To clarify the target of this chapter’s arguments, this section has provided background on the various ways in which theorists have understood the project of analysis. Given the decline of the traditional project of conceptual analysis, the arguments ahead are primarily directed towards proponents ofmetaphysicalanalysis.
Let’s say that X has a unified nature iff there is some single, natural feature that is necessary and sufficient for being an instance of X. Let’s say that X has a discoverable nature iff learning X’s nature involves genuine discovery (i.e., extends beyond mere re- flection on our concepts). Then, differences notwithstanding, proponents of metaphysical
8See Fair (1979, 231), Bigelow & Pargetter (1990), and Dowe (2000, 3).
9While some theorists (e.g., Earman (1976, 24), Tooley (1987, ch. 7)) require metaphysical analyses to
be necessary, other theorists (e.g., Aronson (1982, 302), Salmon (1984, 239-242), and Dowe (2000, 3)) only require truth in the actual world. Paul & Hall (2013, 40-41) take a middle position, claiming that metaphysical analyses should not rely on facts “too specific to one’s own world.”
10Given this sufficient condition, here are some examples of metaphysical analyses in the literature.Cau-
sation: Aronson (1971), Mellor (1988, 229-230), Bigelow & Pargetter (1990, 278), Dowe (2000, 11), Wood- ward (2003, 115-117). Composition: van Inwagen (1990, see esp. 66-68), Merricks (2000, ch. 1). Desire: Morillo (1990), Arpaly & Schroeder (2013, ch. 6).Knowledge: Kornblith (2002, 1-2).Possibility: Armstrong (1989, ch. 4).Laws of nature: Swoyer (1982, 221-222).
analysis share a major assumption:that the itemXunder analysis has a discoverable, uni- fied nature. The arguments ahead will show how we might challenge this assumption in a given particular case.
To make this challenge concrete, consider again the contrast between water and game- hood. Intuitively, the project of metaphysical analysis seems interesting for water but not for gamehood. This is because, unlike water, gamehood doesn’t seem to be a thing with a discoverable, unified nature. I want to raise the following question: why should we as- sume that things like the causal relation, dispositions, etc. are like water instead of like gamehood?
2.3 ‘water’ vs. ‘game’
In this section, I will identify two important semantic differences between the terms ‘water’ and ‘game’. These differences explain why the project of metaphysical analy- sis is legitimate for water but not for gamehood. We will then have a test for whether a philosophically-interesting itemXcan be given a metaphysical analysis by seeing whether the term ‘X’ shares these features.
2.3.1 Epistemic security
Suppose that all of our ordinary evidence suggests that a certain sampleX is water: it has the right appearance, taste, boiling point, etc. On the basis of this evidence, we would justifiably endorse the sentenceP ≡‘Xis a sample of water’. But notice thatP is hostage to empirical fortune: if we were to find out thatX has a different chemical structure from the one shared by all the other water-like liquids in our environment, we would revise our judgment and conclude thatP is false. This shows thatP’s truth presupposes thatXshares some unified nature with other water-like samples in our environment.11
11This is not to say that we can knowa priorithat water has a unified nature. For example, if it had turned
out that half of the water-like stuff around us was H2O while the other half was some other chemical, it is plausible that our term ‘water’ would have referred to both types of substances (see Block & Stalnaker (1999,
But the following thought experiment shows that game judgments are different. Sup- pose we become convinced that a certain theoryAis the best possible analysis of gamehood (on whatever desiderata for theory choice one prefers12). But suppose (plausibly enough)
that there are stillsome cases whereAconflicts with our intuitions; for example, perhaps Aclassifies multiplication tables as games.
