while to have a trajectory at all is to have, eternally, many locations) because the trajectory of the observer happens to coincide in some sense with the trajectory of the object—it’s my changing perspective on the trajectory which gives rise to the appearance of change of location, since from now the object’s trajectory looks like it is located here, and from then it looked like it was located there. So too, it turns out, with change in other properties: the appearance of change is generated by my changing perspective on the trajectory. This changing perspective, of course, needn’t be changing in some genuinely passage-y sense of change; it can perfectly well be the much more mundane matter of being located at different places at different times.

And the trajectory framework also gives us recombination. For if we take a general property space, populated by objects with certain trajectories (which means a general property space and a location relation), then we’ll be able to carve that space up arbitrarily and recombine it with other spaces as long as we don’t try to recombine spaces that would end up ‘clashing’. (How to make sense of the no-clashing provision is made easier if we assume that locations in general property spaces are identifiable across worlds, but thought behind it is just the same as goes into the patchwork characterisation of recombination.) We don’t have objects at a location having to stand in relations to other regions to say how they are at that region—how they are at a region is a matter of what part of the general property space that region corresponds to. So we have reintroduced genuinely intrinsic properties, since all changeable properties will be intrinsic (or at least internal) features of the region of property space where an object is located at the time of property ascription.

This move also gives endurantists a response to the problem of temporary intrinsics. As I just mentioned, I think we have a principled distinction between intrinsic and extrinsic properties, corresponding to whether the property is characterisable as an intrinsic property of a property space region or not. But we also now have a principled reason to reject the Lewisian intuition that some things are ’just plain straight’—for straightness, no more or less than any other property, is about occupying a certain region of the location-property product space. There is nothing more to being an object that having a certain trajectory through this product space, so no sense in which an object is somehow ‘lacking’ a shape if it occupies a certain region of shape space. The intuition that Lewis has now comes to have a new character: it is the intuition that some properties are special, had in some different way than the relevant object being located in the appropriate region of a property space. We can quite legitimately insist, I think, on being given an answer to the question of what this ‘instantiation’ relation is if its not like the familiar occupation relation—and what these mysterious properties are. (Since I also think that the debate over the metaphysics of properties is one of the less interesting bits of contemporary metaphysics, I’m happy to simply sidestep those debates in favour of the much better understood notion of location—though I note with some displeasure a certain traitorous character to this claim given my Australian roots.)

There are a couple of residual problems I’m aware of.

1. What if location relations aren’t fundamental? I’ve got some thoughts about why this might be so. One is the argument that fundamental location has bad consequences. For one thing, the existence of extended simples falls out pretty directly:
   “The possibility of extended simples follows from plausible principles about location and possibility; mereology has nothing to do with it. The principle about location is that location is a fundamental relation between objects and points of space. The principle about possibility is a combinatorial principle requiring, roughly, that any pattern of instantiation of a fundamental relation be possible. These principles imply the possibility of the location relation’s holding in a one-many pattern between a mereo- logically simple object and points of space—an extended simple” (Sider, ‘Parthood’, §1). But, we might worry, it shouldn’t be so easy to prove a substantive metaphysical thesis. Worse consequences still than the possibility of easy arguments in tricky bits of metaphysics might be various kinds of locational exotica that we can apparently get via recombination (such as the ‘crowded simples’ and ‘compact fusions’ that Raul Saucedo discusses in ‘Parthood and Location’, OSM 6). We might find the implausibility of these exotica to be a reason to deny the recombination claim which generates them. Less indirectly, there are reasons (controversial reasons, it has to be said) from quantum mechanics to be suspicious of fundamental location. In particular, since position is an observable that we don’t need to represent the quantum state in terms of (if we choose as a basis the conjugate quality of momentum), we can completely describe a fundamental supervenience basis for quantum reality without mention of location—and if locational facts supervene on position facts, and position is supervenes, then location isn’t fundamental. I find these lines of objection not extremely compelling, but they are enough to give us pause. And, actually, the present framework has a natural response to the quantum mechanical case: namely, that the position observable itself supervenes on a system being located somewhere in quantum mechanical state space in the fundamental sense. Location is fundamental; location in position space versus momentum space is not fundamental, no more fundamental than location in any of the other constituent property spaces.
2. What to say about mereological relations? For it doesn’t seem straightforward to regard x and y’s standing in the part-whole relation as a matter of location of x and yin some mereological property space.
   • We could make the attractive move of identifying objects with their trajectories (so-called supersubstantivalism); then our fundamental location relation is just identity. And since we have, as part of the setting up of our property spaces, a relation of subregion, we could adapt that into a parthood relation: x is part of y iff x is a subregion of y. We could do this without identifying objects and their trajectories, and accept the doctrine Gabriel Uzquiano calls ‘Mereological Harmony’ (OSM 6), though accepting Harmony without identifying objects and locations is subject to counterexamples from recombination if location is fundamental.
   • We could deny that there are any interesting mereological relations amongst material objects. That is, we could be nihilists. Indeed, this view gives an argument for nihilism: there is one fundamental relation on this view, location; if there were composite objects alongside the simple objects that are their basic parts, we get puzzling restrictions on the pattern of obtaining of the fundamental relations (specifically, that we cannot generally locate the parts of an object elsewhere than the object). (We might also think that enduring simples are more readily believed in because not subject to intrinsic change of the sort sometimes taken to be problematic for endurantists.)

