Ethics

The question this subject is about is what you should do, and it is not answered by finding out what anybody, including you, happens to approve of.

That sentence looks obvious and is denied constantly, usually without anyone noticing. The free will course ended with responsibility: who is answerable for what they did, what excuses, what exempts, and whether desert survives determinism. It never asked what the person should have done in the first place. A perfectly reasons-responsive agent, fully the source of their action and fit to be praised or blamed, still has to decide what to do on Tuesday morning. That decision is this course.

Three questions under one word

"Ethics" names three enquiries that use different methods and can be answered independently, and most confused arguments come from sliding between them.

Applied ethics asks about particular cases. Is it permissible to end the life of a patient who asks for it? May a state conscript? Should you eat animals? These are the questions people actually argue about, and they are the last ones a course can usefully take, because arguing about them productively requires the other two.

Normative ethics asks what makes any act right or wrong. Not "is this permissible" but "what is the general property in virtue of which anything is permissible". Three families of answer have survived: that what matters is the outcome, that what matters is conformity to duties that hold regardless of outcome, and that what matters is the character the act expresses. The middle of this course builds all three.

Metaethics asks what a moral claim is. When someone says torture is wrong, are they describing a fact, expressing an attitude, reporting a convention, or saying something false because there are no moral facts to make it true? This is a question in metaphysics and philosophy of language rather than a question about conduct, and the next lesson is spent on it, mostly to show that it does not let you avoid the other two.

The order matters. A dispute about euthanasia that is really a dispute about whether killing and letting die differ is a normative dispute wearing applied clothes, and it will not be settled by more facts about patients. Locating a disagreement in the right layer is half the work.

Example. Classify each claim: (a) "In Sparta, exposing weak infants was accepted." (b) "An act is right if it produces at least as much good as any alternative." (c) "Moral judgements express the speaker's attitude rather than stating a fact." (d) "You should not have lied to her."

(a) is not a moral claim at all: it is a historical claim about what a society approved, and it belongs to anthropology. (b) is normative ethics, a candidate criterion of rightness. (c) is metaethics, a claim about what moral sentences do. (d) is applied, a verdict on a case. Notice that (a) is the one people most often deploy as though it settled something, and it is the only one that is not about ethics at all.

Now you. Where does "there is no objective right and wrong, so you should not impose your values on other cultures" belong, and is it consistent?

What people approve of is not the answer

The most common move in a moral argument is to cite what is accepted, and it never works, in either of its two forms.

The personal form is subjectivism about the speaker: "wrong" means "I disapprove". It has an immediate cost. If your claim that factory farming is wrong just reports your disapproval, and mine that it is fine just reports my approval, then both statements are true and we are not disagreeing about anything, any more than we disagree when you say you like anchovies and I say I do not. But we plainly are disagreeing, and one of us expects to change the other's mind. A theory that makes moral disagreement impossible has misdescribed the phenomenon it set out to explain.

The cultural form is relativism: "wrong" means "forbidden by the speaker's society". It has the same defect one level up, and adds two of its own. It makes the moral reformer automatically wrong, since by definition they dissent from their own society's code, so abolitionists in 1800 were mistaken about slavery and became right only once enough people agreed with them. And it needs a way to say which society a person belongs to, which is hopeless for anyone who belongs to several.

Both are worth distinguishing from the plain and correct observation that societies have differed enormously in what they approve. That is descriptive relativism, it is well established, and by itself it entails nothing about whether any of those societies was right. Physics has also been believed differently in different places.

The gap between is and ought

The general point behind all this was made by David Hume in the Treatise of Human Nature in 1739, in one of the most cited paragraphs in philosophy.

Hume observes that authors proceed for some time with ordinary reasonings about what is and is not the case, and then, imperceptibly, every proposition becomes connected by "ought" or "ought not" instead. He remarks that this change is of the last consequence, that the new relation needs to be explained, and that a reason should be given for how it is deduced from things entirely different from it.

Read as a logical claim, it is the requirement that a valid argument cannot have a conclusion containing a term absent from all its premises: no set of purely descriptive premises entails an evaluative conclusion. From "this causes suffering" nothing follows about what to do until you add "causing suffering is wrong" or something like it. That extra premise is the moral premise, it is doing all the work, and the discipline of the rest of this course is largely the discipline of finding it and arguing about it directly.

G. E. Moore sharpened the point in Principia Ethica in 1903 with the open question argument. Suppose someone claims that "good" just means "what we desire to desire", or "pleasant", or "what furthers evolution". Then "this is pleasant, but is it good?" ought to sound like "this is a bachelor, but is he unmarried?", a question closed by the meanings of the words. It does not sound like that. It sounds like a real question, which suggests that "good" is not synonymous with any natural property. Moore called the confusion the naturalistic fallacy and thought it fatal. Later work, particularly on how names refer, weakened it considerably: two terms can pick out the same property without a competent speaker knowing it, which is how water turned out to be H₂O without "is water H₂O?" being a closed question. What survives is a warning, not a proof: any definition of the good in natural terms owes an argument, and cannot be helped to itself.

Example. Someone argues: "Homosexuality is unnatural, since it does not lead to reproduction. Therefore it is wrong." Diagnose the argument.

The premise is descriptive and the conclusion evaluative, so a moral premise is missing, and when it is supplied the argument collapses. Written out fully, it needs "whatever does not lead to reproduction is wrong", which condemns celibacy, contraception, and every hour anyone has ever spent reading. Weaker versions fare no better: "whatever is statistically unusual is wrong" condemns left-handedness, and "whatever does not serve a biological function is wrong" condemns music. The general lesson is that arguments from nature almost always hide a moral premise that nobody would accept once it is written on the page, and writing it on the page is the whole technique.

Now you. A doctor argues: "This treatment has a 4 percent mortality rate and the alternative has 11 percent. So you should choose the first." Is a moral premise missing, and does that matter?

What free will settled and what it left open

This subject follows the free will course, and the relationship between the two is worth stating, because they are often run together.

Free will is about the conditions on being an apt target of praise, blame and punishment. It asks what has to be true of an agent for responsibility to attach to what they did: whether they could have done otherwise, whether they were the source of their action, whether a determined agent can deserve anything. Its subject is the agent.