Now consider: how would our belief thatAis the best possible analysis of gamehood affect the game-judgments we are disposed to make in ordinary contexts? For example, how would it affect our disposition to assert Q≡‘Multiplication tables are not games’ in ordinary contexts? I submit that, even if we were convinced thatA was the best possible analysis of gamehood, we would feel no pressure at all to revise our judgment that Q(in any ordinary context). We would continue to assert Q just as we always had.13 This is
because, when judging whetherY is a game in ordinary contexts, we do notcarewhether it shares some unifying feature with other cases of games.14
This thought experiment reveals an important difference between ‘water’ and ‘game’. The fact that we revise our water-judgments when we learn that X does not share some unifying feature with other water-like samples suggests that ‘water’ expresses a property with a unified nature. But the fact that we feel no similar pressure to revise our game- judgments suggests that ‘game’ does not express a property with a unified nature.15
21ff) for discussion). To speak more precisely:P’s truth presupposes thatX shares some unified nature with the other water-like samples around usconditional on the supposition that these other samples (generally) share a unified nature. I will leave this precisification tacit because it will not affect the arguments ahead.
12For example, perhaps we become convinced thatAupholds more of our our intuitive judgments than
any other comparatively simple analysis.
13By “ordinary context”, I mean: contexts where we aren’t engaged in the philosophical task of analyzing
“gamehood”. An example of such a context might be: a competition where players quickly write down the names of objects within various categories beginning with a given letter. If someone wrote down “multipli- cation tables” as their response to the category “game” and the letter “M”, we would protest.
14Note that nothing in this argument hinged on the choice ofQas an example; we wouldn’t feel pressure
to changeanyof our determinate game judgments in order to make them conform to some neat analysis.
15Might the difference in revisability betweenP andQinstead be explained by a difference in our con-
fidence in these judgments? For example, might our unwillingness to revise our judgment thatQreflect the fact that, unlike in the water case, we are more confident inQthan in any proposed analysis conflicting with
To capture this difference, I will say that game judgments areepistemically securein a way that water judgments are not: our judgment thatX is a game is not threatened by the possibility thatXfails to share some unifying feature with other games.16
2.3.2 Epistemic rigidity
The term ‘water’ is metaphysically rigid: it refers to the same substance across meta- physical possibilities. But ‘water’ is epistemically non-rigid: it refers to difference sub- stances across epistemic possibilities. For example, since the dominant water-like sub- stance in our actual environment turned out to be H2O, the term ‘water’ refers to H2O in
the actual world. But if the dominant water-like substance in our actual environment had turned out to be XYZ, the term ‘water’ would have referred to XYZ.17
The epistemic non-rigidity of the term ‘water’ explains the sense in which an analysis of water involves substantive discovery. Before we did chemistry, there were many open epistemic possibilities about what ‘water’ refers to; perhaps it refers to H2O, perhaps it
refers to an element, etc. But by performing experiments, we learned which substance ‘water’ actually refers to. This result closes epistemic space and is justifiably considered to be a substantive discovery.
In contrast, the term ‘game’ seems epistemically rigid. Although we can imagine
it?
In response: while it may be true that we are more confident inQthan any analysis conflicting with it, this difference betweenPandQpotentially obscures a more important difference between them. The more important difference is that, when judging thatX is water, wecare(at least implicitly) thatX shares some unifying feature with other water-like samples — this is why we revise our judgment thatP upon learning thatXhas some different chemical structure. But when making game judgments, we donotcare whether the given practice shares some unifying feature with other games.
This difference — a difference in what we care about — is the real explanation of why game-judgments cannot be threatened in the same way as water judgments. Indeed, this explanation shows that it would be a mistake to think of our game-judgments as being weighed against theories of gamehood in the same way that our water-judgments are (apparently) implicitly weighed against theories of water.
16Of course, this is not to say that game judgments are epistemically securesimpliciter. For example, we
might mistakenly judge thatX is a game because we are not fully aware of the details ofX (its rules, its aims, its history, etc.).
“twin-worlds” where functional and phenomenal duplicates of us use the term ‘water’ to refer to XYZ, we cannot imagine twin-worlds where functional and phenomenal duplicates of us use the term ‘game’ to refer something other than games.18 So unlike the water case,
there are no discoveries for us to make about what ‘gamehood’ refers to in the actual world. Given this disanalogy, it is unclear in what sense an analysis of gamehood could involve substantive discovery.
