Rationality and Time

Richard Chappell

Do temporal relations have intrinsic rational significance? We may reasonably discount risks and benefits in the distant future due to uncertainty, but are they in addition less important simply in virtue of being more distant? This paper seeks to defend temporal neutrality, or the view that a mere difference in timing can have no rational influence over our preferences. If a difference in timing matters at all, it is only because it is accompanied by other differences which matter. I have argued elsewhere that we have reason to recognize an atemporal constraint on our decision procedures. After briefly summarizing this argument, the core of this paper will address Parfit’s objections to temporal neutrality.

Global Rationality (revisited)

We may initially be tempted to conceive of rationality as a matter of doing what seems best from the perspective of the moment. But this can foreseeably lead agents astray. Suppose an immortal has been gifted a bottle of wine that will improve indefinitely with age.¹  When should she drink it? On any given day, she may reason that it would be better to drink it tomorrow.  If this pairwise comparison is taken to be sufficient reason to postpone drinking the wine then she will never drink it at all.  The guidance of this localized perspective would then predictably yield the worst possible result.  Admittedly, if this risk is incorporated into the agent's initial reasoning then it is no longer so clear that postponement has a higher expected utility even from the local perspective.  But a sure solution is for the reasoning agent to grant normative authority to a more 'global' perspective.  This would have her abstract away from her present temporal location, and resolutely settle (somewhat arbitrarily, if necessary) on an outcome that would be good enough.  Global rationality further requires that – when the fated day finally arrives – she sticks to her resolution and drinks the wine, rather than postponing it another day as might seem more rational from her then-local perspective. I thus suggest the following atemporal constraint on reasoning:

(AC) In any given situation: one rationally ought to act and reason as one would recommend from a timeless perspective.

The incremental reasoning which leads to eternal postponement cannot be endorsed from a timeless perspective, and so is disqualified by AC. It is important to be clear about the precise sense in which this principle is ‘temporally neutral’.  It is not to say that you must always make the same decision at all times. For example, one might reasonably wish to have a carefree youth but act responsibly upon maturity. So long as one’s younger self endorses (or should endorse) the idea of growing into responsibility, and the older self does not (reasonably) regret their carefree past, then there is no temporal inconsistency here. All time-slices of the person can agree that it is good to be carefree in youth and responsible in maturity. So, different things may be recommended consistently from the timeless perspective, depending on which time in the life is being assessed.  (Note that, as per the parenthetical emendations, the timeless perspective is somewhat idealized, and not to be held hostage by the irrational attitudes of one's actual time-slices.  Also: I have assumed full information for ease of exposition, but decision-making under uncertainty may be accommodated by stipulating that the 'timeless perspective' remains under the same informational constraints as the actual deciding time-slice.)  Relatedly, an agent who is temporally neutral in my sense might nonetheless take timing into account insofar as it has narrative significance: one might prefer to experience hardship earlier in life so that it may later be redeemed, for example.  Again, the crucial point is that this does not give rise to changing or inconsistent global preferences as time passes.

The above discussion may seem somewhat esoteric, so let me expand on the intuitive connection to the issue of temporal neutrality. The core idea is that we start with the ‘big picture’ – the global preferences one has about the shape of their life as a whole – and work our way down to the recommendations that are appropriate for each particular moment.  In other words, each momentary stage submits to the rational authority of their idealized self, conceived of as a temporally-extended and unified whole.  For any given momentary stage, the recommendations offered it are temporally neutral, in the sense that they do not differ depending on the temporal location of the assessor. (As we have seen, even one’s elderly self can agree that it is good to be carefree in youth.)  So, the degree to which it is bad for me to suffer pain at t1 does not depend on what time I consider it from. Whether t1 is past, present, or future, its quality remains the same.

