by Magnus Vinding. First published in July 2022.
Some ethical views grant a lexical priority to the prevention of extreme suffering over mild forms of suffering, meaning that the prevention of extreme suffering takes precedence over the prevention of mild suffering.
Such views have been claimed to have implausible practical implications. For instance, one objection is that such a lexical priority implies that we should neglect all endeavors that do not aim directly at the reduction of extreme suffering. My goal in this post is to reply to a couple of these objections, and to clarify some key aspects regarding how one might think about prioritization in light of lexical views.
Different kinds of de facto lexical views: Theoretical and practical
Before diving into the practical implications, let us start by reviewing some different ways in which one can end up with a de facto lexical view in practice.
Lexicality in theory
Perhaps the most straightforward way to end up with a lexical view is to endorse it directly at the theoretical level. For instance, one may hold that a certain amount of extreme suffering is worse than any amount of mild suffering even in theory. Such views can then be further divided into lexical views that entail abrupt thresholds and lexical views that do not (cf. Klocksiem, 2016; Knutsson, 2016; 2021).
Strong lexicality in practice without lexicality in theory
Alternatively, one may endorse a lexical view in practice without endorsing lexicality at a purely theoretical level. For example, one might hold that the badness of any amount of extreme suffering could be exceeded by a sufficiently large amount of mild suffering in theory, while also holding that a single instance of the most extreme suffering is worse than any amount of mild suffering that could ever be physically realized in the accessible universe.
In particular, if one thinks that the badness (or disvalue) of suffering increases superlinearly as its intensity increases, and if one thinks that the worst suffering is sufficiently intense, then one could in effect think that there exists a threshold of suffering intensity such that no physically realizable amount of sub-threshold suffering could be worse than a single instance of the most extreme suffering (because the disvalue of the latter is so great).1
Weaker lexicality in practice: “Lexicality in expectation”
Weaker versions of practical (de facto) lexical views are also possible, and have been endorsed by some thinkers. One such view is what we may call a “lexical in expectation” view, which holds that the expected amount of extreme suffering is so large that no physically realizable amount of mild suffering could be worse.
Of course, such a view can take various forms, as one can defend a variety of distributions regarding how much extreme suffering will occur in the future. And these distributions will tend to imply different probabilities assigned to the claim that the amount of extreme suffering in the future (that we can influence) is lexically worse than any physically realizable amount of mild suffering. (It will obviously also vary depending on how we define “mild suffering”, as well as on the relative disvalue we assign to different states of suffering.)
Such a view may thus entail that there is, say, a 0.1 percent probability that the amount of extreme suffering in the future could be less bad than the largest amount of mild suffering (e.g. mild headaches) that is physically realizable. Yet note that this would still practically be a lexical view, as it entails that with 99.9 percent probability, no amount of physically realizable mild suffering could be worse than the amount of extreme suffering that will occur in the future. And if we hold the expected amount of extreme suffering up against the expected amount of mild suffering rather than what is physically realizable, the probability of “practical extreme suffering dominance” would become much higher still.2
Indeed, someone who endorses lexicality at the theoretical level with a significant degree of moral uncertainty could easily end up having a lower credence in lexicality compared to someone who holds the above-mentioned version of the “lexicality in expectation” view with a low degree of moral uncertainty.3
Practical implications of lexical views
Say we were to accept one of the lexical views outlined above, such that we effectively grant a lexical priority to the reduction of extreme suffering over mild suffering in practice. How, then, should we think about practical ethics and prioritization? Perhaps a good way to address this question is to start by looking at some of the practical objections that have been raised against lexical views.
Devoting resources to a narrow range of endeavors?
One objection against granting a lexical priority to the reduction of extreme suffering over mild suffering is that it implies that we should devote all our resources toward an implausibly narrow range of actions that aim directly at the reduction of extreme suffering (cf. Huemer, 2010, p. 338). Yet there are various reasons why this implication need not — and indeed does not — follow.
A narrow focus need not follow at the theoretical level
It is worth noting that value lexicality between extreme and mild suffering does not imply value lexicality between extreme suffering and other potential bads, such as rights violations or premature death. Hence, granting a lexical priority to the prevention of extreme suffering over mild suffering is compatible with granting a similarly strong priority to the prevention of other potential bads. Furthermore, one might endorse non-consequentialist duties that imply that we should — at least in some cases — pursue other actions than just those that strictly minimize extreme suffering, even if we granted a lexical priority to the reduction of extreme suffering over all other value entities at the axiological level.4
A narrow focus does not follow at the practical level
If we disregard the points made in the previous section, and assume that the reduction of extreme suffering is always our sole priority, we still find good practical reasons not to devote all our resources toward a narrow range of actions.
