This very valuable book addresses this sort of problem: how much weight should present you give to the values and tastes of past and future you? Should you now in your twenties commit to the artistic career that you’ve always wanted, when you expect that in your forties you’ll wish you’d studied to be a doctor instead? Should you now go on that amusement park ride knowing that you will be regretting it quite a lot in about five minutes?

Pettigrew concedes (17–18) that whether this even looks like a problem will depend on what ‘values’ and ‘tastes’ are meant to be. On a behaviouristic or ‘revealed preference’ view your present choices settle your present values: if you choose x when y is available then you do presently value x over y. But the present values that they settle include your present values regarding your future values. If you take the amusement park ride you are thereby revealing that you are not much concerned about your feelings in five minutes. So your choices reveal that you already value future goods, like the satisfaction of future desires, at some rate; they therefore already encode an answer to Pettigrew’s question. But I don’t agree with his concession that this would eliminate the problem. You can agree that your present values give your future values such and such weights. But you can still ask whether those weights are right.

Pettigrew’s own ‘realist’ approach identifies values with tastes that are not always revealed by choice but which do typically inform it. Formally he distinguishes psychological parameters Cr and u i.e. credences and utilities – degrees of belief and of desire – from a higher-level parameter V, the evaluation function, which choice does reveal (15). So the question is: given the agent’s past, present and future credences and utilities, what should her present evaluation function be? How should her choice between present options be sensitive to her past and future as well as her present credences and utilities?

Pettigrew treats this as a problem of judgment aggregation. Here we have a lot of selves, past, present and future, each with their own credence and utility functions. What we want is an appropriate way to combine these inputs to give an evaluation function: one that selects from among present options in a way that gives due respect to your past and future as well as to your present attitudes towards their consequences.

His solution is that your choice at a time should maximize the expectation of a weighted sum of your past, present and future utilities, the expectation being calculated with respect to your present credences. Simplifying slightly, the figure of merit for any option a is then given by:

(1) $$V\left( a \right) = \sum\nolimits_Z {Cr} \left( {z{\rm{|}}a} \right)\sum\nolimits_t {{\alpha _t}{u_t}\left( {az} \right)} $$

In (1), t ranges over past, present and future times, z ranges over possible outcomes of your choice and Cr is your present credence. Each α_t represents the weight attaching to the utility function u_t of your time-t self. There is some simplification in (1) because the weights and utility functions may vary not only with times but with the state of the world at those times: if you drink tonight then tomorrow you might have an aversion to loud noises or bright lights that you could tolerate perfectly well when not hungover. Pettigrew’s version takes account of this fact (cf. 230).

The book is structured around this solution: after the introductory Chapters 1 and 2, Chapters 3–5 argue against three alternative approaches, Chapters 6–7 set out the main proposal, and Chapters 8–10 address various questions about it and objections to it: these cover amongst other things the issue of interpersonal and intertemporal utility comparisons, the question of what is so good about linear pooling (e.g. vs geometric pooling) and challenges from Sarah Moss and Laurie Paul. Chapters 11–15 discuss general principles for setting the weights α_t : for instance whether it is reasonable to care less about your distant future self than about your near future self; or whether you ever may or ever should give some weight to the utilities of your past selves. (Pettigrew’s answer in all three cases is yes.) In this review I’ll discuss two of these points: Pettigrew’s solution to the problem of intertemporal utility comparison and his arguments concerning what weight to give one’s past utilities.

Suppose you want a utility function that aggregates the values of two individuals – me-now and you-now, or you-now and you-tomorrow – by giving (say) equal weight to both. The trouble is that if u represents an individual’s values then so does any positive affine transformation of it i.e. any $$au + b\,\left( {a \gt 0} \right)$$ . So if ${u_1}$ and ${u_2}$ represent these individuals’ values then there is no reason to think that it is $${u_1} + {u_2}$$ rather than (say) $${u_1} + 2{u_2}$$ that is giving them ‘equal weight’. It’s not even clear what it could mean to give them equal or any other weights. Trying to aggregate two value profiles in an equally weighted way is like trying to aggregate (say) the temperature of an object and its price into a single measure that gives them equal weight.

