NOTE-160: Rule Consequentialism (SEP)

Contribution

A synoptic, argued survey of rule-consequentialism as it stood in 2023. It separates full from partial rule-consequentialism by three components (criterion of wrongness, decision procedure, conditions for sanctions; §4). It works through the formulation choices: actual vs. expected good (§6.1), compliance vs. acceptance vs. teaching (§6.2), complete vs. incomplete acceptance (§6.3). It sets out three arguments for the view (§7) and gives the defender’s replies to the classic collapse, incoherence and rule-worship objections (§8). It also lists the objections still open (§9). The entry adds no new theorem or argument. Its value is that the formulations are stated exactly and the replies are organised by the author who made most of them (Hooker 2000, Ideal Code, Real World).

Key insight

Rule-consequentialism is not “act-consequentialism with shortcuts”. It ranks codes by the expected good of teaching them to, and having them internalised by, real people. Those costs grow with a code’s number of rules, their complexity and their demandingness. A code with many exception clauses also invites self-serving misapplication and undermines others’ assurance. So the best code is simpler and less demanding than one that is optimal act by act, and it will sometimes require sub-optimal acts. That is why the view does not collapse into act-consequentialism. It is incoherent only if one assumes an overriding commitment to maximise that the theory does not contain.

Assumptions

Premises and authorities the entry relies on (non-formal; stated as the entry states them):

• Mill’s link (§6.1): “If an act is wrong, it is blameworthy” (from Mill 1861 ch. 5 ¶14), plus: blame requires that the rules’ sub-optimal consequences were foreseeable. From these the entry derives the case for expectablist over actualist criteria.
• Expected good is computed from “rational or justified probability estimates”, tied to evidence available at the time, not to agents’ actual credences (§6.1). The entry concedes such calculations will be “impressionistic and imprecise” (Lenman 2000 is sceptical).
• Human nature (§6.2): children start with natural partiality toward themselves and a few others; each rule learned carries a cognitive cost. The entry calls these “realistic assumptions” without empirical citation. It also reports, without endorsing, the Sidgwickian premise that strong and impartial concern for everyone is not a realistic human possibility.
• Transition costs are measured from new generations (§6.2), not from whatever code a society already holds. This avoids giving weight to unjustified existing codes (the sexist-code example) and avoids social relativism (the same code for “1960s Michigan as for 1960s Mississippi”). Kahn 2012, D. Miller 2014/2021 and T. Miller 2021 argue the opposite.
• Imperfect uptake (§6.3): internalisation “will not be achieved in every last person” (D. Miller 2021: 130).
• Method of justification (§7): reflective equilibrium / moderate foundationalism. Intuitive moral rules get independent but fallible credibility.

Key results

The formulations and arguments, with locations:

• Act-consequentialist criterion (§4): an act is wrong iff it results in less good than some available alternative. Nearly all consequentialists reject the matching decision procedure for four reasons: lack of information, cost of information, calculation error and bias, and expectation effects. The fourth is contested (Hodgson, Hospers and Harsanyi say trust would break down; Singer and Lewis say it would not).
• Partial rule-consequentialism = “indirect consequentialism” (§4): an act-consequentialist criterion of wrongness plus a rule-consequentialist decision procedure. Sidgwick, Moore, Hare and Pettit are placed here.
• Global consequentialism (§5; Pettit and Smith 2000, Kagan 2000) assesses everything directly. The entry charges it with paradoxical gaps between required act, recommended procedure and blamelessness (Crisp 1992; Streumer 2003; Lang 2004), and cites replies by Driver 2014 and de Lazari-Radek and Singer 2014.
• Sophisticated expectablist criterion (§6.1): an act is wrong iff forbidden by the rules with the greatest expected good, or, if two or more codes tie, by the tied code closest to conventional morality. The tie-break rests on the claim that social change’s unexpected consequences “usually seem to be negative”.
• Teaching costs decide between codes (§6.2, worked example): Code A has expected value n and internalisation cost −1; Code B has n+5 and cost −7. Net, A = n−1 beats B = n−2.
• Incomplete-acceptance formulation (§6.3): an act is wrong iff prohibited by the code whose acceptance “by the overwhelming majority of people in each new generation” has the greatest expected value. It is needed because a 100%-acceptance formulation would not endorse rules about punishment. About 90% is defended as a compromise. H. Smith 2010 shows the outcome depends on what the remaining ~10% accept.
• Teaching formulation (§6.3): acts are prohibited, obligatory or optional “iff and because” a code, “the teaching of which to everyone has at least as much expected value as the teaching of any other code”, prohibits, requires or permits them. Its advantages: no precise acceptance percentage is needed, and it ties to the publicity condition.
• Three arguments for the view (§7): (1) it best supplies an impartial justification for intuitively plausible rules (reflective equilibrium); (2) consequentialist premises plus the claim that indirect assessment produces better consequences; (3) a contractualist route (Harsanyi’s veil; Brandt; Parfit 2011; Levy 2013), suggesting “extensional equivalence of contractualism and rule-consequentialism”. Wolf 2016 and Copp 2020 add a route from the view that morality is a social practice with a function.
• Collapse answered (§8) by ranking codes by teaching and acceptance, not compliance. Incoherence answered by denying that either the agent’s moral psychology or the theory contains an overriding maximising commitment. The agent’s fundamental motive can be “to do what is impartially defensible”. Rule-worship answered by a “prevent disaster” rule (Brandt 1992), which is triggered only by “a very much larger difference” in expected value and so does not re-collapse. Woodard 2022 recasts it; Woollard 2022 accepts its vagueness.
• Open objections (§9): real-world demandingness (Carson 1991) and possible-world counterintuitiveness (Mulgan 2001); necessary vs. contingent justification of rules; justification by the wrong facts; the incoherent description of the teachers; conflicts among rules (Eggleston 2007).