Ethics is about the act. It asks which of the options in front of someone they ought to take. That question arises identically whether or not anyone is ever responsible for anything. Suppose the hard incompatibilist is right and no one deserves blame for anything ever. There is still a difference between administering the drug and withholding it, and a doctor still has to choose. The question of what to do does not go away when desert does.

The connection runs in one direction only, and it is worth being precise about it. If nobody can be blamed, the reactive part of ethics changes: guilt, resentment and punishment need rethinking, which is where the free will course ended. The deliberative part is untouched. This is why sceptics about free will are not, and do not need to be, sceptics about ethics, and why the strongest of them write books about how to reform punishment rather than books saying that nothing matters.

Why the subject is not optional

There is a tempting position that says all this is interesting but idle, since people act on their intuitions anyway and the theory is decoration.

It would be more convincing if the intuitions were consistent. They are not. Most people hold that it is wrong to let a child drown in front of them to save a suit, and also that they have no obligation to send the price of a suit to a charity that would prevent a child's death elsewhere. Most people hold that killing is worse than letting die, and also that a doctor who switches off a ventilator at a family's request is not a murderer. Most people hold both that the numbers count and that you may not cut up one healthy patient to save five. Each pair can be reconciled, and each reconciliation commits you to a principle with further consequences that you then have to accept or reject. That is what theory is for: not to replace intuitions, but to find out what they cost.

The second answer is that the cases where intuitions are agreed are not the cases that matter. Nobody needs a theory to know that torturing children for entertainment is wrong. Theory is for the cases where careful and decent people disagree, which is why this course spends its last third on exactly those.

Example. Two people agree entirely about the biology of a foetus at twelve weeks and disagree completely about abortion. What kind of disagreement is it, and what would move it?

It is normative, not factual, and no further embryology will touch it. What they disagree about is which property confers moral status: species membership, sentience, the capacity for a future, or something else. Progress requires each to state the criterion they are using and then test it against cases neither has a stake in, which is exactly what the lesson on moral status does. Notice that this diagnosis does not favour either side; it says only that the argument is in the wrong place, and that both parties are wasting their time trading ultrasound images.

Now you. Name a factual question whose answer would change what a consequentialist should say about the death penalty, and a normative question that no fact could settle.

What this course will and will not deliver

It will not deliver a formula. There is no algorithm at the end of ethics that takes a case and returns a verdict, and any course that offers one is selling something.

What it delivers is narrower and more useful. By the end you should be able to take a moral argument, state it so that its premises are separately attackable, identify which family of theory each premise belongs to, generate the counterexample that tests it, and say what would have to be true for your own view to be wrong. That is not a small skill, and almost nobody has it. It is the difference between having opinions and being able to defend them.

The next lesson takes the obvious objection head on. If ethics is not a report of what anyone approves, and cannot be deduced from facts about the world, then perhaps there is nothing there at all, and the whole subject is an elaborate way of expressing preferences. That position has serious defenders and serious arguments, and it turns out to change less than it promises.

If moral conclusions cannot be deduced from facts about the world, the obvious next thought is that there are no moral facts for them to be about.

That is a serious position with serious defenders, and the previous lesson's argument leads straight to it. This lesson takes the anti-realist family at full strength, since a version stated weakly is easy to knock down and teaches nothing. It ends somewhere that surprises most people: whichever of these views is right, the reader still has to work out what to do, and by the same means.

The menu

Start by laying out the options, because the vocabulary is a mess and the same word is used for three positions.

Realism says there are moral facts, they are not constituted by anyone's attitudes, and moral claims are true or false by whether they match. It comes in a naturalist version, on which moral facts are ordinary facts about welfare or flourishing, and a non-naturalist version, on which they are a further kind of fact not reducible to any other.

Subjectivism says moral claims state facts, but facts about attitudes: "wrong" means "disapproved of by me" or "by my society". These are the relativisms of the previous lesson, and they are cognitivist, meaning that moral sentences say something capable of truth.

Error theory says moral claims genuinely try to state objective facts, and there are none, so all positive moral claims are false. John Mackie set this out in Ethics: Inventing Right and Wrong in 1977, and the position is the most radical on the list: torture is not wrong, and it is not right either, in the way that no witch is a real witch.

Expressivism says moral claims do not try to state facts at all. Saying that torture is wrong is more like booing torture than like reporting anything. A. J. Ayer's 1936 version was crude; Simon Blackburn and Allan Gibbard spent the 1980s and 1990s building versions that behave in argument almost exactly like realism, which is called quasi-realism.

Two axes cut across each other here. Does a moral sentence state something true or false, and is what makes it true independent of attitudes? Realism says yes and yes, subjectivism yes and no, error theory yes and there is nothing, expressivism no to the first question so the second does not arise.

The argument from disagreement, and why it is weak

The most popular argument against realism is that people disagree, endlessly, and that this is best explained by there being nothing to be right about.

Mackie's version, the argument from relativity, is more careful than the popular one. He does not claim that disagreement proves there are no facts. He claims that the pattern of variation is better explained by the hypothesis that moral codes reflect ways of life than by the hypothesis that they are imperfect perceptions of an objective order. People do not just disagree; their disagreements track their societies' economies and histories in exactly the way you would expect if the codes were adaptations rather than discoveries.

The standard reply has three parts, and it is strong. First, disagreement is compatible with realism everywhere else: physicists disagreed for centuries about the age of the earth without anyone concluding there was no fact of the matter. Second, much moral disagreement is downstream of factual disagreement. Societies that burned witches were not operating an alien value system; they accepted that maleficent supernatural harm was real and that those who inflicted it should be stopped, which is a value most modern readers share applied to a false belief. Third, agreement is broader than the argument suggests. Gratuitous cruelty, betrayal of the helpless and bad faith are condemned essentially everywhere, and the variation is in scope, in who counts, rather than in the underlying principles.

What that reply cannot do is explain away the residue. Some disagreements survive every correction of the facts, and Mackie's explanation of them is not obviously worse than the realist's.

Example. Two societies both hold that killing the innocent is gravely wrong. One practises infanticide of disabled newborns and the other does not. Is this a disagreement in values?

