La educación para la ciudadanía en Inglaterra

I have argued that it is instrumentally irrational to be money pumped according to desire-based instru- mental rationality, at least as long as it is possible to avoid being money pumped without taking actions that are themselves suspect in terms of the agent’s desires at the time of action. This will only give an agent conclusive reason to have acyclical preferences if having acyclical preferences is the only way of avoiding being money pumped.

However, it is frequently argued that having foresight can save an agent with cyclical preferences from being money pumped without her having to give up her cyclical preferences. Here I consider whether foresight can really save the agent from being money pumped. The short answer is that it can only do so if the agent violates maximization at some point. However, as we will see in the next section, such a violation need not be in conflict with desire-based instrumental rationality.

Schick (1986) argues that an agent with foresight will stop trading early in the series of trades offered to her in money pump scenarios. Let us consider his argument in the context of the example of my apartment choice introduced in the last chapter. First, we need to represent the series of choices I am

offered by the rental agency as a dynamic choice problem. As we saw in Chapter 2, such problems are usually represented in decision trees like the one in Figure 4.1. At the square nodes, an agent can decide whether to go up or down, starting at the left-most node. We assume here that the agent is not offered further trades once she has refused one.

Apartment A − Apartment C Apartmen t C Apartment B Apartmen t B Apartment A

Figure 4.1: Dynamic Money Pump Problem

Schick argues that in a decision problem like this one, agents should decide using a process of backward induction. They should consider how they will choose in the last choice, assuming they will be guided by their preferences over the outcomes available then. They then take their prediction as given when considering the second to last choice, etc. As we noted in Chapter 2, this approach to dynamic choice has come to be known as ‘sophisticated choice’. We there characterized it as continued application of maximization, coupled with the belief that one will maximize in the future.

Now that the agent has cyclical preferences, we cannot assume she follows maximization. But as we saw in the last chapter, we can still make sense of the idea of preference-guidance in the context of intransitive preferences, for instance by appealing to Schwartz’s rule. In the binary choice the agent faces here, this rule, like maximization, implies that the agent should choose the option with the strictly preferred associated outcome. We can thus still make sense of sophistication when the agent’s prefer- ences are cyclical: It can be characterized as continued application of some rule to be guided by one’s preferences in action.

A sophisticated agent facing the series of trades in our example will indeed not be money pumped, but instead end up with Apartment B. She will predict that, if she got to make the last choice, she would choose to pay $25 to get Apartment A back. To predict what she would do at the second to last choice, we thus consult the agent’s preferences between Apartment B and Apartment A −. Since Apartment B is preferred to Apartment A, it is also preferred to Apartment A having payed $25. And so the agent would choose to stick with Apartment B in the second to last choice. Knowing that, she would choose Apartment B in the first choice.

However, Rabinowicz (2000, 2001) shows that sophistication will not save an agent from being money pumped in all situations. Appeal to sophistication alone thus does not decisively speak against money pump arguments in favour of acyclicity. In particular, sophistication may not save an agent from being money pumped in cases where she faces an opponent that is persistent.

To modify Rabinowicz’s (2000) example slightly, and adapt it to our purposes, suppose that the rental agency offers me a trade exactly three times, even if I have refused the first or second trade. I start with Apartment A. They offer me Apartment B. If I accept the offer, they offer me Apartment C. If I accept that offer, they offer me Apartment A - . If I have refused a trade, they simply offer me the same trade again. The dynamic choice problem I now face is represented in Figure 4.2.

Apartment A -  Apartment C Apartmen t C Apartment C Apartment B Apartmen t B Apartmen t B Apartment C Apartment B Apartmen t B Apartment B Apartment A Apartmen t A Apartmen t A

Figure 4.2: Modified Dynamic Money Pump Problem

In this modified money pump, even a sophisticated agent will be money pumped. That is, she will end up with Apartment A -  even though she could have had Apartment A. To see that, note that at each of the last decision nodes, the sophisticated agent will choose to trade. Given that, she will also trade at the second to last choice node. The choice the sophisticated agent thus effectively faces at the first choice node is between Apartment C (if she refuses to trade now), and Apartment A -  (if she trades straight away).

