Consequentialism and demandingness

In what sense, if any, is consequentialism alienating? (2000 Q4)

• Three different versions of the alienation objection:
  • (1) Integrity objection (Williams) – can gloss it as: consequentialism undermines your relationship with your own personal projects and identity
  • (2) Alienation in relationships with other people (Stocker, Wolf)
  • (3) Consequentialism undermines your relationship with your own agency
• Consequentialists can address (1) and (2) relatively easily with sophisticated consequentialism, though residual objections remain; (3) is more challenging to rebut.
• (1) Integrity: George the chemist
  • However, a utilitarian George wouldn’t be acting against his convictions by taking the weapons job! There’s only integrity-violation if you think he is non-utilitarian.
  • A reformulation: George lacks distinctive projects as a utilitarian, and that is a problem.
• (2) Relationships: visiting friend at hospital
  • “One thought too many” – the multi-level utilitarian justification goes something like “I should visit my friend because that promotes friendship, and that’s (heuristically) something to promote the good”.
• Stocker and “moral schizophrenia”
  • The complaint is that the multi-level utilitarian exhibits a worrisome disharmony between their justifications and motives.
    • The theory’s justification for visiting a friend is that it promotes utility.
    • But the motivating reason for the agent is something like love for their friend.
  • Stocker thinks a correct moral theory should let you live by it in an integrated way – i.e., the considerations that make your actions right are also the considerations which move you to act.
• One way you might try to address this: Scheffler’s agent-centred prerogatives. But these seem to fail and aren’t really consequentialist.
• Sophisticated consequentialism (Railton) can address this: adopt non-utilitarian goals.
  • Multi-level utilitarians distinguish between criterion of rightness and decision procedure, licensing use of e.g. rights-based heuristics to make decisions. But all they intrinsically value is the utilitarian good.
  • Sophisticated utilitarians go further, and adopt non-utilitarian goals/desires when this would have good results.
    • So, at the point of deliberation, there’s no gap between justifications and motives.
    • And adopting friendship / personal projects as goals does seem like it would be better on utilitarian grounds, given facts about human psychology!
• A problem with sophisticated consequentialism: genesis and conditionality.
  • From the inside and the outside, a sophisticated consequentialist looks identical to someone who straightforwardly intrinsically values those goods.
  • But the sophisticated consequentialist adopted those intrinsic goals only because they promoted utility. If valuing friendship ceased to promote utility, they’d drop it.
  • And moreover, they seem to have their motivations (valuing friendship) for the wrong sort of second-order reasons (to promote aggregate welfare).
• Chappell can partly address this with a “subsumption” strategy, based on his value receptacles argument.
  • Key idea: overall wellbeing matters only because each particular individual matters.
  • So concern for the overall good is instead built up out of concern for each individual. And you ought to go to visit your friend at the hospital because it would cheer him up – i.e., grounded in his individual welfare.
  • This doesn’t address Williams’s integrity objection so well, because e.g. sports & hobbies don’t have any other subject whose welfare you promote by engaging in them.
    • But this seems OK – intuitively sports & hobbies do just have instrumental value, as means to happiness / friendship.
    • And so sophisticated consequentialism (i.e. intrinsically valuing the sport, adopted because this will help utility-maximisation) appears less objectionable.
• Global utilitarianism might help separately: it says you can evaluate motives (and all other objects of interest) by utilitarian standards, not only acts.
  • So we can say that the motive of direct concern for your friend is itself good, which seems intuitive.
    • But still, it’s good only conditionally, because of consequences.
  • On hybrid utilitarianism, you can remove the conditionality: motive of direct concern can be good e.g., because of friendship.
    • But then you’ve given up on consequences as the sole standard of moral value.
    • If you’re committed to consequentialism, then you will just be OK accepting that ultimately your justifications need to bottom out at consequences.
• (3) Negative responsibility: alienation from your own agency
  • Williams objects that because consequentialism denies the doing/allowing distinction, it absurdly implies that you’re just as responsible for failing to mitigate omissions of other agents as you are for your own actions.
  • It’s true that utilitarianism can’t really get rid of this sense of alienation.
  • But this can be turned on its head like Sobel: other theories rob the less-fortunate of their agency. Cf also Ashford on the global poverty emergency.
• Insofar as it is alienating, that needn’t count against it too much.
  • Williams thought Kantianism could also be alienating; to some extent all moral theories will make demands of us on our agency, etc – so it’s not a weighty objection.