Pettigrew’s solution (102–103) is that we can sometimes settle, without first choosing any particular utility functions, that the difference between one individual’s values for each of two outcomes is the same as the difference between that individual’s and the other individual’s value for one of those outcomes. For instance, I may be able to settle that the difference for me now between e.g. a camping holiday next year and a beach holiday then is the same as the difference between a beach holiday next year for me now and a beach holiday next year for me-in-a-week (after I have seen Jaws). As Pettigrew notes, if there are two such scale-independent facts then the problem is solved, for if we are given a utility function ${u_1}$ representing present me then there is exactly one utility function ${u_2}$ that both represents future me and gets both intervals to match. For instance, if it is a constraint on utility measures that: (i) present me’s utility for the beach holiday is exactly halfway between present me’s utility for the camping holiday and future me’s utility for the beach holiday; (ii) present me’s utility for a holiday in Skegness is exactly halfway between present me’s utility for a holiday in Blackpool and future me’s utility for a holiday in Skegness, then we can pin down what it means to give equal weight to the utility functions of present me and future me.

This is not reassuring. Of course Pettigrew is right that facts like (i) and (ii) would make intertemporal utility comparisons meaningful. The problem is finding these facts. His ‘matching intervals’ solution presupposes that there are facts about whether (a) an interval bounded by present me’s utility for two things matches (b) the interval between present me and future me’s utility for one of them. But whether (b) can match anything is what is at issue: anyone who worried about making sense of intertemporal utility comparisons would not feel any better to be told that everything would be all right if only we could make sense of this intertemporal utility comparison. It would be like saying that we could meaningfully aggregate price and temperature, if only we could find a cup of tea and a cup of coffee such that the difference in temperature between the tea and the coffee was equal to the difference between the temperature and the price of the coffee.

Pettigrew is more optimistic on this point. He writes:

Similarly then, you might think that my future self has enough insight to tell by how much more or less it now values the beach holiday than any of its past selves did, in particular whether this increment is more or less than the difference in value that it once perceived between the beach holiday and the camping holiday.

Maybe Pettigrew’s insight into his own past and future selves is more acute than mine. I personally can’t remember much about my past utilities for anything, not if you go back more than a week or two; and what I can remember is more reconstruction than insight. I know I used to like ice cream a lot more than coffee, but not any longer: but I know this because I remember that I used to order ice cream at restaurants when nowadays I often order coffee. But comparing past and present choice behaviour is not enough to settle whether (say) my present value for coffee lies half-way between, or a third of the way between, my past value for ice cream and my past value for coffee. And introspection is no help either: when I introspect it is ‘as if a snapshot of a scene had been taken, but only a few scattered details of it were to be seen: here a hand, there a bit of a face, or a hat – the rest is dark’ (Wittgenstein Reference Wittgenstein2009: §635). Now move from comparisons across relatively short times involving relatively trivial options, to long-term shifts in fundamental values – cases like Parfit’s Russian nobleman (212–213). Think of all the opportunities for time to obscure such things and for self-interest to distort them. In these cases it would be a wonder if I ever got a grip on ordinal changes in my utilities, let alone a precise idea of the size of these variations – if indeed (as I continue to doubt) such quantities exist at all.

Still, the most interesting parts of Pettigrew’s discussion survive this kind of scepticism. For instance, Chapter 12 discusses whether present you ought to give any weight to the values of past you. Even if we can’t make sense of the difference between giving (say) equal weight to your past and present values as opposed to giving (say) twice as much weight to the former as to the latter, we can still make sense of the difference between giving no weight to the values of your past self and giving some weight to the values of your past self. Pettigrew argues that you are obliged to give some weight to the values of your past self.

In the example that is meant to motivate this (167), Roshni is trying to decide whether to attend university in the knowledge that her grandmother, who has just died, (a) made heavy and permanent sacrifices throughout her life so as to give Roshni the best possible education and (b) always wanted Roshni to attend university. Pettigrew argues (plausibly enough) that it is permissible for Roshni to take account of her grandmother’s wishes. He then argues that given certain further assumptions, for instance that Roshni’s having this opportunity meets the norms of justice, it is mandatory for Roshni to take account of her grandmother’s wishes. He then argues that the analogy between your obligations to other people who no longer exist and to versions of your own self that no longer exist means that in similar cases it is mandatory to take account of your own past values. In short, it is mandatory to take account of your past self’s wish that you now do X, if your past self has made sacrifices in order that you now have (and justly have) the opportunity to do X. This obligation need not always be decisive, but there are cases (perhaps like Roshni’s) where it may tip the scales.