The OTT VANMOOF No. 3

It also attracts some admiring glances—white wall tires and leather grips/saddle, along with the unusual geometry, make it stand out, and I do really enjoy riding something so cleanly designed. But I most look forward to attaching a child trailer and hauling little Sylvester around town once he’s old enough.

Sleeping Beauty is a paradigm of rationality. On Sunday she learns for certain that she is to be the subject of an experiment. The experimenters will wake her up on Monday morning, and tell her some time later that it is Monday. When she goes back to sleep, they will toss a fair coin. If the outcome of the toss is Heads, they will do nothing. If the outcome of the toss is Tails, they will administer a drug whose effect is to destroy all memories from the previous day, so that when she wakes up on Tuesday, she will be unable to tell that it is not Monday.

Let P be Beauty’s credence on being awakened on Monday, and P⁺ be her credence after being told that it is Monday. The problem is this: it seems that upon awakening on Monday, Beauty should have her credence in the coin landing Heads unchanged: so P(Heads) = 0.5. After all, she knew all along on Sunday that she would be awakened, and awakened she has been; and the mere awakening is not evidence of bias in the coin. But this seems unsustainable, since she should also have credence in Heads of 0.5 after being told that it is Monday (since she then finds out that a coin she believes to be fair has yet to be tossed), so P⁺(Heads) = 0.5. And these claims are inconsistent.

These two claims are inconsistent, given these further truths about the scenario:

• P and P⁺ should be related by conditionalization, so that P⁺(⋅) = P(⋅|Monday);
• P(Heads|¬Monday) = 0;
• 0 < P(Monday) < 1.

I used to favour (though I did not like) the solution which denied the claim that P and P⁺ were related by conditionalization. I thought that there were good reasons to think that P(Heads) = 0.5, given by Lewis (in ‘Sleeping Beauty: reply to Elga’); and I thought that there were good reasons to think that P⁺(Heads) = 0.5, given by Elga (in ‘Self-locating belief and the Sleeping Beauty problem’). But now I’m convinced that Elga was right, and Lewis was wrong.

Here’s the line of thought which has convinced me. Beauty forgets something between Sunday night and Monday morning. For on Sunday night she knew the self-locating proposition expressed by a Sunday utterance of ‘Tomorrow will be Monday’; whereas she did not know that proposition on being awakened on Monday, for she did not know the proposition expressed by a Monday utterance of ‘Today is Monday’, and they are the same proposition.