Caring about the Present

Temporal neutrality seems most plausible when it is contrasted with the bias towards the near, i.e. a preference for instant gratification at greater future costs. Parfit identifies what he thinks are two further – and seemingly more defensible – ways that people commonly violate temporal neutrality. We exhibit a bias towards the future, such that we are relieved when our suffering is past, and also the bias towards the present, such that we “care more about present pains.”² Parfit invites us to imagine a character, Timeless, who lacks these biases, so that his emotional responses are impervious to the passage of time. Such affective constancy seems bizarre, as Parfit rightly notes. Even when it comes to the bias towards the near, it seems perfectly reasonable for us to feel “mounting excitement… as some good event approaches the present—as in the moment in the theatre when the house-lights dim.”³

I want to suggest that the temporal neutralist can grant all this, for temporal neutrality is a thesis about what we are rationally required to prefer or choose, not what we are required to feel.  These may be expected to come apart.  Brad Skow, however, argues that the two are intimately connected: pleasant affect, for example, may be generated from the thought that our preferences have been satisfied.  We are more pleased to be reminded of future benefits than past ones, Skow suggests, because we more strongly prefer the future benefit to obtain.  The fact that we are more affectively attuned to our future than past welfare is thus taken to indicate a temporally biased preference structure.  Skow concludes that "biased [affective] attitudes will be rational just in case biased preferences are."⁴  But we may question whether the connection is really so tight, as there are compelling reasons to expect the two to sometimes come apart.  Our affective responses are designed to help us respond successfully to threats and opportunities in our environment, and this changes over time. I should feel fearful and alert when faced with a dangerous predator, and all my time-slices can endorse this. It does not follow that all my time-slices should themselves maintain such a state of arousal – once the danger has passed, there would be no point.

So I agree with Parfit that attention and affect should change over time, in sync with our changing circumstances. But rational preferences are another matter. I may feel greater excitement about an imminent lesser benefit, as I attend to it in the present, even though I judge the distant greater benefit to be preferable, and so would sooner give up the imminent benefit if forced to choose. Note that I do not later regret feeling excitement before the show, whereas I would regret choosing a lesser nearby good over a greater distant one. This is a revealing difference. Changing feelings may be endorsed from a timeless perspective, and so are consistent with temporal neutrality, unlike fluctuating preferences.

It may be objected that I prefer water over bread when I am thirsty (at time t1), but reasonably reverse this preference when I am hungry (at t2). But these are not really conflicting preferences, for we may expand each to include an implicit temporal parameter. I prefer that I receive water at t1 over bread at t1, and that I receive bread at t2 over water at t2. At no time do these preferences change. All that changes is which tenseless preference gets expressed by my tensed expression ‘I prefer water (now) over bread (now).’ All this is revealed by the fact that if you grant my t1-self all the relevant information, and ask him to choose whether at t2 he will get to receive bread or water, he would presumably disregard his present thirst and choose the bread.

Preferring Pains to be Past

So much for caring. Parfit has a more direct objection, though. As part of the bias towards the future, he thinks that we might really prefer to suffer a greater pain in the past rather than a lesser pain in the future. Assuming that we cannot affect the past, it is difficult to see what this preference amounts to, if it is meant to be distinct from mere feelings, e.g. relief upon learning that a painful operation is now past.⁵ There is no genuinely possible choice in which the preference would be directly revealed. But perhaps we can imagine an impossible choice, involving time travel or divine intervention (in foreknowledge of one’s backward-looking prayer, say). Can we then imagine, in this impossible scenario, reasonably choosing to have one’s pains amplified and shifted into the past?  It may initially seem so, but I think there are at least three reasons to be skeptical of this apparent intuition.

First, we are used to the equivalence between being over and ceasing to extend in time. So a typical preference for a pain’s being over is inseparable from the preference that it become no longer – a preference that the temporal neutralist can certainly endorse. Granted, the fact that our pain extends no further beyond time t1 will gives us more satisfaction at t1 than beforehand, but this is just to return to the familiar point that we are affectively attuned to our present situation. The tenseless preference may thus be more salient at some times than others, but that does not make its content irreducibly tensed.

Second, even if we appreciate that in this special case the past pain is no shorter, we may reasonably doubt – given the special circumstances – that this really matters. In Parfit’s ‘Case Two’, for example, you wake up the day after a painful operation, though you cannot remember exactly how long it was.⁶  A nurse tells you there are two possibilities: (1) You had five hours of pain, but the operation is now over; or (2) You had just two hours of pain yesterday, and have another hour still to come. Parfit suggests that the first seems preferable, despite being worse for your life as a whole. But one might question the latter judgment. Plausibly, the disvalue of something so subjective as pain depends in part on how we conceive of the experience, and how we incorporate it into the narrative structure of our larger life-story. One episode of extended pain may have a roughly constant disvalue no matter its actual duration, at least if you cannot subjectively tell the difference. If this is so – a controversial position that I argue for in the Appendix to this paper – then the first option is actually better for your life as a whole (containing but one episode of pain, rather than two), and so there is no 'bias' involved in preferring it.

