Naive Characterisational Representationism is deeply flawed, for reasons pointed out by Lyle Zynda (2000) It is easy to show that whenever S can be para
3.3 Problems with Classical Characterisational Representationism Naive, Extreme, and Classical Characterisational Representationism have been the
main focus of criticisms against characterisational representationism. In a recent critical review, Meacham and Weisberg (2011) present a number o f arguments against five variations on the basic theme o f characterisational representationism,
each o f which entail the classical version— and the majority o f problems that they raise result from this entailment (and are summarised below). The same can be said more generally: Objections to characterisational representationism often come in the form of arguments against either Naive, Extreme, or Classical Characterisa tional Representationism. See, for example, the critical arguments o f Hampton
(1994), Christensen (2001), Häjek and Eriksson (2007), Häjek (2008), and Easwa- ran (2014).
In the rest o f this section, I will discuss the main concerns which have been raised in the literature (along with some additional concerns o f my own). These I have divided into two classes: Those which arise from the very strong connections that these views posit between preferences, credences, and utilities (§3.3.1), and those which arise from the use o f CEU theorems in particular (§3.3.2).
3.3.1 The connection with preferences
There are, I think, good reasons to reject CCR and any view which entails it. Per haps the most common concern is that CCR suggests (and is often held by those with) an anti-realist stance towards the graded propositional attitudes. That is, CCR implies that having certain preferences is sufficient for having such-and-such cre dences and utilities, as though being in those latter kinds o f state were nothing over and above being in a particular kind o f preference state. In Joyce’s words, CCR can “foster a kind o f pragmatism that sees belief [or credence] as a second-class propo sitional attitude that can only be understood in terms o f its relationship to desire [or preference]” (1999, 89).24 The worry is even more apparent with Extreme Charac- terisational Representation ism, or with the suggested approximation-based exten sion o f CCR outlined above— according to either view, to have credences and util ities at all ju st is to have a particular pattern o f preferences.
Worse still, where a theorem’s > is understood behaviouristically (as it’s usually intended to be), these positions suggest an outdated form o f behaviourism— that there is nothing more to having credences and utilities than behaving (or being dis posed to behave) in a particular kind o f way. Such a position is contrary to our shared, pre-theoretic conception of these things, where our credences and basic util ities for outcomes are understood as each playing a part in the causal explanation o f our choices. On intuitive grounds, this strongly suggests that credences, utilities, and preferences (whether understood mentalistically or behaviourally) should be kept conceptually and metaphysically separated (cf. Joyce 1999, 21-2).
24 Note that how Joyce officially defines ‘pragmaUsm’ makes it the meta-normative claim that epistemic norms are grounded in practical norms (the “laws of desire”). This kind of view perhaps makes the most sense under an anti-realist or behaviouristic construal of credences in terms of be havioural preferences, but it need not be committed to those construals.
There is, I think, another problem here, and one which goes beyond a simple knee-jerk reaction to the anti-realism or behaviourism suggested by CCR. The (nec essary) preference conditions o f a CEU theorem are most plausibly read as norma tive constraints on preferences. Descriptively, however, ordinary agents frequently fall foul o f basic norms o f rationality (whether for systematic or non-systematic reasons), and this creates problems for CCR. Importantly, it’s immensely plausible that ordinary agents will sometimes fa il to have preferences that maximise their expected utility, given their credences and utilities. This means, for one thing, that an agent might have probabilistically coherent credences “Bel and utilities Ves but
not have expected utility maximising preferences. More importantly, it means that an agent might, due to some irrational state o f mind, have preferences which could be paramorphically modelled as maximising expected utility relative to some Bel- Ves pair despite not having credences ‘Bel or utilities Ves. In effect, CCR allows for no wriggle room between preferences on the one hand and credences and utili ties on the other, in the event that those preferences satisfy the relevant theorem’s conditions. It implies that it’s impossible for anyone to be preference-rational by accident—that, whenever someone’s preferences conform to the conditions, it must
be because they were acting rationally given their credences and utilities. CCR im plies that irrational agents cannot satisfy a CEU theorem’s preference conditions, and this seems utterly unmotivated.
There are some obvious changes that one could make to CCR to loosen the con nection it posits between preferences on the one hand, and credences and utilities on the other. For reasons that I will return to shortly, I doubt that these will be quite enough, but it’s worth highlighting them briefly first. To begin with, we might con trast CCR with the following account, inspired by Lewis (1980a):
S has credences Bel and utilities Ves iff S is in some psychophysical state R*, where
R* would tend to cause a typical subject S' to be preference-rational such that she would be modelled, by *F, as an expected utility maximiser with credences Bel and utilities Ves
This kind of view would require that credence states are identifiable independently o f their functional role— i.e., as a particular neurobiological kind. Like Lewis, one
might cash out the tends to cause relation by reference to the causal role that R*
would play in a typical member o f some pre-specified population. There are, how ever, other ways to flesh out the relation, which we need not consider here; the important point for our purposes is that it’s not CCR: the fact that an underlying psychophysical state R* tends to cause preference-rationality does not mean that whenever the agent is preference-rational, they are therefore in R*. Perhaps CCR holds much o f the time, or holds for a perfectly typical subject, but it need not hold in general.