What Beauty knew on Sunday rules out a certain hypothesis, given everything else that Beauty knows about the scenario:

(NotYet) The coin toss hasn’t yet occurred by the time of Beauty’s next awakening.

Since she forgets the proposition expressed by ‘Tomorrow will be Monday’, she is on awakening on Monday unable to rule out the hypothesis (NotYet), since for all she knows, it is Tuesday, and on Tuesday the coin toss has already occurred. So on forgetting what she knew on Sunday, Beauty should reason as follows on being awakened on Monday:

Since I don’t know whether it is now Monday or now Tuesday, there are two hypotheses about the chance of Heads that are compatible with what I know. Either the coin has not yet been tossed, and the chance of Heads is 0.5; or the coin has already been tossed, and the chance of heads is either 0 or 1 (since the past is no longer chancy). If the chance of heads is 1, it must be Tuesday, and moreover, I would know it was Tuesday (since I would remember being told yesterday that it was then Monday). So if the chance of Heads is either 0 or 1, it must be 0, since otherwise I would know it was Tuesday, and I don’t. So my credence in Heads now should—in line with the Principal Principle—be a weighted average of the possible values of the current chances. So my credence in Heads now should be a weighted average of 0 and 0.5, and must therefore fall strictly between 0 and 0.5.

(The Principal Principle entails this (suppressing other admissible evidence):
P(Heads)
= P(Monday)⋅P(Heads|Monday) + P(Tuesday)⋅P(Heads|Tuesday) [Theorem of total probability]
= P(Monday)⋅Ch(Heads|Monday) + P(Tuesday)⋅Ch(Heads|Tuesday) [the PP]
= P(Monday)⋅0.5; [Given the actual values of the chances]
< 0.5. [given that P(Monday) < 1]
)

This line of argument already shows that something in Lewis‘ argument for P(Heads)=0.5 is incorrect. Lewis’ premise is

(L1) Only new relevant evidence, centred or uncentred, produces a change in credence; and the evidence [of being awakened with one’s last memory being Sunday evening] is not relevant to Heads versus Tails. (p. 174)

But the second clause of this claim is false, as Beauty forgets some information that ensures the evidence about awakening is irrelevant. And indeed she does; for she forgets whether or not yesterday evening was indeed Sunday evening. And if yesterday evening wasn’t Sunday evening, then while she still has no relevant evidence about the fairness or otherwise of the coin, she does have evidence that—given what she knows—might bear on the outcome of the coin. For it might be that the coin has already landed tails, for all she knows. (Lewis may be being misled by Elga’s claim that, during the relevant interval, Beauty ‘suffers no cognitive mishap’ (p. 146), whereas forgetting clearly is a cognitive mishap of some sort.)

The reasoning Beauty has already gone through already shows that P(Heads) ≠ 0.5, as long as she assigns non-zero credence then to its being Tuesday (which she does, having forgotten that yesterday was Sunday). But what weights should she assign to the hypotheses about the chances? It will be Beauty’s credence on awakening that it is Monday that does the work; and there are a number of arguments in the literature to the effect that, having forgotten what day it is on awakening, she should have equal credence in all the possible awakenings with this character (namely, the character of not knowing what day it is). There are three possible awakenings that have this character: namely, the coin lands Heads and it’s Monday; the coin lands Tails and it’s Monday; and the coin lands Tails and it’s Tuesday. (The arguments on offer for this equal credence conclusion range from applications of a restricted principle of indifference, to appeals to the relative frequency of the different kinds of awakenings—however this goes, there look like good reasons for Beauty to be indifferent between these possible awakenings, whether or not that indifference arises from the application of anything like a Principle of Indifference.) In any case, given her equal credence in these awakenings, P(Monday) = 2/3; and therefore, by the PP argument above, P(Heads) = 1/3.