Is it an objection to a consequentialist theory that it cannot be used as a guide to action? (2001 Q5)

• If consequentialist theories were indeed unusable as a guide to action, then this would be an objection to them, insofar as most consequentialists aim for their theories to guide action. However, lack of action-guidingness would not necessarily count against the truth of consequentialism – and in any case, such objections are largely unjustified, in that consequentialism seems hardly less action-guiding than any other major moral theories.
• Many flavours of consequentialism, but they all share in common the claim that the rightness of an action is solely a function of the value of the consequences it produces, and moreover that this value is what makes the actions right.
  • Maximising act-utilitarianism, for example, says that an action is permissible iff and because it leads to the outcome with the greatest aggregate welfare among all open to an agent.
  • For specificity, we’ll focus on maximising act-utilitarianism (henceforth “utilitarianism”), but arguments and replies apply to other consequentialist theories mutatis mutandis, except for where indicated.
• The putative problem:
  • 1. The consequences of an action extend into the future by thousands of years, or more
  • 2. Nobody can (or could) foresee all those consequences
  • 3. So nobody could determine what the optimal action would be, in terms of consequences
  • 4. So utilitarianism is not action-guiding.
  • 5. So utilitarianism is a bad ethical theory.
• For 2, we actually need something a little stronger – more like “… and these consequences continue to be morally-relevant”.
  • If you’re a pure-time discounter with high discount rate, then you might be able to loosen things a bit, but not really – still ex ante you can’t figure out what’s going to be optimal since there’s uncertainty even over very short periods
• Relaxing the maximising constraint does not really help, although it loosens informational requirements a little.
  • E.g., if we adopt satisficing act-utilitarianism, we no longer need to find the best action – but still, one must calculate which action is above the threshold of goodness.
• The step from 3 to 4 is open to challenge.
  • For a theory to be action-guiding, what we really want is something like: it provides a decision procedure capable of use by agents, which reliably recommends actions in accordance with the theory’s criterion of rightness
  • We needn’t expect that the decision procedure perfectly tracks the criterion of rightness.
    • So if the consequences average out either way, then cluelessness doesn’t matter much anyway, in expectation we’ll do the right thing.
    • And as Bentham, Mill, and virtually every utilitarian since argued, we can use heuristics as part of that decision procedure, meaning it is actionable.
• Now consider the step from 4 to 5. Generally consequentialists do want their theory to be action-guiding (e.g., recommending you to donate to animal welfare charities, or work on reducing existential risks). So on their own account it would be an objection.
  • But in principle there’s little reason to think the true ethical theory is easily action-guiding.
  • And indeed, looking at other theories they often aren’t action-guiding
    • E.g. uncodifiability of virtue ethics
    • Or impossibility of obtaining the set of principles to which nobody could reasonably reject in contractualism.

Post-Claude & examiner report notes

• Spend more time on the step from 4 to 5, i.e. whether, conditional on being true, non-action-guidingness is an objection.
  • And make the criterion vs decision procedure point more obvious, it’s the core of what we need
  • Likewise can mention the objective vs subjective rightness (i.e. ex post vs ex ante)
  • But might drop the satisficing aside.
• Some different senses of action-guidingness:
  • An algorithm which given inputs will actually yield a verdict
  • A reliably correct algorithm
  • Something which recommends heuristics / rules of thumb that are approximately correct
• Other action-guidingness worries, besides cluelessness
  • Computability (i.e. why we need to appeal to heuristics)
  • Regress (how do we know what is the decision procedure, if not criterion of rightness?)
    • Maybe we can generalise from empirical observations about patterns, e.g. in general honesty seems to result in better consequences than lying
    • We can also use reflective equilibrium with specific cases
• Note that Kantian deontology with universal law formulation might appear algorithmic, and in general rule-based theories do better here. But they have problems too.
  • It’s very indeterminate which maxim describes an action (at what level should you individuate?), and they might all universalise differently.

