Probabilistic moral reasoning is unintuitive

"If there's a 1% chance that shrimp feel pain, and we farm 440 billion of them annually, the expected harm is..." assumes access to a pretty weird way of thinking about what's good and bad.

At the Animal Welfare Economics conference last week, some folks and I discussed how we think about the harms associated with different animal products, e.g., a glass of milk is less bad than an omelette because cows produce like 12 tons of milk a year (etc.). At which point someone wisely pointed out: a utilitarian framework is not how any of us got involved in this work in the first place. We’re in it for much squishier reasons about how animals deserve to be treated, their rights, their dignity, etc. Why would we expect cost benefit analysis to persuade others when it didn’t persuade us?

Today I remembered a nice paper touching on the distinction between these modes of moral reasoning: Berman & Kupor (2019), Moral Choice When Harming Is Unavoidable (emphasis added):

Past research suggests that actors often seek to minimize harm at the cost of maximizing social welfare. However, this prior research has confounded a desire to minimize the negative impact caused by one’s actions (harm aversion) with a desire to avoid causing any harm whatsoever (harm avoidance). Across six studies (N = 2,152), we demonstrate that these two motives are distinct. When decision-makers can completely avoid committing a harmful act, they strongly prefer to do so. However, harming cannot always be avoided. Often, decision-makers must choose between committing less harm for less benefit and committing more harm for more benefit. In these cases, harm aversion diminishes substantially, and decision-makers become increasingly willing to commit greater harm to obtain greater benefits. Thus, value trade-offs that decision-makers refuse to accept when it is possible to completely avoid committing harm can suddenly become desirable when some harm must be committed.

Experiment 1 gives the flavor:

Each participant was randomly assigned to either an avoidable-harm condition or an unavoidable-harm condition. All participants assumed the role of a doctor who faced a decision regarding whether to pull life support from a dying patient to save money that would be used for cancer research. In the avoidable-harm condition, participants could choose to pull life support from one dying child to save $60,000 for cancer research or choose not to do so. In the unavoidable-harm condition, participants instead faced a decision about whether to pull life support from one dying child to save $800 for cancer research or to pull life support from two dying children to save $50,000 for cancer research…participants were more likely to prefer to commit greater harm in exchange for greater benefits in the unavoidable-harm condition (69.9%) than in the avoidable-harm condition (28.6%).

In plain English: when the choice was between letting zero or one kid die, participants valued the kid’s life at > $60,000,¹ but in a situation where one kid is definitely going to die, most people value the life of the second kid at ≤ $49,200. In other words, participants have both deontological (rule-based) and utilitarian (utility-maximizing) moral frameworks available to them, and you can activate one or the other depending on how you ask the question. Also, people do not have a fixed value for how much a kid’s life is worth. Once you are forced to value the first kid’s life at ≤ $800 (which is absurdly low), the second kid’s life seems worth somewhere between that number and infinity dollars, but apparently less than $49,200.

The ‘basic argument’ for vegetarianism is much less uncertain than the argument for veganism

James Rachels (2004), The Basic Argument for Vegetarianism, makes the following case (emphasis added):

It begins with the principle that it is wrong to cause pain unless there is a good enough reason. The qualification is important, because causing pain is not always wrong. My dentist causes me pain, but there’s a good reason for it, and besides, I consent…The second step in the argument is to notice that in the modern meat production business, animals are made to suffer terribly. There is a reason for this suffering, too. We eat the meat, and it helps to nourish us. But there is a catch: we could just as easily nourish ourselves in other ways. Vegetarian meals are also good. Nonetheless, most people prefer a diet that includes meat because they like the way it tastes. The question, then, is whether our enjoyment of the way meat tastes is a good enough reason to justify the amount of suffering that the animals are made to endure. It seems obvious that it is not. Therefore, we should stop eating the products of this business. We should be vegetarians instead.

This is an appeal to an intuitive rule, and the language is definitive. It does not depend on committing to a particular moral theory but can instead be grounded in widespread, common-sense moral terms. Sure, you might argue that there’s some chance that animals don’t matter at all, that there's just no one inside to suffer, but that, I think, is the unintuitive position. Most people look in a pig’s eyes and see a light there the same way there is in a dog’s eyes or a child’s. (Historically babies were a harder case.) No difficult utility-maximization under uncertainty is asked for.

Smaller, more evolutionarily distant to us animals require more of a leap. Back to Rachels (emphasis again added):

The question of which other animals feel pain is a real and important issue, not to be settled by appeals to “common sense.” Only a completed scientific understanding of pain, which we do not yet have, could tell us all that we need to know. In the meantime, however, we do have a rough idea of what to look for. If we want to know whether it is reasonable to believe that a particular kind of animal is capable of feeling pain, we may ask: Are there nocioceptors present? Are they connected to a central nervous system? What happens in that nervous system to the signals from the nocioceptors? And are there endogenous opiods? In our present state of understanding, this sort of information, together with the obvious behavioral signs of distress, is the best evidence we can have that an animal is capable of feeling pain.