Not necessarily, and the distinction matters. If the practising society believes that a newborn is not yet a person, or believes that a disabled infant faces certain suffering and early death that exposure shortens, then the two societies share the principle and differ about its application, which is a factual disagreement about status or about consequences. Anthropological work on infanticide finds a great deal of this. What would be a genuine disagreement in values is a society that agreed on every fact, including the infant's status and prospects, and still held that its life carried no weight. Real cases are mostly the first kind, which is why the argument from disagreement is far less potent than it looks, and why establishing that a residue exists takes careful work rather than a list of shocking customs.

Now you. Someone argues that widespread moral progress, such as the abolition of slavery, proves realism, since progress requires something to progress towards. Assess it.

Mackie's queerness argument

The stronger anti-realist argument is metaphysical rather than sociological.

Mackie asks what an objective moral fact would have to be like. It would have to be a feature of a situation that is both there in the world and, merely by being perceived, motivating: seeing that an act is wrong would have to move you not to do it, without any desire of yours being involved. Ordinary facts are not like that. Knowing where the door is does not make you want to leave. So a moral fact would be, in Mackie's word, queer: an entity of a kind utterly different from anything else in the universe, requiring a faculty of moral perception equally unlike any other.

There are two components, and they are worth separating. The metaphysical strand says such properties would not fit into a scientific picture of the world. The epistemological strand says we would have no credible account of how we came to know about them, and this is the sharper one. Evolution explains why we have the moral responses we have without any reference to their truth: tribes of cooperators outcompeted tribes of defectors, and that story is complete. Sharon Street pressed this in 2006 as the Darwinian dilemma. Either our evaluative attitudes are tracking mind-independent moral truths, in which case the realist owes an account of the coincidence between what selection installed and what is true, or they are not, in which case realism gives us no reason to trust any of them.

The realist replies are real but expensive. One is the companions-in-guilt move: mathematical and modal facts are just as queer, and nobody abandons arithmetic. Another is naturalism: identify moral facts with facts about welfare, which are perfectly ordinary, at the cost of the open question. A third is to deny the motivational requirement, holding that a person can judge an act wrong and simply not care, which is what the amoralist seems to be, and to let moral knowledge work like any other knowledge.

Expressivism and the Frege-Geach problem

Expressivism escapes queerness completely, because it needs no moral facts, and it runs into a technical problem that has shaped forty years of work.

If "lying is wrong" expresses disapproval rather than stating anything, what does it do inside a larger sentence where it is not being asserted? Consider the argument: if lying is wrong, then getting your brother to lie is wrong; lying is wrong; therefore getting your brother to lie is wrong. That is a plain modus ponens and it is obviously valid. But in the first premise nobody is disapproving of lying; the speaker is supposing it. For the argument to be valid, "lying is wrong" must mean the same in both premises, and expressivism seems committed to it meaning something different, since in one place it expresses an attitude and in the other it does not. Peter Geach pressed the point in the 1960s, building on an objection of John Searle's, and it is called the Frege-Geach problem.

Blackburn's answer is to build a logic of attitudes: to accept the conditional is to hold a higher-order attitude, disapproving of the combination of disapproving of lying while not disapproving of procuring it, so inconsistency becomes a kind of practical incoherence rather than a logical falsehood. Gibbard's is to model judgements as plans and treat inconsistency as the impossibility of a plan covering all contingencies. Both work, in the sense that they reproduce the inferences. Whether they explain validity or merely mimic it is still argued, and the honest summary is that expressivism survived the objection at the cost of enormous complexity.

Example. A friend says: "Morality is just evolution. We help our relatives because it spread our genes, and that is all there is to it." What is right and what is wrong in this?

The evolutionary claim is largely correct and by itself establishes nothing, since it is an explanation of why we make moral judgements rather than an evaluation of them. Taken as an argument it commits the genetic fallacy: showing how a belief arose does not show it false, or arithmetic would be in trouble too, since counting also has a selective explanation. The serious version is Street's, which does not say the beliefs are false but that a realist has no story about why selection would have tracked truth, so the beliefs are unreliable on realist assumptions. That is an argument about justification rather than origin, and it is much harder to answer. Notice also that the friend's claim, taken at face value, licenses nothing about how to behave now, since we routinely act against inclusive fitness and do not regard that as an error.

Now you. Does the following argument work? "There are no objective moral facts. So nothing is really wrong. So I may do as I please."

Why the metaethics does not settle anything

Now the point of the lesson. Suppose you accept, having read the arguments, that expressivism is true and there are no moral facts. What changes about the argument over euthanasia you have to have tomorrow?

Almost nothing. You still have attitudes, and you still find some of them inconsistent with others, and consistency is exactly what moral argument trades on. You still discover that your objection to euthanasia rests on a principle about killing that you do not apply to switching off ventilators. You still find that the principle you would need to defend your position condemns things you approve of. Every move made in the remaining lessons of this course is available to you, because every one of them works by exposing tension between commitments, and tension between attitudes is as real as tension between beliefs.

This is the point of quasi-realism, and it is why the position was built. Blackburn's programme was to earn the right to talk exactly like a realist, to say that some moral views are better than others, that we can be mistaken, that torture would still be wrong if everyone approved of it, all from expressivist materials. To the extent it succeeds, the practical difference between the two positions is nil.

The one thing that does change is the character of a final standoff. If two people have run out of shared premises and disagree at the foundations, the realist thinks one of them is wrong about something, and the expressivist thinks they simply differ. That matters for what you say at the end of an argument, and not for anything you say during it. Since almost no real disagreement gets that far, the metaethics has less practical purchase than its prominence suggests.

Example. Which position does each claim belong to? (a) "Nothing is wrong, though we should keep talking as if things were." (b) "Wrong means forbidden here." (c) "Saying it is wrong is not saying anything true or false." (d) "It would be wrong even if everyone approved."

(a) is error theory in its fictionalist form, keeping moral talk as a useful pretence. (b) is cultural relativism, a species of subjectivism. (c) is expressivism, denying that moral sentences are truth-apt. (d) is what realism straightforwardly asserts, and it is also what quasi-realism claims to be able to say, which is exactly why sorting positions by their slogans is unreliable and you have to ask what makes the slogan true.