For while it is true that the very best ways to reduce extreme suffering on the margin will tend to fall within a fairly narrow range of causes, the same is decidedly not true from a broader perspective that includes all the endeavors necessary for humanity as a whole to reduce suffering in effective ways. After all, if humanity were to change its resource allocation such that it devoted vast amounts of resources to the best causes on the current margin, the marginal analysis would change, and new causes would become more promising on the margin. And if those causes were to be fully covered or even overprioritized, then other things would become more pressing, and so on.
When we look at the totality of endeavors that are necessary for the reduction of extreme suffering — from a broad as opposed to a momentary marginal perspective — we find that they are numerous and diverse indeed. They include the acquisition of knowledge in a wide range of fields, from mathematics to sociology, as well as the skillful application of such knowledge, at every level ranging from grassroots activism to the highest political offices. They also include many endeavors that reduce extreme suffering in rather indirect ways, such as increasing humanity’s ability to cooperate, improving humanity’s values, and increasing our overall capacity to reduce suffering (Vinding, 2022a, ch. 9).
Moreover, even if we zoom in on a single individual who aspires to reduce extreme suffering as effectively as possible, it is still dubious to claim that such a person should adopt a very narrow focus, for at least two reasons.
First, even if such an individual should ideally focus on a single “most effective cause”, there will likely be great empirical uncertainty as to what that cause is, which may warrant broad exploration into a wide range of plausible causes in order to identify that most promising cause (relative to one’s talents, motivations, etc.).
Second, if we assume that a given individual had already identified their single “most effective cause”, it by no means follows that this single cause will imply a particularly narrow focus. Indeed, competent action within virtually any promising cause — whether it be the abolition of factory farming or the reduction of s-risks due to AI conflicts — will tend to require a wide range of insights and practical implementation skills.
In short, a singular focus on the reduction of extreme suffering does not imply a narrow practical focus (Vinding, 2020, sec. 9.4-9.5).
Ignoring mild suffering?
A related objection is that views that give lexical priority to extreme suffering over mild suffering will imply that we should ignore all (seemingly) mild suffering in practice, which is arguably implausible. Yet there are various reasons why this supposed implication does not follow (Vinding, 2020, sec. 8.11).
First, views that grant a lexical priority to the reduction of extreme suffering still tend to hold that the reduction of mild suffering is valuable when other things are equal. Therefore, if one were wholly uncertain as to the eventual effects on extreme suffering, these views would deem it worthwhile to reduce the mild suffering in question.
It may then be objected that other things are virtually never equal in practice, and hence the impact that our actions have on mild suffering per se should virtually always be completely disregarded in practice. Yet even if we grant that claim, there are still good reasons to reduce mild suffering in practice. One reason is that ignoring mild suffering may condition us to also ignore more intense suffering. In contrast, if we make an active effort to reduce all suffering — including mild suffering — then this likely reinforces a commitment to the reduction of suffering, which suggests that such efforts tend to (slightly) reduce intense suffering in expectation.
Second, and more importantly, there is the point that we face considerable empirical uncertainty. Unlike the theoretical case in which we can simply stipulate that some being experiences mild rather than extreme suffering, the practical reality is that we do not know from the outside which beings are — or shortly will be — experiencing extreme suffering. This means that there is a risk that beings who appear to merely experience mild suffering are in fact experiencing extreme suffering. And this is not purely hypothetical.
After all, even in the case of humans, there are many forms of intense suffering that can be difficult to verify based on outward appearances — e.g. a state of severe depression may not look all that bad from the outside; a state of intense paranoia may give little external clues of horror; and, as an extreme case, those who wake up and feel excruciating pain during anesthesia may look wholly unconscious. Yet our uncertainty tends to be much greater in the case of non-human animals, especially when it comes to beings who look less like us, such as birds, fish, and insects. And our uncertainty gets greater still when it comes to new potential forms of sentience.
Hence, when we are confronted with beings who are most likely experiencing mild suffering, it still seems right — from our uncertain vantage point — to assign a non-zero probability to the possibility that there are instances of extreme suffering among their experiences. In other words, it seems right to think that there is some amount of extreme suffering in expectation among the experiences of those beings.
Our empirical uncertainty thus highlights the importance of thinking in terms of expected value when trying to reduce extreme suffering in practice, and it likewise reveals why lexical views do not entail a discontinuity between (what appears to be) mild suffering and (what appears to be) extreme suffering at the practical level. Instead, such views entail continuous probabilities — and continuous expected amounts — of extreme suffering among different beings.
Lexical views can thereby end up resembling non-lexical views to some extent, since these continuous probabilities concerning the presence of extreme suffering will tend to render practical priorities more continuous than one might naively assume.5
Example: Helping more beings who seem less likely to experience intense suffering
As a toy example, consider a case in which we can either help a billion small beings who, on our best guess, can experience “lexically bad” suffering with 10 percent probability, or we can help a million slightly larger beings whom we believe can experience “lexically bad” suffering with 51 percent probability. (Say that these beings all have equally long lives, and that these respective groups of beings tend to experience their most intense forms of suffering with the same frequency).