2.3.3 A test
Both ‘water’ and ‘game’ are famous examples of the failure of traditional concep- tual analysis. But only in the case of ‘water’ does the alternative project of metaphysical analysis seem well-motivated. I’ve explained this intuition by identifying two semantic dif- ferences between the terms ‘water’ and ‘game’. Because of the epistemic security of game judgments, there is no reason to think that ‘gamehood’ refers to a property with aunified nature. And because this term is epistemically rigid, there is no reason to think that it refers to a property with adiscoverablenature.
This discussion provides a test for when metaphysical analysis is appropriate. Con- sider any philosophically-interesting term ‘X’ (e.g., ‘cause’, ‘free’, ‘law’, etc.). If we de- termine thatX-judgments are epistemically secure and that the term ‘X’ is epistemically rigid, it is inappropriate to ask for a metaphysical analysis ofX.
2.4 ‘cause’ vs. ‘water’
To illustrate this test, I will consider the term ‘cause’. I will argue that, when we compare the term ‘cause’ to terms like ‘game’ and ‘water’, the close analogy is with the term ‘game’.
18Of course, we can imagine worlds where speakers use the term ‘game’ to refer to other types of things:
2.4.1 Epistemic security
As discussed in 2.2.2, metaphysical analyses are not required to agree with our intu- itive judgments across all possible cases. For example, on Aronson’s (1971) transference theory of causation, it is incorrect to say that the ice’s melting caused the water to cool (424- 425). On Dowe’s (2000) conserved quantity theory, causation by omission (e.g., ‘John’s failure to take the medicine caused his illness’) is not genuine causation at all (124ff). Paul & Hall (2013) argue that metaphysical analyses need not uphold our causal intuitions about exotic possible cases, such as those involving magical spells (40-41).
The common problem with these proposals is that our causal judgments are not the types of things that can be falsified through philosophical theorizing. To be sure, we often revise our causal judgments in response to receiving moreordinaryevidence; for example, we would revise our judgmentR≡‘The melting ice caused the water to cool’ upon learning that there wasn’t actually any ice. But we would never revise this judgment simply because a philosophical theory of causation conflicted with it.
For example, suppose we become convinced that the best theory of causation (on whatever desiderata for theory choice one prefers) is Aronson’s transference theory. Even if this were the case, we would feel no pressurein ordinary contextsto revise our judgment thatR. We would continue (in ordinary contexts — say, when swimming in a cold lake) to assertR just as we always had. Similar remarks apply toanyof our determinate causal judgments. We would never revise our robust causal judgments just to make them conform with some attractive theory of causation. This is because, in ordinary contexts, we do not care whether our causal judgments align with any theory. Again, it is worth contrasting this case with the case of our water judgments.
Of course, philosophers sometimes make provisions for cases where ordinary judg- ments conflict with their preferred theory. For example, Dowe (2000, ch. 6) claims that, while judgments like V ≡ ‘John’s failure to take the medicine caused his illness’ do not express genuine cases of causation, it is still correct to make such judgments since they are
pragmatically useful. I agree with Dowe that sentences likeV are often correctly assertible, but there are better and worse explanations ofwhythey are correctly assertible. On Dowe’s explanation,V is assertible because it paraphrases sentences that describe “real” causation. But a better explanation is that the term ‘cause’ doesn’t express a relation with a unified nature in the way that Dowe supposes.