So what happens to P⁺(Heads)? The natural thing to say is this. On being told that it’s Monday, Beauty learns again the proposition she’s forgotten. And that puts her back into the epistemic position she was in on Sunday night, knowing (as she did then) that she would be awakened on Monday and told that it is Monday. So the natural thing to think is that this information rules out exactly the hypothesis that it might be Tuesday; and the only hypothesis about the value of the current chances that remains in contention after (re-)learning that fact is the hypothesis that the chance of Heads is 0.5. So, by the earlier reasoning, P⁺(Heads) = Ch_Monday(Heads) = 0.5.

This argument that P⁺(Heads)=0.5 is an application of the Principal Principle: that Beauty lacks inadmissible information about the future on being told that it is Monday. But the original Lewis argument in favour of P(Heads) = 0.5 also uses the Principle Principle, claiming that Beauty lacks inadmissible information about the future on Sunday, and she continues to lack in on being awakened. Dorr notes that

however plausible it is that she lacks inadmissible evidence about the coin-toss afterwards, it is at least as plausible that she also lacked inadmissible evidence about the coin-toss to begin with. (p. 293)

I agree. But Beauty’s forgetfulness shows another way that one can come to have a different opinion; one loses some admissible information, and comes thereby to regard a hypothesis about the chances that one had previously ruled out, as being consistent with the admissible evidence one has. In short, I’m persuaded by this line of argument that something like Monton’s contention is correct (in ‘Sleeping Beauty and the forgetful Bayesian’):

it is Beauty’s forgetting her temporal location which justifies her change in credence. (p. 52)

In part the debate over the Principal Principle has concerned whether Beauty gains inadmissible information on waking—e.g., Chris Meacham claims that ‘if Elga’s argument is sound, you do get inadmissible evidence when you wake up on Monday’ (‘Sleeping Beauty and the Dynamics of De Se Beliefs’, §5). But if the above discussion is right, that’s not correct—Beauty loses information on going to sleep, not through trauma or drugs but simply through losing the ability to discriminate her own present temporal location. But that is not at all gaining inadmissible information. But this actually puts me in agreement with Meacham, who continues the remarks quoted above by saying that admissibility is the ‘wrong way’ to approach the problem. It’s the loss of evidence, permitting new chance hypotheses to be considered, rather than the gain of evidence on awakening that matters in this case.

*

Added 26 July 2011: Alastair Wilson gave me some comments by email, and I’ve made a couple of changes to clarify the above argument. He also emphasised something I didn’t emphasise enough: the role of chance, and even more importantly the Principal Principle, in setting up the Beauty scenario. (I think he has a draft paper on why chance is important in Sleeping Beauty, if you’re interested.)

The reasoning above, as you’ll have noticed, doesn’t make much of the self-locating nature of the content that Beauty forgets—she could have forgotten anything, as long as forgetting it opened up new possibilities about the chances. After those new possibilities are available, she is forced to adjust her credences to be the expectation of the chances in line with the Principal Principle. Because I’ve said nothing about the relationship between Beauty’s credences on Sunday and her credences on being awakened, and I’ve said nothing about the dynamics of de se content involved, it’s harder to see how to extend the reasoning directly to some of the interesting variant cases.

But in the most natural variant, where the coin is tossed on Sunday and Beauty is awakened once or twice in line with its outcome, it seems clear that Beauty’s credences on Monday should be the same as in the case I discussed. (Elga, in comparing the two variants, says ‘Your credence (upon awakening) in the coin’s landing Heads ought to be the same regardless of whether the researchers use method 1 [toss on Sunday] or 2 [toss on Monday]. So without loss of generality suppose that they use — and you know that they use — method 2′ (Elga, §2).) But the argument for thirderism doesn’t directly go through in the variant case, since there isn’t any obvious way in which Beauty can reason in the way sketched above. There are of course other arguments she can use—maybe Elga’s original argument; maybe an argument that she does get some evidence on awakening, namely the evidence ‘I’m awake and I don’t know what day it is!’, and that evidence is more abundant if the coin landed Tails on Sunday than if it landed heads, giving P(Heads) < 1/2 (if that evidence confirms Tails more than Heads and Beauty knew the coin was fair on Sunday night). Maybe she can even use the above argument, and agree with Elga’s contention that she should treat the original case and this variant equally. But these arguments are going to be quite different from the one I’ve given above.