While the smaller beings most likely do not experience “lexically bad” states of suffering, whereas the slightly larger beings most likely do, a standard expected value framework would still recommend that we prioritize helping the smaller beings in this hypothetical example. Indeed, such a framework would entail that there is over two orders of magnitude as many instances of “lexically bad” states of suffering — in expectation — among the smaller beings.6
This conclusion departs quite radically from a naive decision procedure that would round off the 10 percent credences to zero while rounding the 51 percent credences to 100 (cf. the human tendency to engage in “belief digitization”).
To be clear, I am not claiming that the expected value approach outlined above is unassailable, or that it should be our only decision procedure for determining priorities. Indeed, I think there are good reasons not to rely exclusively on this approach. After all, the expected value approach might not be the best way to make decisions when our probability estimates are at a high risk of being unreliable or misguided, which suggests that we may benefit from supplementing our decision procedure with additional heuristics, to help render it more robust to miscalculation (cf. Karnofsky, 2011).
That said, it still seems worth using expected value calculations as a key yardstick — arguably even the main one — in our practical deliberations, and it certainly beats simpler alternatives such as the crude belief digitization approach that simply rounds all credences off to 0 or 100 without any justification.
Acknowledgments
For helpful comments, I thank Teo Ajantaival, Tobias Baumann, Simon Knutsson, and Winston Oswald-Drummond.
References
Huemer, M. (2010). Lexical priority and the problem of risk. Pacific Philosophical Quarterly, 91(3), pp. 332-351.
Karnofsky, H. (2011). Why we can’t take expected value estimates literally (even when they’re unbiased). Ungated
Klocksiem, J. (2016). How to accept the transitivity of better than. Philosophical Studies, 173, pp. 1309-1334.
Knutsson, S. (2016). Value lexicality. Ungated
Knutsson, S. (2019). Lars Bergström on pessimism, ethics, consequentialism, Ingemar Hedenius, and quantifying well-being. Ungated
Knutsson, S. (2021). Many-valued logic and sequence arguments in value theory. Synthese, 199, pp. 10793-10825. Ungated
Tomasik, B. (2013). Three Types of Negative Utilitarianism. Ungated
Vinding, M. (2020). Suffering-Focused Ethics: Defense and Implications. Ratio Ethica. Ungated
Vinding, M. (2022a). Reasoned Politics. Ratio Ethica. Ungated
Vinding, M. (2022b). Lexicality between mild discomfort and unbearable suffering: A variety of possible views. Ungated
- This is assuming that we are talking about suffering that we could causally influence, and assuming the validity of cosmological models that imply that the potential suffering that we can influence is finite. Such assumptions may be wrong, of course. Yet note that one might still endorse strong practical lexicality in an infinite universe, e.g. if one holds that any realistic lower bound of the density of extreme suffering throughout the universe will imply that no amount of physically realizable mild suffering could be worse.[↩]
- A similar point is made in Tomasik, 2013. Philosopher Lars Bergström likewise seems to endorse a form of practical lexicality: “There is nothing realistic that could happen that could counterbalance the bad [e.g. extreme suffering]”.[↩]
- Similarly, lexical views that entail gradual thresholds and lexical views that entail abrupt thresholds may ultimately end up being equivalent at the practical level, since empirical uncertainty means that there will be a range of uncertainty as to what qualifies as lexically bad suffering in any case. That is, if one holds that lexicality kicks in abruptly when suffering becomes sufficiently intense, one will still have uncertainty in practice as to which beings experience such “abruptly lexical” suffering, and this range of uncertainty may be roughly equivalent to the range of practical uncertainty that would be implied by a non-abrupt lexical view, even as the latter view would entail a graded “priority range” both at the empirical and the evaluative level, cf. Knutsson, 2021; Vinding, 2022b. The empirical uncertainty may be so significant that it causes plausible versions of these respective views to effectively converge in practice.[↩]
- In moral philosophy, it is common to distinguish that which has value from that which is morally right or what we have moral duties to do. While some views entail that what is morally right to do is to maximize value or minimize disvalue, other views entail that we have moral duties that do not reflect pure value optimization, cf. Vinding, 2020, sec. 6.4.[↩]
- One can also wonder whether this practical point might potentially be a distorting factor in people’s evaluations of thought experiments such as “torture vs dust specks”, since there could from a practical perspective — as opposed to from a purely theoretical perspective — be more extreme suffering (in expectation) in an unfathomably large number of “seemingly mild states of suffering” than in an inconceivably smaller number of “seemingly extreme states of suffering”.[↩]
- Note how this line of reasoning may support a focus on reducing intense suffering among insects, given how numerous insects are and given that it hardly seems justified to have extremely low credences in “intense insect suffering”. Of course, future beings who may exist in even greater numbers could well warrant greater priority still.[↩]