At the first choice node, the sophisticated agent knows she will trade in the future, and thus knows that Apartments A and B are inaccessible to her. Given a choice between Apartment C and Apartment A - , she chooses the latter. Thus, we have shown that sophistication does not save the agent from being money pumped. We only need some persistence on the side of our rental agency. Does this mean that foresight in general does not help?

It has been proposed that different ways of deciding in dynamic choice problems may help an agent with cyclical preferences to never be money pumped. One such choice strategy is what McClennen (1990) calls ‘resolute’ choice, as we already encountered it in Chapter 2. Another is what Rabinowicz (2014) refers to as ‘unified choice’. Resolute agents choose a sequence of actions in the beginning of the dynamic choice problem, in accordance with their preferences then, and then simply go through with that sequence of actions. Unified agents decide at each point in a sequence of choices as they would were they to then decide on the whole sequence in one single choice. Given stable preferences over outcomes, this should lead to the same course of action. Since this stability will be presupposed in all dynamic

choice problems from now on,1Rabinowicz’ unified choice should be understood as included whenever I mention resolute choice.

In our case, a resolute agent would choose as she would in the ‘synchronic’ money pump. Rabinowicz (2014) claims that the agent would not choose to be money pumped in this case:

Were she to make a single choice, [...] we may safely assume, she would not choose to accept all the three exchanges. A simple calculation would show that refusing all of them would save her the extra costs and still result in the same outcome. (p. 373)

Note here that this solution seems to rely on abandoning preference-based instrumental rationality. As we have seen, as far as preference-based instrumental rationality is concerned, we cannot say that the agent is rationally required to not choose A - . But money pumps only have an intuitive hold on us in the first place because we think it is in fact instrumentally irrational to end up paying for something you could have had for free (especially in a one-off choice). We can appeal to principle Q (Section 3.7) to explain why a resolute agent should not select a course of action that would leave her money pumped. And so neither money pump arguments, nor the most common response to them work without abandoning preference-based instrumental rationality.

Resolute, or unified choice, combined with something like principle Q, keeps an agent with cyclical preferences from being money pumped. Does this mean that the money pump arguments have been refuted? The central controversy here has been that it is not clear whether resolute choice is itself compatible with instrumental rationality. Moreover, if it is not, then it is not clear whether our principle Q really condemns being money pumped here as irrational.

This worry is usually expressed in terms of preference-based instrumental rationality, and thus, I think, misses the point. Sophisticated agents make a prediction of their own future behaviour, assuming they will be guided by their preferences over the outcomes still available at each choice point in the future. Given this prediction, they choose the action that is favoured by their preferences over the outcomes that are still available to them. Sophisticated agents thus merely seem to act continually in accordance with their preferences. If we think that being guided by one’s preferences, by following maximization or Schwartz’s rule, is a principle of instrumental rationality, then instrumental rationality seems to demand that an agent be sophisticated.

By contrast, on one natural way of understanding resolute choice, a resolute agent will sometimes choose against her binary preferences over the outcomes still available to her at the time of action. Namely, this is so if the resolute agent’s preferences throughout remain the same as they would be outside of the dynamic choice problem. We already saw in Chapter 2 that resolution understood in this way may require counter-preferential choice in temptation cases. In the present case, the agent will act against her preferences at some point whenever resolute and sophisticated choice come apart.

