This piece is part of a short series on how the Wild Animal Initiative research team thinks about uncertainty. These posts represent the opinions of individual staff members, and not necessarily the organizational position of Wild Animal Initiative on uncertainty. This summer, we are raising $50,000 to further our research—you can learn more here and support our work here. This essay is crossposted from the WAI blog.

A recent article offered new estimates of wild animal numbers that differed for some taxa by an order of magnitude from Brian Tomasik’s often-cited estimates, highlighting the uncertainty we face about something as basic as how many wild animals there are. In introducing the problem of wild animal suffering, the author wrote, “We can all agree for example that 10 units of suffering by one individual is much worse than 10 individuals suffering one unit of pain.” I was very surprised to read this. While the idea is intuitive to me, I don’t consider the intuition justified, and would class it under ‘extreme moral uncertainty’. Uncertainty like this about fundamental ethics is, in my view, the greatest roadblock to large-scale intervention for wild animal welfare. However, there is also plenty of uncertainty on more empirical questions. Fortunately, these questions seem much more tractable, and answers to some may reveal actions we can take that are robustly good under a range of ethics. Here I will summarize the major points of moral and practical uncertainty that impact my own research at Wild Animal Initiative and how I think uncertainty should be addressed in this kind of work going forward.

Which animals are moral patients?

If two animals are both capable of experiencing suffering and pleasure, their interests matter equally. However, individuals of different species may have dramatically different needs and interests. For example, for an animal with a greater capacity for learning there may be a greater variety of stimuli that can trigger a feeling of fear, or their psychological response to a simple stimulus like food deprivation may be more complex if they have a larger repertoire of behaviors to express, instincts to satisfy, or experiences to associate with. It is not clear how to weigh up these differences. If individuals of one species have fewer/simpler preferences, do those preferences count more on an individual basis because they consume more of the animal’s emotional attention at any given time? For example, an animal with a longer time horizon, who is able to form preferences about tomorrow, might feel less strongly about a bad time today than an animal who cannot conceive of anything beyond what they are experiencing in the present. This is ultimately an empirical question, whether something as primal as the feeling of hunger generates more psychological stress in cognitively simpler animals.

The capacity to differentiate enjoyable and unenjoyable experiences has often been used as a proxy for moral patienthood. Much great work has gone into the question of which species have this capacity; this is especially important to address for the most numerous animal groups, such as the arthropods and nematodes. However, there is also a question as to which developmental stages of a given species are moral patients and to what degree. Insects vastly outnumber vertebrates, and in most species, eggs vastly outnumber larvae which in turn vastly outnumber mature animals. For example, while we may be confident that adult fish are sentient, many of them produce numbers of eggs/larvae rivaling insects, the overwhelming majority of whom die within a few days. The answers to these questions will determine which interventions we should ultimately prioritize.

A linear scale of suffering and pleasure?

Returning to the claim that “10 units of suffering by one individual is much worse than 10 individuals suffering one unit of pain,” let’s consider the layers of uncertainty facing this statement, and their implications for studying wild animal welfare. First, the real-world meaning of a unit of suffering is ambiguous. A unit of suffering may depend on an individual’s perception, as with the self-assessed pain scale used in hospitals. Unfortunately, most wild animals can’t self-assess, so it may be more practical to measure suffering based on something like the duration of a pleasant or unpleasant experience – a day of hunger, a minute of sex, or a second of being eaten alive. How an animal’s psychological well-being scales with these real-world measures presents another layer of empirical uncertainty on top of the controversial population ethics.

Uncertainty surrounding the scaling of suffering against time or other metrics has major implications for which parameters are important in models of wild animal welfare and what ecological data we need to collect. For example, if extremes of suffering/pleasure are of exponentially greater concern than moderate values, then the variance in welfare across a population or even across an individual’s lifetime may be even more important than its average. Similarly, exponentially-scaling suffering might require us to place great importance on cause of death in models of wild animal welfare based on the severity of the experience, even if it only consumes a small fraction of an individual’s lifetime. As a ‘unit’ of suffering or pleasure remains poorly defined, it is also not clear whether the placement of these experiences at opposite ends of a single scale of ‘welfare’ is ethically justified or just a convenient way of summarizing a collection of affective states.