Claims

id	claim	strength	support
C1	Consequentialists nearly all reject the act-by-act calculating decision procedure	moderate	long citation list in §4 (Mill, Sidgwick, Moore, Smart, Hare, Parfit, Railton, Pettit …); the expectation-effects reason is conceded to be contested
C2	If wrongness entails blameworthiness, the criterion of wrongness must be expectablist, not actualist	moderate	two-premise informal argument in §6.1; the first premise is taken from Mill and not defended further
C3	Ties in expected value should be broken by closeness to conventional morality	weak	informal argument (§6.1) resting on an unsupported empirical generalisation that unexpected consequences of social change “usually seem to be negative”
C4	Teaching and internalisation costs favour simpler, less demanding codes	moderate	informal argument plus a toy numerical example (§6.2); the “realistic assumptions” about human nature are asserted, not evidenced
C5	Transition costs should be computed from new generations, not from a society’s existing code	moderate	two arguments (§6.2): no weight to unjustified codes, and avoiding relativism; opposing literature cited but not answered
C6	A code accepted by about 90% is a defensible target	weak	explicitly a “reasonable compromise” between two pressures (§6.3); the entry admits the arbitrariness and cites a dissent (Ridge 2006)
C7	Rule-consequentialism does not collapse into act-consequentialism	moderate	informal argument from teaching costs, temptation and assurance effects (§8); convincing given the acceptance/teaching formulation, which it presupposes
C8	Rule-consequentialism is not incoherent	moderate	conceptual argument (§8) that neither agent nor theory holds an overriding maximising commitment; conceded “less viable” if the view is argued for from consequentialist premises
C9	The “best known objection … dies”	weak (as stated)	the headline sentence in §8 outruns the caveat that immediately follows it; holds only on the reflective-equilibrium justification
C10	A prevent-disaster rule answers rule-worship without re-collapse	moderate	informal argument (§8); the entry concedes vagueness about what counts as a disaster
C11	Contractualist arguments suggest extensional equivalence with rule-consequentialism	weak	reported (“such arguments suggest”), with assessments (J. Ross 2009; Nebel 2012; Hooker 2014) cited but not summarised

Concepts

• Full rule-consequentialism — rule-consequentialist answers to all three questions: what makes acts wrong, which decision procedure to use, and when sanctions are appropriate (§4).
• Partial rule-consequentialism / indirect consequentialism — an act-consequentialist criterion of wrongness plus a rule-consequentialist decision procedure (§4).
• Global consequentialism — every evaluand (acts, rules, motives, sanctions) assessed directly by its own consequences (§5).
• Expectablist vs. actualist — rules ranked by rationally expected good vs. by actual good (§6.1).
• Beyond-compliance consequences — effects of accepting a rule other than compliance, such as assurance, incentive and deterrence (§6.2, after Kagan 2000).
• Transition / teaching costs — the costs of teaching a code and of recipients internalising it (§6.2).
• Teaching formulation — rules ranked by the expected value of teaching them to everyone, without assuming full internalisation (§6.3).
• Publicity condition — a moral code must be suitable for being public knowledge (§6.3).
• Rossian pluralism — the intuitive rules are the most basic level of morality, with no unifying principle (§7).
• Rule-worship — sticking to rules even when that is obviously disastrous (§8).

Connections

The entry sits in the consequentialist family (SEP “Consequentialism” is the related entry) and argues against Rossian pluralism, act-consequentialism, global consequentialism and, obliquely, Kantian and contractualist rivals, while treating contractualism as a possible ally (§7). The held work closest to it is LIT-186 (Pettigrew 2025, “Wolfian epistemology”). That work transposes Susan Wolf’s rule consequentialism to epistemic norms, and this entry cites the relevant Wolf paper (“Two Concepts of Rule Utilitarianism”, Oxford Studies in Normative Ethics 6, 2016) twice: in §7, for the functionalist route to rule-consequentialism, and in §8, on the incoherence reply. It also confirms that the Susan Wolf in LIT-186 is the same Susan Wolf whose meaning-in-life view LIT-183 discusses. The entry reports Hooker’s reply that the functionalist route itself needs reflective-equilibrium justification.