Third, and most importantly, even if we come up with a case where the past pain truly is worse for us, we may fail to fully appreciate that the change is really experienced at all. By the time we are in the present, our past experiences have already been and gone.  Because we did not experience them the first time through, it may seem that shifting pains into the past is simply a way to make them disappear altogether. (Of course, this intuition illicitly reflects our assumption that we cannot really change the past – an assumption that must be rejected in the scenario under consideration.)

This debunking explanation may be supported by noting that our intuitive concern is for the dimension of personal-time, not objective time.  If I will soon time-travel back to 1984, I do not want any pains to be transferred to my 1984-self, for that stage is in my future, despite being in the past.  The peculiarities of my personal trajectory make vivid to me the fact that the 1984 experiences really happen – that my 1984-self really suffers any extra pains that get sent his way.  But it must be recognized that, objectively speaking, 1984 is no less past than 2004.  The two predicaments have the same ontological status, so there is no difference in being/reality that could ground normative significance in the one case but not the other.  That is, I take the supervenience of ethics on metaphysics to suggest the following argument:

(P1) It is bad / undesirable for my 1984-self to suffer.

(P2) My ordinary past selves possess the same relevant metaphysical qualities.

Hence: (C) It is bad / undesirable for my ordinary past selves to suffer.

The upshot from this section so far is that when we take care to avoid the three outlined mistakes, and instead vividly appreciate the genuinely worse pain that our past self really would suffer if we made the temporally biased choice, it no longer seems so obviously preferable. Parfit's objection is thus defused, as we find on reflection that there is no longer any clear intuition weighing against temporal neutrality.

This conclusion may be further supported by considering cases in which temporally biased preferences may be indirectly revealed – through the buying of insurance options.⁷ A risk-averse agent will buy different insurance options depending on which outcomes he would most prefer to avoid.  In general, he will be willing to pay to improve his least-preferred outcome at greater cost to the quality of his most-preferred outcome.  Parfit's temporally biased patient, for example, might accept the following deal: (i) he pays $50 if it turns out he is in the first predicament, and his long operation is over, and (ii) he receives $49 if it turns out he is in the second predicament, whereby he had a short operation but a little more is still to come.  Such a deal allows him to make the "worse" outcome slightly less bad, since at least he gets $49 compensation if it turns out he has more pain still to come.  To a risk-averse agent, this is worth imposing a $50 penalty on the "best" outcome, since in that case he'll just be happy that the operation is over.  Tom Dougherty argues that a patient with these biased preferences is a "chump", i.e. subject to being money-pumped by a series of offers that will (transparently) leave him worse-off whatever happens.  After all, his earlier preference would have been to undergo the shorter operation.  So, back then he would have been willing to pay $50 for an insurance offer that would pay back this plus $49 just in case he ends up in the first predicament, i.e. having suffered the long operation.  Combining these two deals, he ends up with a guaranteed dollar loss no matter which outcome results in the end.  This is a strong (if not conclusive) indicator of irrationality.  At the very least, such temporal inconsistency looks like a defect in the constitution of a temporally-extended agent, for whom rational unity and coherence would seem to be constitutive norms.  This suggests an idea to be developed in the next section: that the temporally biased agent risks being less of a person than the agent who embraces norms of temporal neutrality.

Personal Identity and Self-Projection

A new challenge may be posed by those who – like Parfit – deny the commonsense view there is an enduring self or ego that persists through time while being wholly present at each moment. For the alternative is to see personal identity as a mere construction of sorts, perhaps consisting in nothing more than the right sorts of physical and psychological connections between various temporal parts, or momentary time-slices, of a person. In that case, it may seem that what I am, at the most fundamental level, is not a whole temporally-extended person at all, but just a momentary stage or time-slice. The relation my momentary self bears to my future selves then rather resembles the relation my person bears to other, more or less similar, persons. However, most of us think that it may be quite rational to be personally biased, in the sense of favouring some persons (e.g. ourselves and those close to us) over others. So why is it not likewise rationally permissible to favour some momentary stages (e.g. our present moment, and those close to us-now) over others?⁸

I grant the underlying metaphysical picture, and will remain neutral on whether full-blown personal neutrality is rationally required. So I grant that if one were to psychologically self-identify with one’s momentary time-slice only, then bias against later time-slices could reasonably follow. But in such a case, one would arguably no longer be a person with a future at all. This is implied by the following endorsement condition (EC) on the construction of personal identity:

(EC) Any temporal proper part of a person must (at least implicitly) endorse its incorporation into the temporally-extended whole.