Now you. Your interlocutor is an error theorist. What can you still say to them about factory farming?

Where this leaves the reader

The state of professional opinion is worth knowing without being deferred to. The 2020 PhilPapers survey of academic philosophers found that just under two thirds accepted or leaned towards moral realism, with about a quarter on the anti-realist side, which is a majority rather than a consensus.

The safe conclusion is that the metaethical question is open, and that its being open does not stop anything. Every position on the list generates the same next task: work out which acts to defend, which principles support them, and what to say when the principles collide.

That task has a method, and the method is teachable. The next lesson sets it out: how a moral argument is put in a form where it can be attacked, what a counterexample does to a principle, what thought experiments are good for, and what it means for a set of moral judgements to be in good order.

A moral argument is worth having only when it is stated in a form where someone can say exactly which line they reject.

Most are not. They arrive as a paragraph in which a case, a principle and a verdict are fused, and the person who disagrees can only repeat their own paragraph louder. The previous two lessons established that the question of what to do is not settled by opinion and not dissolved by metaethics, so it has to be argued. This lesson is the method, and it is the most immediately useful thing in the course. Everything after it is an application.

Put it in premises

The first move on any moral argument, including your own, is to write it as numbered lines with the conclusion at the bottom.

Take a real one. "Eating meat is wrong because it causes unnecessary suffering." Written out: (1) Factory farming causes severe suffering to animals. (2) Eating meat from factory farms is not necessary for human health or flourishing. (3) It is wrong to cause severe suffering for no necessary purpose. (4) Therefore eating meat from factory farms is wrong.

Now the argument has surfaces. Premise 1 is empirical and can be attacked with evidence about husbandry. Premise 2 is empirical and can be attacked with nutrition. Premise 3 is the moral premise, and it is where the real disagreement usually lives, though almost nobody attacks it, because as stated it looks unimpeachable. Notice also what has become visible: the conclusion is narrower than the slogan, since nothing in the argument touches meat from an animal that lived well and died painlessly, and an opponent who points that out has done something useful rather than merely obstructive.

Two habits go with this. State the argument in its strongest form, including repairs the other side has not thought of, because refuting a weak version teaches nobody anything. And identify the moral premise explicitly, since the previous lesson established that there is always at least one and it is always the load-bearing line.

Example. Put this in premises: "Voluntary euthanasia should be legal because people have the right to control their own bodies."

(1) Competent adults have a right to control what is done to their own bodies. (2) A right to control what is done to your own body includes the right to refuse or request medical interventions, including one that ends your life. (3) The law should protect the rights of competent adults where doing so does not seriously harm others. (4) Legalising voluntary euthanasia under safeguards does not seriously harm others. (5) Therefore voluntary euthanasia should be legal. Now look at where the argument is actually contested. Almost no opponent denies (1). Most attack (2), by holding that a right to refuse treatment does not extend to a right to be killed, which is the acts and omissions question of a later lesson. Serious opponents also attack (4) with an empirical claim about pressure on the vulnerable, which is a testable prediction rather than a principle. Two minutes of writing has relocated the disagreement from a slogan to two specific lines, one philosophical and one empirical.

Now you. Put in premises: "Capital punishment is wrong because the state should not kill."

One counterexample kills a universal principle

Moral principles are usually stated universally, and a universal claim is refuted by a single instance. This is the workhorse technique of the subject, and it is borrowed straight from logic.

Consider: "It is always wrong to break a promise." You promised to meet a friend for coffee at three. On the way you pass a child face down in a pond. Breaking the promise is obviously permissible, so the principle as stated is false. Note what has and has not been shown. The counterexample does not show that promises do not matter, and it does not establish any rival theory. It shows that this formulation is too strong, and it forces a repair.

The repairs are where the philosophy happens. You might weaken the principle to "it is wrong to break a promise unless doing so prevents a much greater harm", which concedes that promissory obligation can be outweighed and raises the question of by how much. Or you might restrict it: "it is wrong to break a promise for one's own convenience", which keeps absoluteness at the cost of covering much less ground. Or you might bite the bullet and insist the child should drown, which almost nobody does but which is at least a position. Each repair has consequences that can be tested with a further case, and the sequence of case, repair, case is what a philosophical literature actually consists of.

The same technique attacks the two halves of a criterion separately. If someone says an act is wrong if and only if it violates consent, you can attack sufficiency by finding a consent violation that is not wrong, such as a surgeon operating on an unconscious accident victim, and you can attack necessity by finding a wrong that violates nobody's consent, such as polluting an uninhabited river. Knowing which half you are attacking keeps an argument from going in circles.

Thought experiments, and what they cannot do

A thought experiment in ethics is a controlled comparison. Its purpose is to hold everything fixed except one feature and see whether the verdict moves, which is exactly what an experiment does and exactly why the cases are so artificial.

Take the two standard trolley cases, treated properly in a later lesson. In one, you divert a runaway trolley from five people onto one. In the other, you push a large man off a bridge to stop it, again saving five at the cost of one. The numbers are identical. If people's verdicts differ, and they do, something other than the numbers is doing the work, and the whole point of the pair is to isolate what. Objecting that no real trolley behaves like this misses the design, in the way that objecting to a frictionless plane misses the design.

The limits are real, though, and honesty requires stating them. First, intuitions about extreme cases may be unreliable precisely because they are extreme: a response system tuned by ordinary life may not have anything sensible to say about a bridge and a fat man. Second, they are sensitive to presentation. Lewis Petrinovich and Patricia O'Neill showed in 1996 that wording the same dilemma in terms of saving rather than killing moved judgements substantially, and that the order in which cases were presented changed the answers to later ones. Eric Schwitzgebel and Fiery Cushman reported in 2012 that professional philosophers with doctorates showed order effects of essentially the same size as everyone else, which undercuts the hope that expertise filters the noise out.

The reasonable conclusion is not to abandon the method but to treat a single intuition about a single case as weak evidence, and to look for the pattern that survives reframing. That is also what an experimentalist does with a noisy measurement.

Consistency and universalisability

The strongest lever in any moral argument is not a principle but an inconsistency in the other person's commitments, because it does not require them to accept anything new.