When arguing that our intuitive causal judgments are defeasible, Paul & Hall (2013, 41) claim: “[when considering intuitions about cases], you should reflect on whether in- tuition has been set up as an arbiter of questions it may not be competent to judge.” But the above thought experiment suggests the opposite result: at least in the case of causation, metaphysical analyses are not competent to judge the truth of our ordinary judgments. A philosophical theory could have authority over our ordinary causal judgments only if we were inclined to defer to that theory in everyday contexts. But we aren’t: we feel no pres- sure to revise our ordinary causal judgments to conform to any theory. So metaphysical analyses cannot “reveal” that any of our ordinary intuitions about causation are misguided. (Of course, we can imagine speakers who areinclined to conform their causal judg- ments to a philosophical theory. For example, we can imagine speakers who, upon ac- cepting Aronson’s theory of causation, no longer judge (in ordinary contexts) that the ice’s melting causes the water to cool. For these speakers, ‘cause’ would express a relation with a unified nature. But this is not howweuse the term ‘cause’.)
Use and reference magnetism
It is plausible that use has at least some role in determining what our linguistic ex- pressions refer to. Some theorists have appealed to this meta-semantic principle in order to criticize theories of causation that depart too radically from ordinary usage.19 The intuitive
objection is that, if a theoryT conflicts with our ordinary causal judgments across a large range of cases,T doesn’t deserve to be called a theory ofcausation(as opposed to a theory
of something else).
While I find this style of argument convincing, the above thought experiment suggests that it does not go far enough. It isn’t just that use makessomecontribution to determining the relation expressed by the term ‘cause’; the thought experiment suggests that use makes the entirecontribution.20 So it is not enough to reject theories of causation that substan-
tiallyconflict with our intuitions. We must reject theories that conflict withanyour robust intuitive judgments.
Proponents of reference magnetism will resist this conclusion.21 According to refer- ence magnetism, certain types of properties are more natural than others, and this natural- ness helps determine the extensions of our linguistic expressions. If reference magnetism applies for the term ‘cause’, then this term will express whatever natural relation best fits our use. So causation may have a unified nature even if our causal judgments themselves do not line up with any natural relation.
My response is that a semantic theory is adequate only to the extent that it accords with our linguistic behavior. It is part of our linguistic behavior that, when we learn that a water-like sampleX has a different chemical structure than other samples, we deny thatX is water. Insofar as reference magnetism explains this result, it is a plausible theory for the term ‘water’. But the situation is different when we consider terms like ‘game’ and ‘cause’. As explained earlier: in these cases, it is not part of our linguistic behavior to revise our judgments so that they conform to some unified nature. So even if there is some unified nature in the vicinity of our judgments about games and causes, reference magnetism isn’t a plausible semantic theory for terms like ‘game’ and ’cause’.
20Of course, this does not mean that all of our causal judgments are correct. For example, we might falsely
judge thatxcausesybecause of ignorance about the basic empirical facts of a case. See section 2.2.5 for further discussion.
2.4.2 Epistemic rigidity
In the literature, it is not uncommon to use nomologically impossible cases as coun- terexamples to theories of causation. For example, Earman’s (1976, 24) counterexample to Aronson’s transference theory is a possible world where collisions do not conserve en- ergy. Similarly, Schaffer’s (2000) counterexample to counterfactual theories of causation involves a “wizard world” where magical laws govern the casting of spells. In offering these cases, Earman and Schaffer implicitly assume that causation is the same kind of thing across the space of epistemic possibilities.
Other philosophers deny this assumption. For example, Aronson (1982) claims that his transference theory “is intended to make sense of how causation takes place in this world, ... not in some alien universe where the laws of physics do not in the least resemble ours” (302). Paul & Hall (2013) also claim that the nature of causation might differ from world to world (41-42).
The above difference amounts to a disagreement over whether the term ‘cause’ is epistemically rigid: Earman and Schaffer endorse the rigid view, while Aronson, Paul, and Hall endorse the non-rigid view. I’ll now present an objection to the non-rigid view.
On the non-rigid view, ‘causation’ is supposed to be like ‘water’: just as ‘water’ would refer to something different if the water-like stuff in our actual environment turned out to be XYZ, so too ‘causation’ would refer to something different if the actual world were Schaffer’s wizard world. But this proposed analogy breaks when we compare the following cases:
(A) Suppose that the water-like stuff in our local environment turns out to be