The sort of tacit endorsement I have in mind is satisfied by (being disposed towards) conceiving of oneself as a temporally-extended person, for example identifying with the subject of one’s memories, anticipating future experiences, etc. We may imagine someone – call her Mini – who, upon rejecting endurantism about the self, goes on to purge herself of all such thoughts. By failing to imaginatively project herself beyond the confines of her present moment or otherwise consent to incorporation into a temporally extended whole, Mini’s person would extend no further than the time-slice. Subsequent time-slices will of course bear various relations to her, being continuous in many notable respects, but they no more comprise a unified person than do people with similar interests automatically comprise a club. Persons, on this view, are voluntary associations.

Mini is not biased against her future, then, because she has no future. We do better to describe the situation as one in which she is biased against the people who later inhabit her body. Granted, if we reject the strict requirements of personal neutrality, then we may consider Mini to be reasonable in her bias. But it is not really temporal bias. It is just an unusual case of personal bias.  Note that the same may be said of Midi – a more moderate version of Mini, who only partially identifies with his past and future stages.  The sort of bias proposed here is not one that tracks distance in time for its own sake.  Rather, it tracks distance in psychological self-identification from the momentary self, and it just so happens that this correlates with temporal distance.

In any case, there are good reasons why we should not aspire to this biased condition. A momentary self is a small and uninteresting being. It can attain little of value. The ends we take to be truly worthwhile require extended time and investment. It is only by situating one’s present stage within the narrative arc of a whole life that it becomes meaningful or even comprehensible. In order to realize these deeper values, then, we must incorporate ourselves as a larger being than our momentary time-slice alone. And, once this larger narrative construction is identified as the locus of our personal concern, it will no longer make sense for us to be biased in our concern for its various temporal parts, or to identify more strongly with some stages rather than others.  Though the whole may not be metaphysically fundamental, we nonetheless recognize that it is fundamentally what matters.

Conclusion

This paper has addressed three Parfitian arguments against temporal neutrality. First, Parfit demonstrates that our patterns of affective concern are temporally biased in rationally permissible ways.  I grant this, but respond that once we distinguish feelings and preferences, the current objection gives us no reason to reject temporal neutrality as regards the latter.  Next, I considered Parfit’s more direct claim that we may reasonably prefer to suffer a greater harm in the past in place of a lesser harm in the future. Here I sought to debunk the intuition by suggesting three reasons we might initially be misled into finding such a claim plausible – most notably, that we were smuggling in the assumption that the past cannot be changed, and so to “shift” a pain into the past is to make it disappear altogether.  Parfit's claim is further undermined by noticing that those who accept it are, in the technical parlance, "chumps".  Finally, I developed a challenge based on Parfit’s reductionist account of personal identity and persistence through time. If we see ourselves as momentary time-slices rather than temporally-extended persons, we may reasonably be biased against subsequent time-slices. However, I showed that this is better understood as a form of interpersonal bias rather than temporal bias within a person, and – furthermore – that it would be irrational to adopt such a self-conception in any case.  So, despite being sympathetic to many of Parfit's insights, I conclude that they do not constitute a solid case against temporal neutrality.⁹

Appendix:  Indiscernible Pains

Here I will defend the principle – appealed to in my section 'Preferring Pains to be Past' – that the normative significance of a pain depends more on its perceived than actual qualities (duration and intensity).  Empirical research suggests that most people prefer an objectively longer experience of pain if the intensity tapers off towards the end; afterwards, the whole experience seems to them to have been more bearable.¹⁰ As a general principle, we should consider each individual to be the best authority when it comes to their own welfare, especially for something so subjective as what sensory experiences they prefer to undergo.  So this provides some initial motivation for the view that objectively longer pains are not necessarily worse for us.  What really matters, we may plausibly think, is the subjective suffering they cause.