The general form runs: you judge case A one way and case B the other; name the difference between them that justifies the different verdicts; and then either the named difference survives testing, in which case you have learned something, or it does not, in which case one of the two verdicts has to go. This is the technique behind almost every famous argument in the subject. Rachels does it with two bathtubs, Singer with a pond and a cheque, the argument from marginal cases with a pig and an infant.

Universalisability is the same tool aimed at the agent. If it is permissible for you to do this, it is permissible for anyone relevantly similar, and "relevantly similar" cannot be filled in by naming yourself. The popular version, "what if everyone did that", is a blunt instrument that overreaches: if everyone became a doctor, nobody would grow food, yet becoming a doctor is fine. The repair is to ask not what would happen if everyone did it but whether you can coherently will that everyone in your circumstances be permitted to, which is Kant's test and gets its own lesson.

Example. A colleague argues that eating pigs is fine because they are not human. You want to test the principle rather than assert the opposite. What do you ask?

You ask what work "human" is doing, and then look for the property behind it. If the claim is that only humans have interests that matter, the test case is an anencephalic infant or a person in a permanent vegetative state, who lack the cognitive capacities a pig has, and whom the colleague presumably does not think may be farmed. If the reply is that the infant belongs to a species whose typical members have those capacities, that principle can be tested too: it implies that a hypothetical alien of high intelligence whose species were typically dull would be fair game. The point of the sequence is not to trap the colleague, which is a bad way to change anyone's mind, but to locate the property they actually rely on, so that the two of you are arguing about the same thing. Many people, pressed this way, end up defending a species-membership criterion openly, which is a real position with real defenders and can then be discussed on its merits.

Now you. Someone holds both that a fifty percent inheritance tax is theft and that a fifty percent income tax on a footballer is not. What do you ask them?

Reflective equilibrium

If principles are tested against cases and cases are tested against principles, what stops the whole thing being circular? The standard answer, from John Rawls in 1951 and again in A Theory of Justice in 1971, is that circularity is not the objection it looks like.

The method is called reflective equilibrium. You start with your considered judgements, meaning the ones you hold calmly, with the facts in front of you, and with no stake in the outcome. You look for principles that systematise them. Where a principle conflicts with a judgement you revise one or the other, choosing whichever you hold with more confidence and whichever is better supported by everything else you believe. You keep going until principles and judgements sit together without conflict. That state is the equilibrium, and it is the only standard of justification the subject has.

Two features stop it being empty. It is a coherence standard applied to a very large body of commitments, and coherence over a big enough set is hard to reach: most people's moral views, written out, are inconsistent in several places, and finding out where is genuinely informative. And in its wide form, the version Rawls and Norman Daniels defend, the equilibrium must also cohere with background theories of the person, of society and of how our judgements were formed, which is what lets a debunking explanation knock out a judgement. If your confident intuition that a caste is inferior is explained by the fact that you were raised in it, that explanation is part of the evidence and counts against the intuition.

The standard objection is that the method is conservative: it takes your existing judgements as data and can only tidy them, so a person who starts with monstrous views reaches a tidy monstrous equilibrium. The reply is that the wide version imports outside constraints, that consistency alone has historically had radical consequences, since almost every extension of moral concern was won by pointing out an inconsistency, and that no alternative method has been proposed that starts from nowhere. That is a real limitation honestly stated rather than a refutation.

Example. You are confident that torture is always wrong, and equally confident that if torturing one terrorist were the only way to find a bomb that would kill a city, it should be done. What does reflective equilibrium require here?

That you stop holding both, and it does not tell you which to drop. What it does tell you is how to decide. Ask which judgement you hold more firmly when calm and disinterested. Ask whether either has a debunking explanation: the second is the ticking bomb case, and it is worth knowing that it is a fictional scenario with no clear real instance, that it stipulates certainties no interrogator ever has, and that it was popularised partly through television, all of which is evidence about the reliability of the intuition it produces. Ask what each option costs elsewhere: dropping the absolute prohibition means accepting a threshold and then defending where it sits, while keeping it means accepting the city. Most people who work through this end up with a very high threshold rather than a true absolute, and the important thing is that they can say why.

Now you. What in this method distinguishes an intuition worth keeping from one worth discarding?

What good looks like

A moral argument has gone well when both parties can state the other's position in a form the other accepts, when the disagreement has been narrowed to one or two identifiable premises, and when each can say what would change their mind.

That last requirement is the sharpest test and it deserves to be applied first. If no possible fact and no possible case would move someone's position, they are not holding a view, they are announcing an allegiance. The same question asked of yourself is uncomfortable in a useful way, and the final lesson of this course is built around answering it.

With the method in place, the substance can start. The next lesson takes the most natural moral principle there is, the one almost everyone reaches for first, and states it precisely enough to be attacked: that the right thing to do is whatever makes things go best.

The most natural answer to what you should do is that you should do whatever makes things go best, and almost everyone reaches for it before they reach for anything else.

Ask a stranger why lying is wrong and you will hear that it hurts people. Ask why a law is bad and you will hear what it will cause. This is consequentialism operating as the default background theory of public argument, usually unexamined. The previous lesson supplied the method for examining it: state it in premises, find its moral premise, and test it against cases. This lesson states it at full strength and makes the case for it. The two after that take it apart.

The idea, and what makes it attractive

Consequentialism holds that the moral quality of an act is entirely a matter of its outcomes. Nothing else counts: not who performs it, not what rule it falls under, not the agent's history or intentions except insofar as those affect what happens.

Two features give it its pull. The first is that it explains rather than stipulates. Rival theories tend to arrive with a list of things that are forbidden; consequentialism says why anything is forbidden at all, namely that it makes things worse, and then derives the list. A theory that generates its prohibitions from one principle is doing more work than one that reports them.

The second is impartiality. Jeremy Bentham's formula, as Mill reports it, is that each is to count for one and nobody for more than one. Your suffering counts exactly as much as mine, and a stranger's exactly as much as your friend's, and the theory has no way of writing your own interests in larger letters. Every historical extension of moral concern, to the enslaved, to women, to foreigners, to animals, is an application of exactly that step, and consequentialists were unusually often ahead of the curve on each of them.