The second way in which CCR might be avoided would be to ignore actual pref erences and instead characterise an agent’s credences and utilities in terms o f what preferences she would have in some idealised state. We have noted that the ordinary subject will often make mistakes, in one way or another failing to have the prag matically optimal preferences given her credences and utilities. However, perhaps under some idealised state o f considered reflection, every agent will conform to decision-theoretic norms:
S has credences “Bel and utilities Ves iff where S in ideal conditions (e.g., she is functioning properly in a normal environment, free from interfering influences such as intoxication, time pressures, and so on), then S would be paramorphically mod elled, by Y, as an expected utility maximiser with credences Bel and utilities Ves
I suspect that something like this is probably true (see §8.5), but as an account o f the nature o f credences it still seems to be missing something. While it may be plausible that utilities straightforwardly reduce to particular patterns o f preferences (especially where ‘preference’ is given a mentalistic construal), our credences seem to be a wholly distinct and independently existing kind o f mental state— and the above two suggestions do not yet capture everything which is important about them.
Importantly, the credences that we have towards specific propositions seems to depend strongly on the evidence that we have accumulated regarding to those prop ositions. However, there is no accommodation for this connection between cre dences and past evidence in CCR (or any o f the proposed refinements). The worry here is expressed nicely in the following passage by David Christensen (see also Weirich 2004, 20, for similar remarks):
True, degrees of belief are intimately connected with preferences and choice behav iour. But they are also massively and intimately connected with all sorts of other aspects of our psychology (and perhaps even physiology). This being so, the move of settling on just one of those connections—even an important one—-as definitional comes to look highly suspicious. (2001,362)
Building off o f Christensen’s discussion, Meacham and Weisberg make the same complaint:
Given that beliefs have connections to so many mental states besides preference— emotions, perception, memory, and so on—it’s implausible that just one of these connections is paramount. With all the pushes and pulls that beliefs and desires are entangled in, we should not expect there to be a rigid and straightforward connection between degrees of belief, utility, and preference. (2011,646)
Indeed, CCR could have us assign credences Bel to an agent on the basis of her preferences even when Bel is entirely at odds with what we would expect her cre dences to be like given her life history o f evidence. And this result seems unac ceptable.
3.3.2 The use o f CEU theorems
Another frequent cause for concern arises from the use o f CEU theorems in partic ular. The focus on CEU theorems is, I suspect, due largely to the attention philoso phers have given to characterising the credences o f ideally rational agents. The use o f CEU theorems has been the grounds o f two basic criticisms, which I will discuss in turn.
The first criticism is that ordinary agents do not satisfy the preference conditions associated with standard CEU theorems. This complaint plays a prominent role in the critical discussions found in (Hampton 1994), (Meacham and Weisberg 2011), and (Dogramaci forthcoming). Much o f the relevant empirical work is summarised in (Tversky 1975), (Camerer 1995), (Schmidt 2002), and (Johanna, Jeleva et al.
2012). The most widely cited evidence here originates with Allais (1953) and Ells berg (1961). Kahneman and Tversky (1979) outline experimental results which (they argue) imply that ordinary decision makers in the kinds o f decision situations that Allais outlined don’t always adhere to Savage’s sure-thingprinciple, which is a common independence condition found in many CEU theorems (see §5.1.2). The adequacy o f other independence conditions also comes under attack from (Birnbaum and Chavez 1997) and (Birnbaum and Beeghley 1997). Some authors have also purported to show through so-called preference reversal experiments that ordinary agents’ preferences are intransitive (Lichtenstein and Slovic 1971, 1973, Fishbum and LaValle 1988).25
It is perhaps not so worrying if most agents don’t satisfy the CEU conditions exactly, so long as they come close to satisfying those conditions (in which case we might appeal to the preference-rational systems they most closely approximate). One reason to think that ordinary agents’ preference don’t vary greatly— or at least, greatly and systematically— from CEU-consistent preferences is that many predic tive models in economics and the social sciences essentially treat the average deci sion-maker as having the kinds o f preferences associated with expected utility max imisers, or thereabouts,26
Even descriptive decision-theoretic models that are explicitly designed to ac commodate the empirical evidence for our deviations from CEU bear a close re semblance to that theory: with few exceptions, they involve a "Bel function (which is at least a capacity if not a probability function) and a utility function combined in something like expectational form, with the basic decision-making principle be ing that an agent will pick the option which has the highest Be/-weighted average utility. This is essentially the case, for example, of Kahneman and Tversky’s (1992) cumulative prospect theory, which is widely considered to be the most empirically accurate decision model so far developed. (See Appendix B for more details.)
25 The vast majority of the empirical work has focused on whether ordinary agents satisfy the necessary conditions associated with CEU theorems; whether they always satisfy the non-necessary, structural conditions is generally taken to be relatively unimportant. The main reason for this attitude will be discussed in §5.2.4.
26 The general point here goes back at least to Fodor (1987), who argued that folk psychology (which is in many respects very close to orthodox expected utility theory) is presupposed so widely within our best explanations of behaviour that it is likely to be at least broadly correct.