Ethics tag. The entry is squarely normative ethics. Its questions are what makes acts morally wrong, when blame and punishment are appropriate, and which moral code is impartially justified. Its position is Hooker’s own: a full, expectablist rule-consequentialism based on teaching and acceptance, justified mainly by reflective equilibrium. The tag is justified, and ethics is the right primary topic.

Bearing on the record

It carries nothing for ML practice. No practice, theory or instruction for training or evaluating models follows from it, and nothing belongs in the Anthology of the SOTA. The analogy the skim flagged has two parts. The first is rule-based vs. outcome-based specification of AI behaviour. The second is imperfect “internalisation” of a code by the agents taught it. The entry does not draw this analogy. Its specific mechanism, the cost of teaching children many, complex or demanding rules, is a claim about human moral education, so anyone using the analogy should mark it as their own. Within nucleation, it is the natural background reference for LIT-186 and should be read before that work.

Limitations

• It is an advocate’s survey. The author is the view’s main contemporary defender, and the §8 replies are largely his. Critics are cited more often than they are stated, especially in §9.
• The central empirical premises are asserted, not evidenced. They concern children’s partiality, per-rule cognitive costs, the Sidgwickian limits on impartial concern, and the claim that unexpected social change is usually bad.
• The ~90% threshold is admitted to be arbitrary. The choice between the “overwhelming majority” and teaching formulations is left open.
• The §8 claim that the incoherence objection “dies” is stronger than the entry’s own caveat allows.
• §§1–3 (welfare, other goods, prioritarianism) are background shared with consequentialism generally, and the entry explicitly sets the prioritarianism debate aside.

Open questions

• Can the teaching formulation describe its teachers coherently (the Andrews/Ehman/Moore objection)? A formulation that fixes the teachers’ own code without reintroducing transition costs would settle it.
• How should rule-consequentialism resolve conflicts among rules (Eggleston 2007), and where exactly is the “disaster” threshold for each rule it overrides?
• Does rule-consequentialism’s acceptable demandingness in the actual world survive Mulgan’s possible-world cases?
• Is the distinction between decision procedure and criterion of rightness legitimate? The entry calls it “contentious” (§7), and the second argument for the view turns on it.

Corrections to the seeded skim

• The summary on LIT-198/NOTE-160 says the view “survives” the objections “provided it is formulated in terms of acceptance by less than everyone and includes a prevent-disaster rule”. The entry is narrower. It is the teaching-and-acceptance formulation (rules ranked by the expected good of “their teaching and acceptance”, §8) that answers collapse, and it is the absence of an “overriding commitment to maximize the good” in either the agent or the theory that answers incoherence. The prevent-disaster rule answers only rule-worship. Acceptance by less than everyone (§6.3) is motivated by the need for rules about punishment, not by the §8 objections. The entry also concedes the incoherence reply “seems less viable” if rule-consequentialism is argued for from consequentialist first premises (§8), so “survives” holds only on the reflective-equilibrium route.
• The skim says “acceptance by less than everyone” as if that were the entry’s settled formulation. §6.3 presents two live options: the “overwhelming majority of people in each new generation” formulation, with 90% defended as “in a defensible range” against the charge of arbitrariness (Ridge 2006 dissents), and a “teaching formulation” (Mulgan, T. Miller, D. Miller 2021), to which the entry credits “definite advantages”. It does not choose between them.
• The skim puts “the problem of teaching new generations” in §9 without its content: the objection (credited to Andrews, Ehman and A. Moore) is that no coherent description of the teachers is available. Teachers who have already internalised the ideal code beg the question, and teachers who have not reintroduce the transition costs the new-generation formulation was designed to exclude.
• The skim omits §9’s contingency objection (McNaughton and Rawling 1998; Gaut; Montague; Suikkanen 2008): some rules are necessarily, not contingently, justified. It also omits its “wrong facts” sibling (Arneson 2005; Portmore 2009).
• The skim reports Mulgan 2001 only as a demandingness critic. The entry says his argument is about counterintuitive implications “in possible worlds”, and that Mulgan “has become a developer of rule-consequentialism rather than a critic”.
• The dossier’s Open question about “§8’s collapse argument [as] the canonical case of a rule-based policy degenerating into case-by-case optimisation” misses the point. The entry’s reply is that the collapse fails once rules are ranked by the costs of teaching and internalising them plus the assurance and temptation effects of exception clauses. That reply is specific to human moral education, so any analogy to specifying AI behaviour is the reader’s own; the entry does not draw it.
• ethics tag: justified — normative ethical theory: what makes acts wrong, when blame is apt, and which moral code is justified.
• primary topic: ethics is right; no other held topic fits better. society-and-governance would be a defensible second tag only for §6.3 on punishment, and it is not recommended.