For instance, in the modified money pump we just presented, resolution demands refusing at least one trade, since the resolute agent would not choose to end up with A - . If the resolute strategy is to refuse the last one, then the agent acts against her preferences over the available outcomes then. If it is to refuse the second to last one, while taking the last one, again, taking into account a correct prediction of one’s future action, the agent then acts against her preferences over the available outcomes. The same

1_{That is, stability of the agent’s ‘given’ preferences will be presupposed. Since on one way of understanding resolution,}

resolution involves making temporary adjustments to one’s preferences, resolution may itself introduce instability in the agent’s ultimate preferences. What we will call ‘given’ preferences are the preferences the agent has outside of specific dynamic choice problems, or before preferences have been adjusted as part of a resolute choice strategy.

holds if the first trade is the one to be refused. Thus, resolution sometimes demands acting against one’s preferences in money pump scenarios.

On this understanding of resolution, the only way that foresight can keep an agent from being money pumped is thus by sometimes acting counter-preferentially. We have said that maximization, or at least preference-guidance in the form of Schwartz’s rule seem to be easily justifiable under preference-based instrumental rationality. As we saw in Chapter 2, two-tier arguments in the face of temptation that try to argue otherwise fail. Being guided by one’s preferences in action seems to guarantee that one does well in terms of one’s preferences. And so, if we assume preference-based instrumental rationality, resolution does not speak against money pumps. This is because according to preference-based instrumental rationality, resolution is itself incompatible with instrumental rationality.

We come to a similar conclusion on another reading of resolute choice. McClennen in fact suggests that resolute agents adjust their preferences within dynamic choice problems, so that being guided by one’s preferences leads one to the resolute choice after all. If, for instance, the resolute choice would require us to end up with Apartment C, the agent would adopt the preference C  A −  at the last choice node in the dynamic choice problem. If agents adjust their preferences as part of a resolute choice strategy, resolute agents in fact end up maximizing at each point in time with regard to those altered preferences. What then still distinguishes it from sophistication? To preserve what I think is a meaningful difference, we have to slightly amend our characterization of sophistication: Sophistication in fact consists in the continued application of some preference-guidance norm like maximization to the agent’s ‘given’ preferences. I will take the agent’s ‘given’ preferences to be those preferences the agent would have outside of a particular dynamic choice problem, or before adjustments to preferences have been made as part of a resolute choice strategy.2

One could be worried that one cannot have instrumental reasons to make changes to one’s preferences like this. And then being resolute would not be a strategy for avoiding being money pumped that one has instrumental reasons to adopt, but just a circumstance some agents are in, and others are not. Again, preference-based instrumental rationality supports these worries. If preferences themselves are the basic standard of instrumental rationality, being resolute in this alternative sense requires the agent to change what she ultimately cares about. And then it seems like any reason to be resolute would be a non-instrumental reason. One caveat here is that having different preferences can serve an agent’s preferences as they are in cases of ‘autonomous benefit’. However, I argued in Section 3.4 that money pumps cannot be construed as such cases of autonomous benefit under the assumption of preference-based instrumental rationality.

Hence, for both interpretations of resoluteness, standard worries rely on preference-based instrumental rationality. But we have argued in the last chapter that preference-based instrumental rationality makes it impossible for us to make money pump arguments anyway. And so if there is any hope for money pump arguments in favour of acyclicity, we cannot rely on preference-based instrumental rationality in our response to the challenge resolute choice poses. We proposed a desire-based notion of instrumental rationality in the last chapter instead. What defenders of resolute choice need to show is that agents may have instrumental reasons to be resolute according to this desire-based notion of instrumental rationality. The more general challenge that resolute choice poses to money pump arguments is that it may be possible to avoid being money pumped while generally keeping one’s cyclical preferences. One could

2_{McClennen (1990) expresses the same idea by requiring that sophisticated agents abide by what he calls ‘separability’}

either do this by acting counter-preferentially at specific points in a decision problem, or by adjusting one’s preferences temporarily within dynamic choice contexts only. The two different interpretations of resolution (or unified choice) we presented here adopt these two alternative strategies respectively. But being resolute may not be the only way of exploiting these strategies successfully in response to money pumps. What I want to investigate now is in how far desire-based instrumental rationality allows for such alternative responses to money pump arguments.