(a) ‘There exists no plausible theoretical account of how we could fail to be obligated to bring about the best available outcome whenever doing so is permissible. Therefore, the objection that consequentialism is too demanding is ineffective.’ Discuss. (2019 Q8)

• Yes, it’s true that there’s no plausible theoretical account, though this isn’t the main reason why the demandingness objection is ineffective. The crucial flaw in it is that it presupposes controversial assumptions analogous to the doing-allowing distinction which consequentialists independently reject.
• Such a theoretical account is one of supererogation: an explanation of how there can exist actions which are (i) admirable, and better than some other permissible actions, but (ii) not required. As Kagan notes, many accounts of supererogation succeed on one while failing on the other.
• Nagel on agent-relative vs agent-neutral reasons.
  • Idea: pain generates agent-neutral reasons, but pleasure (e.g. from being a pianist) generates agent-relative reasons only. So if you are weighing up your desire to play piano against another’s non-essential desire to, say, have a tidied garden, then it’s just agent-relative reasons at play.
    • The problem: doesn’t explain why you are allowed to help your friend play piano rather than tidy their garden.
• Scheffler’s agent-centred prerogatives. His goal is to permit supererogation without appealing to a doing/allowing distinction.
  • You’re permitted (but not required) to upweight benefits & costs to yourself by a factor M, when weighing them against benefits & costs to others.
    • Setting M=1 gives impartial consequentialism; M=infinity gives egoism.
    • One issue is that it’s difficult to conceptually motivate why you’re not required to upweight the prudential effects. But the thought might be: I have a special relationship with myself; I can’t willfully harm myself and so I can waive the flexibility if I want.
    • The problem is that without doing/allowing distinction, any threshold of M will seem to be overly permissive. Rich Uncle.
• This is a microcosm of the wider issues with demandingness objection: rests on doing/allowing distinction.
• Sobel, the usual stuff on compliance costs.
  • If you take a rules-based morality with doing/allowing distinction, then it will deliver supererogation! It lets you fail to improve outcomes even at small cost to yourself.
  • But this seems implausible, because it contradicts, for example, Singer’s principle. And in any case, it’s not really about demandingness so much as how we count costs.

Post-Claude & examiner report notes

• Rather than objecting on the “presupposes other assumptions” point, you could disagree by denying the implicit premiss in the question that an objection needs a worked-out theory to be effective
  • [But of course it just is a bad objection.]
• Some non-consequentialist accounts of options & permissions worth having handy:
  • Kant’s imperfect duties – it’s up to you how much and when to help.
    • But you’re required to adopt others’ happiness as an end; it’s just discretionary how to fulfil that end
  • Murphy’s fairness – you need to do what’s demanded of you when everyone does their fair share; more is optional

/ (b) ‘There exists no plausible theoretical account of how it could be wrong to harm others if we thereby bring about the best available outcome. Therefore, the objection that consequentialism is too permissive is ineffective.’ Discuss. (2019 Q8)

/ Does consequentialism pose a threat to individual rights? (2002 Q7)

“Since just about all consequences of actions are unforeseen and unforeseeable, we are ignorant about which actions to perform if consequentialism is true.” Is that so? What does your answer suggest about the truth of consequentialism? (2020 Q7)

• We are ignorant about which actions will turn out to be right if consequentialism is true, but that doesn’t mean consequentialism fails to enlighten us about which actions we ought to take.
• Are almost all consequences unforeseen and unforeseeable? Yes, that’s quite plausible; cluelessness about the long-term future.
  • Why does the unforeseeable part matter – it’s that even if we got better at forecasting etc, we couldn’t determine the right action, so even in an epistemically ideal situation it would be a problem
• What does “which actions to perform” mean? → It’s meant to be picking out the actions which are right, i.e. those which maximise value (whatever the axiology takes it to be).
• So assuming consequentialism to be true, we would be ignorant about which action will ex post have maximised value.
• But, we can still make decisions! Using ex ante consequentialism
  • In particular, we can appeal to heuristics that have been developed by observing the past and how they play out, then applying to present decisions
• Different purposes of a moral theory: one thing we care about is extensional correctness, and another is providing a guide to action.
  • If consequentialism were unable to provide action guidance, that would not necessarily bear on its truth – there’s little reason to think a priori that the true ethical theory will be easy to follow or comprehend.
  • But it would still count against the theory overall, given that one of its goals is to provide action guidance.
  • However, the theory can provide action guidance, as described
• The bigger problem is that it’s unclear how we would verify whether the decision criterion and criterion of rightness do really align in expectation.
  • Also, there’s some inelegance to the fact that they are different, but it seems perfectly sensible that the method of judging actions is different to the method of selecting them in advance.

