El currículo de educación para la ciudadanía en Inglaterra

Maximization requires an agent to choose an outcome to which no other is strictly preferred. Schwartz’s rule for intransitive preferences requires that an agent be similarly guided by her preferences in action. For instance, in a binary choice between two outcomes, where one of the outcomes is strictly preferred, both rules require the agent to choose the most preferred one. As we have seen, such a choice behaviour is required for the money pump arguments to get off the ground.

Does the characterization of preference we just gave provide a desire-based justification for a require- ment to be guided by one’s preferences in either of these two senses, at least when preferences correctly capture the agent’s desires? Letting one’s preferences guide one’s actions seems to be rationally de- manded under this conception of preference only if there is only one preference relation over outcomes that is admissible given the agent’s desires.

outcomes into preferences over full outcomes, one preference relation that correctly summarizes her desires. For each pair of outcomes, the agent’s desires over features of those outcomes completely determine whether the agent should prefer one over the other or be indifferent. Let me call this condition uniqueness.4 _{If uniqueness holds, and the agent’s preferences correctly capture her desires, it seems like,}

if the agent fails to be guided by her preferences in action, she is not doing well by her desires.

Now suppose uniqueness fails: Several different preference relations would equally well express the agent’s desires over features of outcomes. The agent only adopts one of these correct preference relations.5

Now it seems like desire-based instrumental rationality does not demand that the agent is guided by her preferences. If she chooses outcomes in accordance with a preference relation that she does not have, but that would have been admissible given her desires, then she does not seem to be instrumentally criticizable according to desire-based instrumental rationality.

I here want to argue that if we want to give an agent with cyclical preferences reason to have acyclical preferences instead, then we cannot assume uniqueness. And thus, if there is to be any hope of justifying acyclicity with a money pump argument, we cannot make the straightforward argument for maximization, or preference-guidance more generally that we mentioned in the last section.

In order for there to be any hope of giving an instrumental reason for an agent to have acyclical preferences, given the notion of preference we are working with here, there must be an acyclical preference relation that actually expresses the agent’s desires correctly. I will grant this here.6 _{But now the question}

of uniqueness arises: Is this the one unique preference relation over outcomes that expresses the agent’s desires correctly? I want to argue that this can’t be so if we want to deliver a money pump argument.

If we do assume that uniqueness holds with regard to an acyclical preference relation, then the original cyclical preferences must have been a mistaken expression of the agent’s desires. This is already a worrying implication, since it does not do justice to the intuitive plausibility of those cyclical preferences. But perhaps the money pump argument could then be understood as an argument that preferences that would express the agent’s desires well would need to be acyclical. The reasoning could be that if the agent’s preferences capture her desires fully, then actions guided by those preferences should not frustrate those very desires.

However, this argument does not work, since it takes for granted that agents should always be guided by their preferences in action. But we cannot do so in this case. If there is one unique acyclical preference relation that expresses the agent’s desires well, we may be able to justify maximization with respect to

4_{There is a parallel debate in formal epistemology about whether there is a uniquely rational credence an agent may}

have given her total body of evidence. See, for instance, White (2005) and Titelbaum and Kopec (2016).

5_{Another possibility would be that in these kinds of cases, an agent should adopt a family of preference relations,}

expressing an ‘imprecise preference’. Imprecise credence, too, is often represented by families of probability functions. See Bradley (2015). However, in that case, we would already have given up on maximization, since we would have to formulate new choice rules for families of preference relations. How we could then give an instrumental justification for each preference relation in the family being transitive is unclear to me. Most plausibly, a choice rule for ‘imprecise preference’ would be fairly permissive, and allow the agent to act in a way that is licensed by at least one of the permissible preference relations (that is, by maximizing with regard to the preference relation, or by following Schwartz’s rule with regard to that preference relation). Principle Q would justify us saying that the agent should choose such that she is not money pumped, insofar as this is compatible with the permissive choice rule. But clearly, to conform to principle Q and such a permissive choice rule, the agent’s individual preference relations need not each be transitive.

6_{There is some reason to be sceptical of this claim, especially in problem cases like the Self-Torturer Problem. It is}

not clear that in that case any acyclical preference order could express the agent’s desires correctly. Such a preference relation would have to include a preference not to go up any further setting at some point if it is to keep the agent from self-torturing. But such a preference seems like a poor representation of the agent’s desires to have money and be pain-free, given that the difference in pain between adjacent settings is not distinguishable to the agent when facing a pairwise comparison. However, I will conclude below that the better notion of preference to try and salvage the money pump argument is one that is closer to action anyway. And so while granting the assumption that some acyclical preference relation may represent the agent’s desires correctly here, I am not committed to this claim.

those preferences. But the argument we just made assumes that the agent would also be guided by her preferences in action if she had cyclical preferences instead. And we cannot take that for granted. In particular, as long as we grant that some acyclical preferences would represent the agent’s desires well, desire-based instrumental rationality would permit the agent to act in accordance with those hypothetical preferences, thereby avoiding being money pumped, while keeping the cyclical preferences.

Perhaps in some cases, we have independent reason for thinking that cyclical preferences misrepresent an agent’s true desires. But given the intuitive plausibility of cyclical preferences in the problem cases we are dealing with here, we do not generally have such independent reason for thinking that cyclical preferences misrepresent the agent’s desires. And in any case, we would then not be in need of money pump arguments. But it is widely agreed that money pump arguments are the best instrumentalist arguments in favour of acyclicity.

Hence, in order to be able to deliver a money pump argument in favour of acyclicity, we have to assume non-uniqueness at least in the sense that the original cyclical, as well as at least one acyclical preference relation would represent the agent’s desires correctly. But this is also enough by way of non-uniqueness to show that money pump arguments fail according to the notion of preference we are dealing with, as the next section will argue.