How do cause-specific mortality rates interact with one another?

As Abraham Rowe discussed in his recent post, non-target effects of interventions are a major concern (though not unique to wild animal welfare). Jane Capozzelli also highlighted numerous interventions from the ‘Conservation Evidence Database’ relevant to wild animal welfare, mentioning each of their potential non-target effects on other species in the same ecosystem.

At the population level, there is a similar need for research into potential non-target effects of interventions aimed at reducing mortality due to specific causes – such as eradicating a disease or excluding predators – or among specific age groups. For example, if many of the animals who are killed by predators would have died shortly thereafter from hunger, then it is imperative to understand the relative severity of these types of deaths. On the other hand, suffering from hunger might leave animals vulnerable to disease or predation, so reducing hunger might have an outsize effect on overall mortality rate. Any reduction in mortality will also increase the population size by some amount, which may be good or bad depending on whether the lives of those animals are typically worth living.

How does population density affect survival and welfare?

If a wild population grows without accompanying growth in its habitat or resource availability, every individual’s life may become a bit more challenging. Thus, increasing population density typically reduces life expectancy as well as, presumably, day-to-day quality of life. Understanding the magnitude of this effect is crucial for predicting the impact of any intervention that would disproportionately affect either population size or habitat. The implications of population density also interact with our uncertainty about fundamental ethics: where to draw the line between net-positive and net-negative welfare, whether to weigh suffering and pleasure equally, and whether to value a population based on the sum, average, or some other statistic of its constituents’ welfare. This is because average per-individual welfare – and therefore the probability that their welfare is net-positive – will be higher in a low-density population, but the number of individuals experiencing that welfare will be lower.

How to deal with uncertainty

For researchers, extreme uncertainty requires resilient model-building. Welfare models should be as philosophically open-ended as possible so that the math works, and their output is informative, on a range of views. Of course, there are limits to this: non-utilitarians might object in principle to quantitative models of welfare, and we can’t realistically account for every conceivable moral virtue. An alternative approach could be to build a comprehensive library of more philosophically detailed models. But apart from the additional workload of that approach, at this early stage of welfare biology it may be especially important to develop relatively simple paradigmatic models, more for their ability to unify the field and inspire new research avenues than for their immediate real-world application. As the field matures, the practicality of exploring more precise models should increase.

An example of resilient model-building under uncertainty comes from the early history of another field: the Drake Equation in astrobiology. This model attempts to estimate the number of intelligent, communicating civilizations in the galaxy based on a mix of traditional physical parameters and massively uncertain probability terms. One term represents the “fraction of stars with habitable planets.” The model’s continued relevance to its field owes to the fact that it doesn’t rely on any specific definition of habitability, for example, allowing the model to accommodate new discoveries and ways of thinking about each parameter.

Welfare-focused models of ecological processes should follow a different approach, being as explicit about biological mechanisms and individual outcomes as possible. Given our profound uncertainty about whether most animals’ lives are dominated by pleasure or pain, population sizes by themselves can’t tell us much about welfare. However, models that break population numbers down according to life history features like age/lifespan or cause-of-death provide information that might be useful for understanding and improving wild animal welfare. Even if we are not confident whether the animals in question have lives worth living, these models can help identify the likely best and worst fates experienced by animals of a given species.

Practical considerations may also override our ethical uncertainty in the near term. Even if we are classical utilitarians, who make no distinction between reducing suffering and increasing happiness, it will be easier as we begin to learn about wild animal welfare to identify sources of pain and how to reduce them than it will be to identify sources of pleasure and how to enhance them. In the longer term, when we are confident enough to intervene, scientific understanding of ecology and how welfare varies in a population can help design courses of action that are especially resilient to philosophical uncertainty.

Extreme uncertainty in wild animal welfare requires resilient model-building

Which animals are moral patients?

A linear scale of suffering and pleasure?

How do cause-specific mortality rates interact with one another?

How does population density affect survival and welfare?

How to deal with uncertainty