In light of this, the real reason I’m now a thirder is probably that I’m convinced by the above reasoning that being a thirder is the correct response to that variation on the Sleeping Beauty case, and that makes thirderism a more palatable view more generally. With thirderism itself rendered thus more attractive, I’m not opposed to extending it across the board in the way I once was.

Stories needn’t display their ‘fictionality’ on their surface. They are written in ordinary natural languages, and generally needn’t exhibit any distinctive grammar or meaning. One could imagine an alien, or a naive but precocious child, venturing into Blackwell’s, and mistaking a copy of Bleak House for a historical work on the Court of Chancery. They could understand the work, so their linguistic competence is not at issue. Their mistake rather lies in thinking that Bleak House was asserted by Dickens as known fact. But, of course, Bleak House doesn’t assert that it is literally true of English justice in the early nineteenth century, and the action involving the main characters isn’t believed by Dickens to have occurred, since he didn’t believe those characters really existed.

This rather prosaic phenomenon opens up an intriguing possibility. If it’s not always possible to tell from grammar and meaning alone whether some sentence is an assertion, then perhaps some ordinary sentences that we generally take to be assertions are, in fact, not assertions at all. Perhaps some ordinary discourse is rather more like storytelling than we normally assume.

Some philosophers have thought this about moral discourse—talk about which actions are ‘right’, or ‘virtuous’, or ‘permitted’, etc. While the form of a moral judgment looks like an ordinary declarative sentence—’Giving to charity is right’ is rather like ‘Giving to charity is popular’ in structure—a number of philosophers have noted that the aim of moral judgment doesn’t appear to be stating truths, but rather to motivate certain behaviours. The meaning of ‘Giving to charity is right’ is that there is a certain queer kind of property, rightness, and that a certain type of action possesses it. But on the proposed view, it needn’t be that I, when I utter that sentence, commit myself to any beliefs about the existence of such a property—rather, I make that utterance to express my motivation to give to charity, and to encourage you to share that motivation (or something along those lines).

Such a view—that moral discourse is in some relevant respects like fictional discourse—is called, obviously enough, moral fictionalism. There are a number of pressing problems for the moral fictionalist. Perhaps the most pressing is: why? If the point of moral utterances is to motivate certain behaviours, why do they take the potentially confusing form of a declarative sentence?

Richard Joyce, in his 2006 book The Evolution of Morality, offers an answer. Joyce is interested in helping behaviour: acts that, while beneficial to others, are somewhat detrimental to the actor. If community life involves helping behaviour, that can be of great benefit to the community as a whole. (Think of a community of hunter gatherers, where a successful hunter can either gorge themselves, thus feeding one member of the community to excess, or share with others, thus giving every member of the community a minimum amount of food for survival.) But individuals have no personal incentive to give up their own wellbeing for others, even if the outcome of doing that is (overall) better. So how could helping behaviour evolve and persist in a community? Joyce’s answer is: because moral judgments claim that certain actions are objectively and impersonally encouraged or forbidden, someone who hears a moral judgment is prompted to believe that certain behaviours are objectively demanded of them. If the actions that are encouraged include the helping behaviours (as seems generally true of our moral judgments), then making moral judgments will tend to cause the presence of helping behaviour in the community. So—in a nutshell—the practice of making moral judgments in the form of declarative sentences exists because it encourages helping behaviour in individuals (even when they would rather not engage in helping behaviour), and helping behaviour generally enhances the relative success of communities in which it exists.

This evolutionary account of moral judgments is very controversial, of course. And attentive readers will already be devising objections to the proposal I described. But it, combined with the fictionalist framework, provides a novel account of a very tricky philosophical problem, an account which promises both to render morality less mysterious, but also allows it to continue to play a crucial role in our lives and relations to others.