It may be objected that memory flaws can distort our retrospective understanding of how bad a pain really felt to us at the time.  However, even if we’re drawn to a more objective theory of hedonistic evaluation, according to which a person may be mistaken in their judgments of how bad their pain was, we may wish to count as distinct experiences only those that are qualitatively distinct.  (This ‘indiscernibility principle’ is supported by the intuition that duplicate universes, or Nietzschean eternal recurrence,¹¹ would make no difference to the value of the world.)  Most of the time, even intrinsically identical pains are embedded in discernibly different experiential contexts, and so count as recognizably distinct. But in Parfit’s hospital case, it seems as though the duration doesn’t introduce sufficient qualitative differences. After a while, ex hypothesi, the many moments of hospitalized agony all blur together, and we may think the reason for this is precisely that there is truly nothing in the experiences to distinguish them. So, by the indiscernibility principle, they count for just one. It would follow that one’s preference for the longer past pain, in this special case, is not an instance of temporal bias at all. It is also what is best for one’s life as a whole, considered from a timeless perspective.

Should we accept the indiscernibility principle?  An in-depth examination of the issue is beyond the scope of this paper, but I will at least highlight some relevant considerations.  Technically, the principle should probably be modified to admit of degree, rather than making a sharp cut-off between 'identical' and 'distinct' experiences.  We could instead say that multiple experiences count for more the more distinct they are.  More pressingly: one might object that the patient would presumably prefer the pain to stop sooner rather than later, if asked during the extended experience.  But the questioning probe is a confounding factor that would significantly change the quality of the experience, dispelling temporal blur.  By using the probe itself as a signpost, the patient could distinguish the pre-probe and post-probe experiences, so this is no counterexample but merely a case where the indiscernibility principle would not even apply.  The critic might revise their objection to appeal to the patient's standing disposition to want the pain to stop as soon as possible (even absent any explicit probes).  But such drives do not necessarily reflect our true preferences: a torture victim may be disposed to "agree" to anything just to make the pain stop, but that doesn't mean that they would reflectively endorse present pain-avoidance as their ultimate value.

Finally, it may be objected that the principle yields intuitively absurd results: that it might be no worse to torture a billion people than just one, for example.  But such intuitions gain their force from "ordinary" cases, where different people have different memories, etc., and thus involve discernibly different experiential contexts.  Special care is required to conceive of a situation where the indiscernibility principle would actually apply – e.g. a dystopian 'farm' of duplicate brains-in-vats, programmed to have exactly the same series of experiences throughout their existence – but then the intuition is less clear.  To ensure that all variables are controlled, imagine a 'digital person' or conscious Artificial Intelligence, whose 'life' is constituted by the running of a computer program.  Would it matter how many duplicate copies of the program were run?

Clear intuitions are difficult to come by in these cases, so we may prefer to decide the issue on more general theoretical grounds.  It is commonly thought, for example, that the welfare value of a life is not purely aggregative: most of us would prefer a life that trends upwards in quality, rather than one that trends downwards, even if the sum total of pleasure and pain comes to the same in either case.  We care about the shape of our life as a whole (and, arguably, not just because feedback from this influences the subjective character of our experiences).  A stronger claim is that this is all that matters.  Mere multiplication – say where every event extends for twice as long – would then be dismissed as lacking in normative significance, so long as the broad contours of the life remained much the same.  This explains why terminal illness is especially tragic in youth: the shape of their life has been disfigured or cut short, in a way that the elderly patient's has not.  The value of an extra year of life depends on what would be achieved with it, and whether it would contribute anything new or significant to the overall structure of the life.  If it is all much of a sameness with what has gone before, it may not have any impact on the quality of the life taken as a whole.  It is these sorts of considerations that I take to motivate the indiscernibility principle, and thus my second objection to Parfit's case: we may reasonably doubt whether the past pain he describes, even if longer, is for that reason worse for us.

Endnotes

Bibliography

Dougherty, T. (ms.) 'Pumps and Chumps'
Kahneman D., Fredrickson B., Schreiber C., Redelmeir, D. (1993) 'When more pain is preferred to less: adding a better ending'. Psychological Science 4: 401–5.
Parfit, D. (1987) Reasons and Persons. Oxford: Clarendon Press.
Pollock, J.L. (1983) 'How do you Maximize Expectation Value?' Noûs 17 (3): 409-421.
Prior, A.N. (1959) 'Thank Goodness That's Over' Philosophy 34 (128): 12-17.
Skow, B. (ms.) 'Time-biased Attitudes and Time-biased Preferences'