Post-Claude & examiner report notes

• One point you can make: the consequentialist might actively want to have a separate decision procedure from criterion of rightness.
  • This is because of the alienation/integrity objections to consequentialism, and self-effacingness-style defences.
    • Cf Railton’s sophisticated consequentialism, where the agent might adopt desires and motivations other than purely consequentialist ones, insofar as this promotes the best consequences.
  • So there’s independent reason for them to distinguish between the two; it’s not just an ad-hoc reply to the epistemic problem.
• Tu quoque – deontologists and virtue ethicists also have epistemic limits.
  • E.g., “does this action violate a perfect duty?”; “would the phronimos do this?”
• Defend the premiss a bit more that almost all consequences indeed are unforeseen and unforeseeable

Can rule-consequentialists specify a compliance level at which candidate rules are to be assessed without falling prey to decisive objections to their view? (2022 Q9)

• No, they cannot.
• Rule consequentialism: an action X is right iff and because it is licensed by a set of rules S which, if they were {complied with / accepted} by p% of the population, would make the outcome as good as possible.
• Rule-consequentialists need to specify (i) an uptake level p, and also (ii) whether they are concerned with compliance or acceptance.
• Start with a simple view: let p=100%.
  • Consider a rule like: drive on the right-hand side of the road.
  • Now, it might be that everyone driving on the right-hand side of the road is better than everyone driving on the left-hand side of the road, because we’re mostly right-handed and it reduces the incidence of accidents when people have their right hand available to change gears rather than left hand.
  • But evidently it’s not the case that it is right in the actual world for you to drive on the right-hand side of the road (indeed, it would be wrong…)
• Hooker favours a lower compliance level, like 90%. But this fails the driving example above, too.
  • Maybe you think majority compliance is important. But here’s another example, decidedly moral: if there was 50% compliance with a set of rules mandating you to donate 1% of your income to highly effective charities, then we would solve global poverty. So then, it would also be permissible under the rules to spend considerable money on yourself, e.g. to buy a private jet, or holiday villa.
  • However, it’s at least plausible (per Ashford and Singer) that it isn’t permissible currently to spend considerable money on yourself.
• More broadly, this shows the problem with rule-consequentialist views: given that the actual world will undoubtedly not have the same compliance level as any one theoretically justified, we can always construct cases where the outcome recommended by rule-consequentialism is worse than another attainable outcome.
  • And then, it seems like rule-consequentialism is not consequentialist at all: it’s overly concerned with a set of rules and arbitrary compliance level, disconnected from the actual world we’re in.

