Yes, rule consequentialism is incoherent, in the sense that all possible formulations of the theory either collapse into act consequentialism or fail to be consequentialist at all. Proponents of rule consequentialism often appeal to concepts like the costs of accepting, internalising, or teaching rules in an attempt to avoid this dilemma, but these strategies are unsuccessful, and moreover introduce additional incoherence into the theory by requiring a specific compliance rate at which sets of rules are evaluated. In this essay, I first define a simple form of rule consequentialism and explain why it falls victim to the dilemma described. Then, I outline a modified version of rule consequentialism which supposedly avoids the dilemma, and demonstrate that it is, in fact, equally vulnerable. Finally, I argue that since rule consequentialism (henceforth RC) is less coherent than act consequentialism and not particularly distinctive, we have extremely limited reason to adopt it.
The criterion of rightness for a simple version of RC can be expressed as follows:
(RC1): An action X is right iff and because it is permitted by a set of rules which, if universally followed, would lead to the best possible consequences among all sets of rules.
Note that RC1 is a maximising theory: we focus on rulesets which lead to the best possible consequences (there may be ties). Satisficing versions of RC also exist, where we replace “best possible consequences” with “consequences at least as good as M”, where M is the minimally-morally-acceptable outcome – but these are subject to the same objections as I present here for maximising RC, and face additional objections regarding gratuitous harms, so I set them aside in what follows. Also note that our definition of RC1 does not explain how consequences are to be ranked, even though any fully-specified consequentialist theory must provide an axiology. Again, the arguments I will present do not rely on RC taking any particular theory of value, so I will leave the axiology implicit. In certain places where concreteness is required, I will use an impartial welfare-based axiology (i.e. investigate rule utilitarianism in particular), but the difficulties encountered apply mutatis mutandis to all RCs.
As outlined above, RC1 faces a dilemma between (i) collapse into act consequentialism or (ii) failure to be a consequentialist theory. We can show this as follows:
-
Suppose that RC1 is in some situation extensionally different from act consequentialism (the theory that an action X is right iff and because it leads to the best possible consequences among all actions open to an agent).
-
Then, there is an action X which our ruleset R says is permissible, but does not lead to the best possible consequences among all available actions (otherwise it would be recommended by act consequentialism too).
-
But then our ruleset R permits actions which have foreseeably worse consequences than some other ruleset R’ which simply told agents to take the action recommended by act utilitarianism in each situation.
So RC1 is incoherent. Either it is extensionally identical to act consequentialism, or it recommends foreseeably suboptimal actions, despite the criterion of rightness specifying that the relevant ruleset is the one which does lead to the optimal consequences.
An initial reply from the rule consequentialist might be to say that even if RC1 is extensionally identical to act consequentialism, that does not mean it’s redundant. RC offers a different explanation of what makes an action right than act consequentialism does: the right-making property is conformance with an optimal ruleset, rather than production of optimal consequences. But this merely takes us back to the second horn of the dilemma. If the rightness of an action really is grounded in the fact that it complies with a particular set of rules, then RC seems to be more concerned with rule-compliance than consequences. Conversely, if the rule consequentialist appeals to the definition of the ruleset in terms of optimal outcomes, they have introduced a layer of indirectness (compared to act consequentialism) without any explanatory value.
Rule consequentialists like Hooker are aware of this objection, and have tried to avoid it by appeal to the so-called “acceptance costs” of a ruleset. A ruleset which just encodes the act-consequentialist-endorsed action in every scenario would be impossibly complicated for moral agents to internalise, so (they argue) we must take into account the costs of individuals in society fully internalising the rules when arriving at the optimal ruleset. This leads us to the following modified criterion of rightness:
(RC2): An action X is right iff and because it is permitted by a set of rules which, if universally accepted, would lead to the best possible consequences among all sets of rules.
According to Hooker, RC2 no longer collapses into act consequentialism, because the costs of internalising such finely-grained rules offset the benefits that would obtain from people behaving exactly optimally in each situation. Seen in this light, RC avoids the charge of incoherence: it is a distinctive theory, focused on promoting the best consequences, but sensibly accounting for the bounded rationality of human agents.
However, there are three problems with this approach. First, act consequentialism is also perfectly capable of accommodating the fact that it would lead to worse consequences were agents to attempt to calculate the optimal action in every scenario, by standard multi-level consequentialism. A theory’s criterion of rightness need not be the same as the decision procedure it recommends to agents, and for act consequentialism the two come apart: it endorses agents relying on heuristics and general principles to reach decisions, precisely because it would be infeasible (and suboptimal) for them to attempt to rederive the correct action from first principles. So RC2 does not really seem to provide anything extensionally distinctive. Second, and similarly to before, it is unclear why we should think that the fact of being permitted by a set of rules with a particular property when universally accepted is what makes an action right, as opposed to the action simply being the consequentially-optimal one.
The third problem is that acceptance-based theories of rightness like RC2 require the rule consequentialist to specify a compliance level at which theories are to be evaluated. Since the costs of internalising a theory count against it, RC2 will pick out the minimal ruleset at 100% acceptance – but given that this world of universal acceptance by definition contains no rule-breakers, RC2 will not, for instance, have rules permitting the punishment of criminals. To address this (since intuitively, it is permitted to punish criminals, and moreover may even be obligatory), the rulesets must be evaluated at an acceptance level below 100%. Hooker suggests 90%, but this figure seems arbitrary – surely we could equally-well say that the permissible actions are those permitted by the 89%-optimal ruleset, and it might well be that the rules differ between these.
More fundamentally, the fact that an action X would be recommended by the optimal ruleset in a world with p% acceptance seems like it has very little bearing on whether that action is right to take in our world. If we can predictably foresee that a different action X’ leads to better consequences in our world, then surely we ought to take X’ instead? Yet fixed-rate rule consequentialism seems stubbornly attached to what is optimal in another non-actual world, and objectionably detached from actual consequences for a supposedly consequentialist theory. And setting the fixed-rate to the actual rate of acceptance would merely collapse into act consequentialism.
So, to conclude, rule consequentialism is incoherent because it either collapses into act consequentialism, or fails to be a properly consequentialist theory. Although there are other formulations of rule consequentialism I have not had time to discuss here, involving variable rates of compliance, they face the same fundamental problem that they define rightness in terms of optimality in a non-actual world, leading to foreseeably worse consequences than could otherwise be attained. For this reason, we should reject rule consequentialist theories.
Post-Claude & examiner report notes
- Touch more on what it means for a theory to be incoherent.
- Logically contradictory?
- No, because it at least has the collapse horn to take.
- Self-undermining / unstable in its motivation?
- Yes. It appeals to consequences to select the code, but then ignores them at the act level.
- Be clear that the complaint is not that it’s not act consequentialism.
- Logically contradictory?
- Can be a bit more concessive towards Hooker – remember that he acknowledges RC doesn’t lead to the best consequences, but he motivates it by reflective equilibrium.
- As in, he thinks it matches our intuitions better than other theories. And he admits that it isn’t consequences-maximising.
- The pushback here would be that (a) not so clear that it really does fit our intuitions (e.g. coordination games cases), and (b) if the rationale is fitting intuitions, why not drop talk of consequences entirely?
- From the literature: it’s Smart who complains about rule-worship; Lyons who brings the collapse objection.