There is a third, less advertised, which is that the theory answers rather than shrugging. Given the facts, it always returns a verdict. Whether that verdict is right is the argument of the following lessons, but a theory that goes silent on hard cases is not obviously better.

Bentham's calculus

Bentham's Introduction to the Principles of Morals and Legislation, printed in 1780 and published in 1789, tries to make the theory operational, and the attempt is more interesting than its failure.

The good, for Bentham, is pleasure and the absence of pain, and nothing else. Since these are the only things that matter, the value of any act is fixed by the quantity of pleasure and pain it produces, and quantity has dimensions. He lists seven: the intensity of a pleasure, its duration, its certainty, its propinquity or nearness in time, its fecundity or tendency to be followed by more of the same, its purity or freedom from following pain, and finally its extent, meaning the number of people affected.

The first six describe a pleasure; the seventh is where the aggregation happens, and it is the radical one. Extent turns a private hedonism into a moral theory, because it says that the arithmetic runs across persons and every person is a term in the sum.

Bentham is candid that nobody can literally run the calculation, and says only that it should be kept in view as the standard the practice approximates. That concession is more important than it looks, and it reappears below as the distinction between a criterion and a procedure.

Example. A city council can spend a fixed budget on one of two health programmes. Programme A averts a year of healthy life lost for every 50 pounds spent; programme B does so for every 500 pounds. The budget is one million pounds. What does the calculus say, and what does the comparison actually show?

Divide: one million at 50 pounds each averts $1\{,\}000\{,\}000/50=20\{,\}000$ years, and at 500 pounds each averts $1\{,\}000\{,\}000/500=2\{,\}000$ years, a factor of ten. On the theory the council should spend the whole budget on A and nothing on B, since splitting it produces strictly less good. The comparison matters because such ratios are not hypothetical: cost-effectiveness estimates across health interventions in the Disease Control Priorities literature span roughly three orders of magnitude, so the gap between a good and a poor allocation is far larger than the gap between spending and not spending. What the calculation does not settle is whether averted years of healthy life are the right measure, whether whose years they are should matter, or whether a council owes anything to the specific patients in front of it, and each of those objections belongs to a later lesson.

Now you. Programme B serves a remote community that programme A cannot reach. Does that change the consequentialist verdict?

Mill's repair

Bentham's hedonism has an immediate embarrassment: if pleasure is pleasure, then a life of trivial amusement, sufficiently prolonged, beats a short and difficult life of achievement, and pushpin is as good as poetry.

John Stuart Mill, in Utilitarianism in 1863, seventy-four years after Bentham's book, tries to keep hedonism while denying the conclusion. Pleasures differ in kind as well as amount, he says, and some are higher. The test he offers is a competent judge: of two pleasures, if all or almost all who have experienced both give a decided preference to one, irrespective of any feeling of moral obligation to prefer it, then that one is higher, and its superiority may be so great that no quantity of the other will outweigh it. Hence the famous line that it is better to be a human being dissatisfied than a pig satisfied, better to be Socrates dissatisfied than a fool satisfied.

The move is popular and unstable. If the higher pleasure is preferred because it is more pleasant, the appeal to kinds is unnecessary and quantity was doing the work all along. If it is preferred for some other reason, then something other than pleasure is being valued and pure hedonism has been abandoned. Mill's own qualification that no quantity of the lower outweighs the higher makes this worse, since it treats one kind of value as lexically prior to another, which is not a hedonic calculation at all.

The lasting significance of the repair is that it is the first crack in the identification of the good with pleasure. The next lesson widens it.

Act consequentialism, stated

Here is the theory in the form later lessons will attack. An act is right if and only if its outcome is at least as good as the outcome of every alternative available to the agent.

Four features of that sentence are doing work. It is maximising: doing a great deal of good is not enough if more was available, which is why the theory has no category of the supererogatory, the praiseworthy but optional. It is agent-neutral: the value of an outcome does not depend on who brings it about, so your killing one person and someone else killing one person are equally bad states of affairs, and this is the feature that generates the objections of the next-but-one lesson. It is welfarist in its standard form, meaning that what makes an outcome good is how well things go for those in it. And it is complete: every act is right or wrong, with no gaps.

Two variants matter. Actual-outcome versions say the right act is the one that in fact turns out best, which makes rightness depend on luck and often unknowable at the time. Expected-value versions say it is the one with the highest expected value given the evidence available, which is what any real decision has to use. Most contemporary consequentialists distinguish the two: the actual-outcome version fixes what would have been best, and the expected-value version fixes what it was reasonable to do, and blame attaches to the second.

Example. A doctor prescribes a drug with a 1 in 1,000 chance of a fatal reaction and a large expected benefit. The patient is the unlucky one and dies. Did the doctor act wrongly?

On the actual-outcome version, yes: the alternative would have gone better, so the act was wrong, though the doctor is blameless because she could not have known. On the expected-value version, no: given the evidence, prescribing had the higher expected value and was therefore right, and the death is a bad outcome of a right act. Most people find the second closer to how they use the word, and the split is worth keeping because it dissolves a common confusion in which a bad result is treated as proof of a bad decision. Notice that the two versions never disagree about what to do, since only one of them is usable in advance; they disagree about what to say afterwards.

Now you. A campaign vaccinates one million people against a disease. The vaccine kills 1 in 100,000 recipients and the disease would have killed 1 in 500. Work the numbers and say what the theory concludes.

Criterion or procedure

A persistent misreading treats consequentialism as instructing you to calculate before every act, which would be absurd and would itself have terrible consequences.

The standard reply, present in Bentham and made explicit by Henry Sidgwick in The Methods of Ethics in 1874, distinguishes the criterion of rightness from the decision procedure. The criterion says what makes an act right. The procedure is whatever method of deciding actually produces the most right acts, and by the theory's own lights that will usually be habits, rules of thumb and firm dispositions rather than case-by-case calculation, because people calculating under pressure make self-serving errors, take too long, and become unreliable to others.

So a consequentialist can consistently recommend that you keep your promises without weighing, that surgeons follow protocols, and that you cultivate a settled aversion to violence, on the grounds that agents with those dispositions produce better outcomes than agents who deliberate afresh. This absorbs a great deal of the intuitive appeal of rival theories.

