Modified from “Expected Utility, Contributory Causation, and Vegetarianism” (PDF)
Some people argue vegetarianism isn’t morally necessary because a single meat purchase will not actually cause more farm animals to be raised or slaughtered. Thus, regardless of whether or not the production of meat is inhumane to animals, someone who buys meat is doing nothing wrong. This argument fails to show that meat purchases are morally permissible, however, because our choice to buy meat affects the expected number of animals bred, raised, and slaughtered.
Given the size of modern animal agriculture, it seems plausible to assume that a single meat purchase is too insignificant, relative to the vast number of other meat purchases, to be noticed by the manager of a factory farm. If the manager cannot perceive any increase in demand caused by a single meat purchase, no additional animals will be raised or slaughtered, and thus no harm will have been done to animals by the purchase. In other words, it is claimed that most meat purchases are “causally inefficacious.”
This may be true but it is irrelevant to how we ought to make moral decisions under uncertainty. When we make a decision about how to act, we can never know for certain all of the actual consequences that will result from al our possible actions. We may, after making a decision to act in a particular way, come to know the actual consequences that resulted from the one action we decided upon. However, this knowledge is not helpful in making the original decision, since it is not only reached after the fact, but also limited to only one of the many possible actions we may have had to choose from. Consequently, it is more reasonable that we should make decisions, not on the basis of actual consequences (which we can’t know for certain), but on the basis of expected consequences – the product of those consequences resulting from an action and the probability of those consequences resulting – that one might reasonably predict given the available evidence. Since the expected consequences, not actual consequences, can be known when making decisions, only expected consequences can help ethical individuals decide what course of action to take.
Acting on expected consequences can be understood in problems of “contributory causation,” where many people seem responsible for causing something to happen. Jonathan Glover provides an example of contributory causation called The100 Bandits, where 100 bandits descend on a village that has 100 villagers, and each villager has one bowl containing 100 baked beans. Each bandit takes one bean from each bowl, so that each bandit ends up with a bowl of 100 beans. Now, no villager can perceive the difference made by one bean being stolen from his bowl (either at the moment or later, due to malnutrition). Thus none of the bandits would seem to have individually harmed any of the villagers and so none of the villagers should have been harmed. Yet 100 villagers are without lunch and hungry. So something is wrong.
Glover suggests we approach contributory problems like The 100 Bandits by employing a “divisibility principle” – in other words, a single agent is causally responsible for the consequences of a contributory result divided by the number of contributing agents. In this case, the hunger of 100 lunch-less villagers is divided over 100 bandits. Glover would thus say that each bandit is responsible for the hunger of one lunch-less villager. If we accept Glover’s divisibility principle, each bandit ought not to steal 100 beans because he would then be causally responsible for the disutility of one lunch-less villager.
There may be a more compelling solution to contributory problems such as this one, however, that does not attempt to reconcile actual causal responsibility with our intuitions about moral responsibility. For in the case of the Bandits, it is not true that none of the bandits is actually causally responsible for harming the villagers. At the very least a handful of the bandits are causally responsible for the villagers’ hunger – those bandits who complete threshold units. While it is true that no villager can perceive the difference made by one bean stolen from their bowl, each can clearly perceive the difference made by 100 beans stolen from their bowl. Thus there must be some number of beans between one and 100 that is the smallest number of beans a villager can perceive. Call this number the threshold unit. Say, for instance, the threshold unit is 20. Any number of beans stolen below 20 cannot be perceived. Any number of beans stolen between 20 and is perceived only as 20 beans being stolen; between 40 and 59, only as 40 beans being stolen; and so on, up to 100 beans. Thus bandits who cause a 20th bean to be stolen are responsible for the disutility of 20 beans being stolen. For instance, bandits who cause the 100th bean to be stolen from a bowl are responsible for the consequence of 20 beans being stolen, since had they not caused the 100th bean to be stolen, only 80 beans would have been perceived as stolen.
This is the approach to take in describing the causal responsibility, after the fact, of agents in similar problems of contributory causation. However, as suggested above, this retrospective description of actual consequences does not help us to decide on a course of action.
For this, ethical individuals must combine the knowledge of thresholds with expected consequences. Imagine that the bandits are contemplating stealing beans again. This time, each bandit knows villagers can perceive only threshold units of 20, but each bandit does not know whether he will be stealing a 20th bean from each bowl. Under this uncertainty, each bandit ought to calculate the expected consequence of stealing 100 beans as the probability of completing a threshold unit in each bowl (1/20) times the consequence of perceiving that threshold unit (20) times the number of bowls (100), which equals 100 – one hungry villager.
Even if each bandit knows neither the size of the threshold unit nor which bean he is stealing, he can still calculate the expected consequences. In each case he will know that the consequences of reaching a threshold unit times the probability of completing a threshold unit in each bowl is one. (This is so because the size of the threshold unit and the probability of completing it always vary inversely.) Hence the expected consequence of stealing 100 beans will always be 100. The only condition under which the expected consequence will be les than 100 is when the Bandit has information about both the exact size of the threshold unit and the exact position of a particular bean within that unit. In most cases of contributory causation, this kind of information will not be available.
As a decision procedure, expected consequences yield the same prescription as Glover’s divisibility principle: don't steal beans. This makes sense, since the sum of all the bandits’ chances of completing a perceptible unit is one and the product of each of these probabilities is also one. One virtue of calculating expected consequences, then, is that it provides the same prescriptions as Glover’s divisibility principle but without a questionable view of actual causal responsibility.
Recognizing the expected consequences of an action, the “causal inefficacy” defense of buying meat no longer holds. There must be some threshold at which point a unit of meat demanded by some group of customers is perceived by the grocer. At the very most, the size of this threshold unit is the difference between the demand for no meat and the current demand for meat. Likewise, there must be some threshold where a unit of meat demanded by some group of grocers is perceived by the butcher. And so on, all the way to the farmer. The expected consequence of completing a threshold unit that affects the production and slaughter of animals is thus the product of al the probabilities of completing each threshold unit [p(Al)=p(Grocer)* p(Butcher) *…* p(Farmer)] times the consequence of that entire threshold unit of animal production. It is likely that the probability is quite small. However, the consequence of completing the threshold unit is the consequence of the entire unit, not some portion of it. This consequence is quite large and terrible, since it involves raising and slaughtering a significant number of animals.
For example, take the case of The 200 Million Consumers. There are 200 million consumers, each of whom eats 50 farmed animals each year. In this market, there are only ten possible annual outputs of animals for farmers: one billion animals, two billion, and so on, up to ten billion. The difference between each of these annual outputs – one billion – is the smallest unit of demand perceivable to the farmer and is thus the threshold unit. Since there are 20 million customers per threshold unit, and only one of these customers will actually complete the unit of which his other purchase is a part, the probability of my completing a unit is one in 20 million. That means by buying meat for the year, an individual has a one-in-20 million chance of affecting the production and slaughter of one billion animals. The expected consequence is then one-20-millionth times one billion, which equals 50 – that is, raising and slaughtering 50 animals per year. Given the horrors of today's animal agriculture, that is a substantial consequence. These hypothetical numbers are close to the actual numbers for meat production and consumption in the United States.
As with The 100 Bandits, in the case of The 200 Million Consumers, only a small fraction of individuals may actually cause harm, as determined after the fact. While at first glance this seems to weaken the argument against buying meat, on closer inspection it makes no difference. Since we can never have perfect knowledge beforehand, only a decision procedure can tell us whether or not we ought to buy meat. An ethical individual must thus use expected consequences to make a decision about buying meat, and the expected consequences of buying meat are terrible.