Post-Claude & examiner report notes

• For the driving rule, the rule-consequentialist can reformulate: the rule would be “drive on the locally conventional/legal side”.
  • Although, it might really be that it’s just better for everyone to drive on the right!
  • Also, the theory then doesn’t actually tell you which side to drive on, if you’re starting de novo.
    • There’s some wrinkles about rule-individuation here.
  • Maybe the better takeaway is: different sorts of rules seem like they ought to be evaluated with reference to different compliance levels. But this is pushing us towards a very complicated, and collapsing, theory.
• Remember the 10 Strangers example:
  • If nobody presses a button, then 1000 people saved. But if someone presses a button, then for each person who does not press their button, someone will die.
  • So it would be best for nobody to press.
  • But at 90% compliance, we’d have 1 person pressing a button, so then everyone ought to press.
• Some other compliance levels they might try to stipulate:
  • The actual compliance level. But then it really does collapse into act-utilitarianism (and moreover seems to be a bit circular/recursive as a guide to action)
  • An optimised compliance level: i.e. something which varies by rule, or the threshold which, among all possible thresholds, maximises welfare.
    • Trouble is it either collapses, or is pretty incoherent/arbitrary as a theory
• Here’s a way to fit in the incoherence vs collapse dilemma: the compliance level is the feature which the rule-consequentialist uses to try and be both consequentialist without act-consequentialism. But their choice of compliance level is directly subject to the incoherence or collapse dilemma.
  • If they choose the actual compliance level, it collapses.
    • Maybe they say: people comply with different rules at different rates, so we have a gap vs the “actual actual” compliance level.
    • But then arbitrariness bites; else they definitely have collapsed.
  • If they set it to any idealisation, then the rules can demand acts which foreseeably make the outcome worse than otherwise attainable – i.e. rule-worship.
  • So, no threshold can escape a decisive objection.
• One motivation for evaluating at a less-than-100% compliance level: this is when you have wrongdoers, so it needs to contain rules for preventing disaster and suchlike.
  • At 100% compliance, the ruleset doesn’t need to have anything to say about punishing wrongdoers. But presumably the correct ruleset does.
  • Note that this wrinkle only appears if you adopt acceptance RC, not compliance RC.
    • For compliance RC, the ruleset can just (costlessly) specify what you would do were someone to not comply. [Just like in game theory a strategy profile is a complete contingent plan of action (incl off-path), the ruleset can specify what would happen if someone were to deviate.]
    • But for acceptance RC, there are costs associated with rules being internalised. So we’d have the minimal ruleset – which means that at 100%, you need to trim out all these never-acted-upon rules.
  • Also, for the variable-rate views below – the reason that they don’t just settle on 100% compliance as optimal is because they’re for acceptance RC views.
    • The marginal costs of internalisation increase as your acceptance rate rises (it’s hard to get the very last, most stubborn & amoral person to internalise a rule).
    • So if you’re doing maximax / similar, the EU net of internalisation costs is going to be somewhere below 100%.
  • For these non-compliers, we need to make assumptions about how they behave. Are they defiers? Do they act randomly? Etc. The ideal code is presumably sensitive to this.
• Individuation of rules goes hand-in-hand with uptake level.
  • Recall that the whole point of acceptance RC is to avoid the initial collapse objection, that the ruleset would be maximally specific and fine-grained.
    • Separately, they might fall onto the incoherence horn – is the rule against lying, or lying to save a life, or lying on a Tuesday when it’s raining? It’s indeterminate which rule applies.
  • One cost is that we then concede that the world is better under non-full compliance.
  • But also, more fundamentally, internalisation cost is exactly what makes the ideal code sensitive to the uptake level we evaluate at, as described above! Otherwise, we’d just find the ideal code at 100% compliance.
    • So avoiding immediate collapse into act-consequentialism is what creates this problem of specifying the uptake level.
• Ridge comes up with variable-rate rule consequentialism:
  • An action is right if and only if it would be required by rules which have the following property: when you take the expected utility of every level of social acceptance between (and including) 0% and 100% for those rules, and compute the average expected utility for all of those different levels of acceptance, the average for these rules is at least as high as the corresponding average for any alternative set of rules (Ridge, 2006, 248).
  • But there are arbitrariness issues here too.
    • Why should we take a uniform average over [0,100] compliance? Surely some compliance levels are more likely to obtain than others.
    • Also, maybe we should choose the code with the largest median EU across different levels of compliance, or the greatest total EU at its argmax level of compliance (though presumably that’s just 100%), or the code with the greatest EU at its least favourable compliance level (robustness).
  • Indeed, Holly Smith has a theory that does just search over all compliance levels and choose the welfare-maximising ruleset according to each ruleset’s preferred compliance level
    • An ideal code is the code whose EU at the optimum acceptance level is no lower than that of any alternative code (Smith, 2010, 418).
    • The problem here is that it’s rather unlikely that this particular optimum level obtains. And maximax doesn’t satisfy EU (fails independence).
  • So, to resolve all these issues we might turn to Maximizing Expectation Rate RC, which works with expected utilities at each acceptance level, weighted by the probability of that level obtaining.
    • This is a very complicated machinery! We need a prior over compliance levels for every possible rule-set…
• Overall, some of my own reflections:
  • The increasing amounts of formalism that rule-consequentialists turn to seem like they’re doing ethics backwards.
    • These are more like ad-hoc epicycles than anything converging on a true ethical theory.
    • There’s no normative justification offered for any of these moves; they’re all to address technical problems with an earlier formalisation.
    • And it is really dubious that the right-making feature of actions is that they’re permitted by the code with the highest probability-weighted acceptance-welfare integral. Why should that have any normative force on me at all?
  • Also, variable-rate RC seems to be overly sensitive to imperfect agents when delivering verdicts about what is right and wrong.
    • Take a world where agents are diligently moral, vs a world where people hardly ever comply with large demands. In each case they’ll get a tailored set of rules, and they’ll be laxer in the latter world. But it seems like we shouldn’t be able to shirk obligations by being less good morally.
    • The problem is that moral rightness is defined by whatever the ideal code dictates, and that is in turn a function of how morally-diligent agents are.
    • For act-consequentialism, although the optimal act is world-relative (I should take the best action given how others actually behave), the standard of rightness is fixed. So we can say that non-compliers are doing something wrong.
  • On the other hand, fixed-rate RC ends up (putatively) over-demanding when the evaluation uptake rate is low.