Example. You have promised to help a friend move house on Saturday. On Friday you are offered a shift of paid work whose earnings, donated well, would do more good than your friend's afternoon is worth. What does the theory say?

At the level of the criterion, it says to take the shift, since the outcome is better, and any consequentialist who denies this has not understood their own theory. At the level of the procedure, it says something more interesting. An agent who reasons this way about promises will break them often, will be known for it, and will find that nobody relies on them, which destroys the very good that made their promises useful in the first place. The theory therefore recommends being the sort of person who keeps promises without weighing, and such a person will keep this one. The two answers do not contradict each other, because they answer different questions: whether the act was optimal, and what disposition produces the most optimal acts over a life. Notice how much of ordinary morality this recovers, and notice the residue, which is that the theory still calls the promise-keeping a mistake in this instance and calls the disposition that produced it correct.

Now you. Is a consequentialist who feels guilty about breaking a promise, in a case where breaking it produced the best outcome, making an error?

The structure also creates a problem that Sidgwick faced squarely and called the esoteric part of the doctrine. If the best outcomes come from most people believing something other than consequentialism, the theory recommends that they be brought to believe it, and that the truth be kept among those who can handle it. Bernard Williams later called this Government House utilitarianism, and it is not an attractive feature: a moral theory that recommends its own concealment has an awkward relationship with the idea that moral reasons are the kind of thing one can state publicly.

The record

Theories are usually judged on their counterexamples, and the next lessons supply plenty. It is worth first noting what this one got right, because a theory with a good predictive record deserves more patience than one without.

Bentham argued for the decriminalisation of homosexuality in an essay written around 1785 and left unpublished in his lifetime, at a time when the offence was capital in England. He argued against slavery and for the legal equality of women. He put the question about animals in a footnote that is still the standard citation in the field: the question is not whether they can reason, nor whether they can talk, but whether they can suffer. He was a systematic prison reformer. Mill wrote The Subjection of Women in 1869 and was the first member of Parliament to move for female suffrage.

None of that is an argument for the theory, and one can reach the same conclusions from other premises. What it establishes is that the theory's central move, insisting that a being's interests count regardless of which being it is, has repeatedly produced conclusions that were regarded as absurd when drawn and obvious a century later. Any objection to consequentialism has to reckon with the possibility that today's counterexample is tomorrow's obvious truth, and that a strong intuition against a theory's verdict is exactly what the slave owner had.

That is the strongest thing that can be said in its favour, and it should be held in mind through the two lessons that follow. The immediate task is narrower. The theory says to maximise the good, and so far the good has been left as pleasure, which Mill already found untenable. Before the theory can be assessed, it needs a filling, and every candidate filling turns out to have a problem of its own.

A theory that tells you to make things go best is not yet a theory until it says what "best" means, and every candidate answer has a case that breaks it.

The previous lesson left the good as pleasure, which is where Bentham left it and where Mill already found it untenable. This lesson works through the three families of answer, then turns to a harder problem the theory of the good cannot avoid: adding welfare up across people, and deciding how many people there should be. The population arithmetic is the part most readers have never seen, and it is the part that does the most damage.

Hedonism and the experience machine

Hedonism says that what is good for a person is pleasure and the absence of pain, and nothing else. It has the virtue of being about something everyone recognises from the inside, and the vice of being refutable by one thought experiment.

Robert Nozick, in Anarchy, State, and Utopia in 1974, asks you to imagine a machine that gives any experience you want. Floating in a tank, electrodes on your skull, you would believe yourself to be writing a novel, making a friend, reading a book. Others can plug in too, so nobody need be neglected. You can preprogramme a lifetime. From the inside it would be indistinguishable from a wonderful life. Would you plug in?

Most people say no, and if hedonism were true there could be no reason to refuse, since the machine delivers strictly more pleasure than any real life. Nozick draws out what the refusal reveals: we want to do things, not merely to have the experience of doing them; we want to be a certain sort of person, and someone floating in a tank is not anything; and we want contact with a reality deeper than what we can construct. Each of those is a value that hedonism cannot accommodate.

The honest qualification is that the intuition is less clean than it looks. Felipe De Brigard reported in 2010 that when subjects are instead told that they have been in a machine all their lives and can now leave, most choose to stay, and that adding an unattractive real life increases the preference for staying further. That suggests a large part of the standard reaction is status quo bias rather than a perception of value. The result does not rescue hedonism, since even a modest residual preference for reality is one hedonism cannot explain, but it does mean the case should be cited as evidence rather than as a demonstration.

Preference satisfaction

The natural repair is to stop asking what feels good and start asking what people want. A person's life goes well to the extent that their preferences are satisfied, whatever those preferences are about.

This fits economics, which measures preference through choice, and it respects people's own authority over their lives. It also handles the experience machine immediately: you prefer that your friendships be real, that preference is not satisfied in the tank, so the tank is worse for you even though it feels the same.

Three problems follow it. The first is preferences whose satisfaction never touches you. You want a stranger you met once on a train to recover from her illness; she does, and you never learn of it. Did your life go better? Something has to be said, and both answers are awkward: yes makes welfare depend on events with no connection to the person at all, and no forces a restriction to preferences about one's own life that is hard to state without circularity.

The second is misinformed and malicious preferences. Someone wants a drink that is poison. Someone else wants their neighbour to suffer. Counting these at face value gives absurd results, so the theory is normally refined to satisfaction of the preferences one would have if fully informed and rational. That helps, and it moves the theory towards the objective view, because the idealisation is now carrying the weight and the question is what makes an idealisation the right one.

The third is the hardest, and it comes from development economics. Amartya Sen documented that people in long deprivation report themselves satisfied: a woman with no education in a society that permits her none may sincerely prefer things as they are, and health surveys find the chronically ill in poor regions reporting less morbidity than better-off people with milder conditions. Martha Nussbaum built her capabilities approach around exactly this. If welfare is preference satisfaction, adaptive preferences make oppression invisible, since the oppressed adjust their wants downwards until they are met.

Example. A hospital must choose between a treatment that leaves patients pain-free but with reduced memory, and one that preserves memory at the cost of chronic moderate pain. Patients split. What does each theory of welfare say?