Roshni (and her counterparts in the intra-personal analogy, like Blandine; 3–4) may indeed have obligations but I am not sure that they are the ones Pettigrew is aiming to defend. From the realist perspective that we are assuming there are two levels at which something conative or desire-like enters psychological description. There is the utility function u, which describes your ‘inner’ values and tastes but is not necessarily revealed in behaviour; and there is the evaluation function V, which combines the utility functions u of various selves with Cr, to yield a ranking of options. When we say that Roshni is obliged to take account of her grandmother’s wishes, or that present you is obliged to take account of the wishes of your younger self, do we mean they must feed directly into present you or Roshni’s V or do we mean that they ought to influence your or her inner values as reflected in u?

It is natural to interpret the example as illustrating the latter. It doesn’t show that Roshni ought to act on her grandmother’s wishes whether she cares about them or not. It shows that she ought to care about them whether she acts on them or not. If Roshni knows about the grandmother’s sacrifice, is completely indifferent about it, has a slightly higher utility for not going to university, and does not go, then what has gone wrong would not have been made all right if she had in fact gone whilst still not giving a fig about the grandmother. On the contrary she would still have been somewhat monstrous – a Kantian monster – not for her choices but for her indifference.

But then it is no longer clear that the example contributes to Pettigrew’s project. That project, recall, was to say what weights your past, present and future utilities and credences should have in determining your present choices. But if we are saying that Roshni ought to go to university because her grandmother’s wishes ought to matter to her, then in the intra-personal analogue of the case we are saying that you ought to let your past utilities determine your present acts because you ought to let them determine your present utilities. But this is consistent with giving the past zero weight in the target sense: that is, it is consistent with an evaluation function of the form (1) such that ${\alpha _t} = 0$ for any t lying in the past, just as long as your present u is responsive to at least some such u_t . Conversely, if we are really asking – as Pettigrew claims – about how we ought to set the weights α_t then Pettigrew’s insightful discussions of Roshni’s and similar cases may not be relevant.

These considerations also bring into relief the strangeness of any non-trivial instance of (1): specifically any view on which ${\alpha _t} \gt 0$ for some t other than the present moment. Towards the end of the book, Pettigrew writes:

There are two ways that we can think of a CEO: as a trustee of the shareholders or as a delegate. On the ‘trustee’ model she uses her own judgment and acts on her own preferences, even when these run contrary to those of the corporation’s shareholders. On the ‘delegate’ model she is simply their mouthpiece, in something of the manner of state electors in US Presidential elections.

It is entirely natural to think of your present self as the trustee of your past and future selves: you have to use your own present judgements and preferences to make choices that will affect the welfare of these other selves; and how it affects them will depend on the extent to which your present utility function takes account of theirs. A good CEO of course ought to take some account of shareholders’ wishes. But the analogue of that wholly natural thought is not a model in which ${\alpha _t} \gt 0$ for some t other than the present moment, but rather one in which your present u is a function of the corresponding u_t s. By contrast, a non-trivial instance of (1) models a person whose present self is more like a delegate: a self whose present concerns, for the past, future and present, are not entirely in control of his present choices but rather in competition with his past concerns, a competition that his present concerns must inevitably sometimes lose. Talk about back-seat driving!

My reservations about these aspects of Pettigrew’s project are not meant to leave the reader with a negative view of the book. On the contrary, they were meant to make clear that it is grappling hard with difficult and important questions about the conduct of our lives as temporally extended beings with changing tastes. It is also full of gems that I have not mentioned: the argument for linear pooling in Chapter 9, for instance, or the argument in Chapter 13 (which I hope to discuss at length elsewhere) that exploitability does not entail irrationality and so cannot be the basis of an objection to hyperbolic discounting. All in all it is a terrific contribution to the project of using formal methods to help each of us analyse, clarify and ultimately address the foreseeable and inevitable mutability of taste.