/ ‘If all that killing does is prematurely end a valuable life, we might wonder why the killing especially wrongs the person who got killed, rather than just wronging all of us collectively, by removing something of value from the world.’ (CHRISTINE KORSGAARD) Do utilitarians have a good answer to this question? (2023 Q10)

/ ‘Demandingness objections to consequentialism presuppose controversial assumptions that, once revealed, undermine the power of those objections.’ Is that true? (2024 Q8)

• Yes. In particular, demandingness objections presuppose an analogy to the distinction between doing and allowing, which consequentialists independently reject. So, demandingness objections do not provide novel reasons against consequentialist theories, but rather repackage existing complaints.
• Note that not all theories of consequentialism are subject to demandingness objections (e.g. satisficing versions are exempt). But for this essay I will focus on maximising consequentialist theories, the most common and the ones subject to demandingness objections.
• For concreteness, start with maximising act-utilitarianism. An action is right & obligatory iff and because it leads to the largest aggregate welfare of all actions open to an agent; all other actions are impermissible (ignoring ties).
• A demandingness objection goes something like the following:
  • 1. Donating a large fraction of my wealth to the global poor is the way for me to maximally increase aggregate welfare.
  • 2. If consequentialism is true, then I am obligated to increase aggregate welfare in the greatest way possible.
  • 3. But I am not obligated to donate a large fraction of my wealth to the global poor.
  • 4. So consequentialism is false.
• Some might reject P1 or P2.
  • For example, perhaps aggregate welfare is not the only good that matters, and I have an obligation to promote autonomy too, which donating would undermine.
    • But this can’t be right; it misses the intuition that it is good just not obligatory to donate.
  • Or, it might be that consequentialism allows me to increase the weighting I put on benefits to myself (Scheffler’s agent-centred prerogatives).
    • But this has Rich Uncle problems.
• One way to defuse the Rich Uncle objection, that proponents of demandingness objections would use: there’s a meaningful difference between failing to help your uncle and actively hurting him. i.e., the objection trades on denying the doing/allowing distinction.
  • This is true. But it illuminates the deeper problem with demandingness objections: they themselves trade on accepting the doing/allowing distinction, which consequentialists independently reject.
• Per Sobel: consider Kim and David. Kim needs a kidney, David is considering being a donor, but does not want to.
  • Consequentialism seems to demand a lot of David, that he donates.
  • But non-consequentialist theories demand that Kim does not, for example, kidnap David and steal his kidney!
  • Demands a theory makes of an agent ≠ suffering a theory permits to befall a patient.
• A consequentialist would say: a moral theory is demanding when we aggregate over the compliance costs for all individuals, including those who’re benefitted from the theory by others’ compliance.
  • But on this account, it’s not at all obvious that consequentialism is more demanding than, say, deontology.
  • Indeed, someone like Ashford who would reject P3 above thinks that it is a desirable feature of consequentialism that it’s appropriately demanding. It doesn’t seem like demandingness should be in itself a reason to doubt a theory, as opposed to the idea that it generates obligations we think are incorrect.
• So demandingness objections rely on an analogue to the doing/allowing distinction, but once you assume this you would already reject consequentialism, which denies that such a distinction exists.

Post-Claude & examiner report notes

• Flag that doing/allowing is a controversial assumption.
  • Also, are there any other presuppositions? Question is in the plural.
    • E.g. it presupposes a category of the supererogatory.
    • So we could bring in Kagan on how an account of supererogation needs to explain (i) why supererogatory acts are good and (ii) why they’re nonetheless not obligatory
• Recall that Murphy pushes back on Sobel, claiming that costs of what a theory merely permits (not requires) are attributable to the agents acting, not the theory.
  • The counterexample: a theory that says “you can do whatever you like to Joe” seems pretty costly to him. And then also we can predict being permissive → being costly for others, since humans are self-interested.
• To round off the point in final paragraph about demandingness not being a reason to doubt a theory: ethical egoism is highly undemanding, but that’s no reason to like it!
• Also, the precise way that McElwee distinguishes what a demandingness objection is: when you grant there are good moral reasons to do X, but the theory asks too much in requiring it
• Can also deploy Ashford to argue that all plausible theories will be demanding.
  • And to frame it in the context of the question: demandingness objections rest on a controversial assumption like “we don’t have obligations to help the poor at little cost to ourself” / negations of the rescue principle or Singer’s principle