Relying on such evidence, some writers, such as Gary Varner, have tentatively suggested that the line between animals that feel pain and those who do not is (approximately) the line between vertebrates and invertebrates. However, research constantly moves forward, and the tendency of research is to extend the number of animals that might be able to suffer, not decrease it.

Now we’re in the land of probability. “Might be able to suffer” — how certain do we have to be to make costly changes? This question, I think, is hard, and how normal people navigate it is understudied.

Shrimp, the limiting case

Do shrimp feel pain? We don’t know. Consciousness is very poorly understood. Self-awareness is metabolically costly, so why would shrimp have evolved it? (The answer, if we ever learn it, is probably as a byproduct of some other selected for trait). But, as Rachels notes, as we develop finer-grained measurement, our understanding of who can suffer moves outwards. Per the Shrimp Welfare Project, “Recent studies suggest that shrimps are far more complex than we once believed,” e.g., they express “Protective and Self-Directed Behavior…This targeted, prolonged behavior suggests discomfort and an attempt to alleviate it, rather than a mere reflex.” But suggests might not be good enough for people. What to do?

One answer is to apply a nonhuman version of the precautionary principle, that when something bad might plausibly happen, “precautionary measures should be taken even if some cause-and-effect relationships are not fully established scientifically.” Another is to round the other way and say, if I’m ≥ 50% sure that an animal is not conscious, I’m in the clear to eat it. A third option is to say, like the Shrimp Welfare Project does, that if we aren’t sure, we can start doing some things that will definitely reduce pain if they’re conscious and aren’t too expensive, e.g., stunning before slaughter, better transport and handling, and improved water quality.

Or you might approach the problem like a philosopher or economist: using the expected value framework, multiplying expected outcomes by their probabilities. Stuart Rachels (James’s son) does this when estimating how many holocausts bad factory farming is. He notes that factory farmed animals so vastly outnumber humans killed in the Holocaust that even if we think human suffering is, say, 100 times more important than an animal’s, then we still get to something that’s many, many holocausts worth of suffering. Or (emphasis added),

Suppose I’m wrong. Suppose that, for whatever reason, human pain is ten times worse than animal pain. On that assumption, factory farming over the last twenty years has still caused pain morally equivalent to five hundred Holocausts. Or suppose there's only a 10% chance that the arguments in this paper are correct. On that assumption, factory farming, again, has had the expected utility of five hundred Holocausts. And if there's only a 10% chance that animal pain is 10% as bad as human pain, then factory farming has had the expected utility of fifty holocausts (or really more, since I’m ignoring a lot of the suffering caused by industrial farming). The philosophical arguments for vegetarianism are easy. What's hard is getting people to stop eating meat.

And finally I get to my point: I think this way of thinking is correct, and extremely unintuitive. The sense in which this argument is “easy” is that it’s easy to explain to other philosophers who are used to defining rationality as maximizing one’s expected utility and viewing that as a reasonable normative ideal of how to think. When money is on the line, maximizing expected utility under uncertainty becomes a lot more intuitive. Professional gamblers get good at it. Appraising a mortgage applicant’s default risk, and setting an appropriate interest rate, foregrounds probabilistic outcomes. But about moral outcomes? I think this would require a lot of prepwork to explain to people.

How would we do it? I am not sure. For starters, how many other moral issues depend on empirical questions we don’t know the answer to? Abortion comes to mind, though of course everyone thinks they do know the answer to the moral status of fetuses. Other clear moral harms, we are pretty certain we know the fundamentals. We don’t have to reason under uncertainty.

And so we default to deontological arguments that attempt to establish the more basic proposition that something is right or wrong, or that emphasize the reality of animals’ pain.² But something is lost, I think. we don’t make our strongest arguments because they require too much of a leap to a weird, professionally philosophical way of thinking about moral issues.

I’m mulling this over because I am interested in why abstract arguments about right and wrong work for some people but not others. Are we weird? Were they made more effectively to us? Is there some “unlock” that can make the case and also explain its epidemic foundations?

Again, I have some hypotheses, but no research design in mind. Suggestions welcome.

1. HPMOR, chapter 82: “Hermione’s replacement value is infinite! There’s nowhere I can go to buy another one!” In chapter 108, the crux of the matter: “The whole point of having deontological ethical injunctions is that arguments for violating them are often much less trustworthy than they look.” One of my most-quoted passages, though first place probably goes to “There was nothing left to do/There was nothing left to plan.”

2. Though, as James Rachels notes, this is not strictly an emotional appeal. Rather, “[t]he point is to fill in the details of the basic argument. The basic argument says that causing pain is not justified unless there is a sufficiently good reason for it. In order to apply this principle to the case of factory farming, we need to know how much pain is involved.