Hedonism gives a determinate answer: compare the aggregate pleasure and pain, and the pain-free option almost certainly wins, since memory loss is not itself painful. Preference satisfaction says there is no single answer, since which is better for a patient depends on that patient's preferences, and it makes the split rational rather than a sign that half of them are confused. An objective list theory says that memory and continuity of self are components of a good life independently of whether the patient wants them, so it can say that some patients are choosing wrongly, which is either its great strength or its great weakness depending on your view of paternalism. The example is worth holding on to because it shows the three theories are not equivalent formulations: they give different clinical advice.

Now you. A man raised in a rigidly hierarchical society sincerely prefers not to be educated and reports himself content. Which theories can say his life is going badly, and on what ground?

Objective lists

The third family says there are things that make a life go well whether or not they are wanted and whether or not they are pleasant. Derek Parfit set out this three-way division in an appendix to Reasons and Persons in 1984, and it has been the standard map since.

Typical lists include knowledge, achievement, friendship and love, autonomy, aesthetic experience and health. The theory's advantage is that it says the obviously right thing about the machine, about the deprived man, and about the person who spends a talented life counting blades of grass, which Rawls used as a test case.

Two objections. The first is that the list is unargued: any item can be questioned, and the theory has no principle generating them, so it looks like the theorist's own preferences promoted to metaphysics. Defenders reply that the list is arrived at by reflective equilibrium like everything else, and that the convergence across very different traditions on roughly the same items is evidence rather than coincidence.

The second is alienation, and Peter Railton stated it best in 1986: something cannot be good for a person if it leaves them entirely cold. A life of achievement that its owner finds hollow is not going well for them, whatever the list says. The usual response is a hybrid: an item counts towards a person's welfare only when they engage with it, so that the good life requires both the objectively worthwhile activity and the subjective endorsement of it. Most contemporary theories are hybrids of this kind, which is a sign of health rather than of confusion.

Adding it up

Whatever the good is, consequentialism has to sum it across people, and that requires welfare to be measurable on a scale where one person's gain can be compared with another's loss.

Nothing guarantees that. Ordinal preference, the kind economics is built on, tells you that I prefer A to B, and says nothing about by how much, and nothing at all about whether my gain from A exceeds yours from B. The interpersonal comparison the theory needs is an extra assumption, and Lionel Robbins argued in the 1930s that it is not scientific at all.

Practice proceeds anyway, and the results are not absurd. Health systems use the quality-adjusted life year, a year of full health scored at 1 and a year in a given health state scored between 0 and 1 by surveyed valuations. The National Institute for Health and Care Excellence in England has for two decades used a threshold in the region of 20,000 to 30,000 pounds per quality-adjusted life year when deciding what the health service should fund. That is interpersonal comparison, done officially, with a published number attached, and a system that refused to do it would still be making the comparison implicitly and worse.

So the practical objection is weaker than the theoretical one. What survives is that the comparisons are rough, contested, and much less precise than the arithmetic they feed, which should make anyone suspicious of an argument whose conclusion turns on a small difference in the sum.

Example. A quality-adjusted life year scores a year in a given health state between 0 and 1, using values obtained by asking survey respondents how much healthy life they would trade to avoid the state. What does that method assume, and where does it go wrong?

It assumes that a health state has a single value, that people can report it reliably, and that the reports of the general public are the right input. Each assumption has a known failure. People who have lived with a condition consistently rate it far less badly than people imagining it, a robust finding often called the disability paradox, so the choice of whose valuations to use changes the answer systematically; using the public's values discounts the lives of disabled people, and using patients' values risks the adaptive preference problem from earlier in this lesson. The method also treats a year of life as equally valuable to everyone, which is what makes the arithmetic possible and is itself a substantive moral choice, since it implies that saving a year for a young person and for an old one count the same while a life-extending treatment for the young buys more years. None of this makes the measure useless; it makes it a moral instrument rather than a measurement, and the practical response has been to publish the valuation method rather than to abandon the exercise.

Now you. Two treatments cost the same. One gives 20 people an extra year each in a state valued at 0.5. The other gives 5 people an extra year each in full health. Which does the arithmetic favour, and what does the comparison hide?

Population, and the conclusion nobody accepts

The worst problem in the theory of the good is not about individuals at all. It is about how many of them there should be, and it was Parfit who made it inescapable.

Total utilitarianism says to maximise the sum of welfare. Then consider a world A of ten billion people, all with excellent lives, and a world Z with a hundred thousand times as many people, each of whose lives is barely worth living: some music, some potatoes, and nothing else. Set the good life at 100 units and the barely-worth-living life at 0.01. World A totals ${10}^{10}\times 100={10}^{12}$. World Z totals ${10}^{15}\times 0.01={10}^{13}$, ten times as much. Total utilitarianism says Z is better, and by a wide margin.

Parfit called this the Repugnant Conclusion, and the mere addition paradox shows it is not easily escaped. Start with A. Add a separate group of people with lives clearly worth living but less good, without affecting anyone in A: this seems no worse, since nobody is harmed and the added lives are worth living. Now level the difference and share the resources, raising the worse-off and lowering the better-off with a net gain in total: this seems better. The result is a larger world with lives slightly less good than A's. Repeat, and each step is defensible while the destination is Z.

Switching to the average view avoids Z and buys worse problems. On the average view, adding a person whose life is worth living but below average makes the world worse, so a happy child born to very happy parents is a moral cost. The same view produces what Gustaf Arrhenius called the sadistic conclusion: where adding lives below the average is bad, it can be better to add a small number of people with lives of suffering than a large number with lives that are worth living but modest. And it makes the value of an act depend on people long dead, since they are in the average.

Arrhenius has proved a series of impossibility theorems showing that no population axiology can satisfy all of a short list of very plausible adequacy conditions. This is not a defect of utilitarianism specifically. It is a result about the structure of any view that ranks outcomes containing different numbers of people, and it is why the area is unsettled rather than merely difficult.

Example. Compute the totals for A and Z above, and say precisely which premise of the mere addition argument you would deny.

The totals are ${10}^{10}\times 100={10}^{12}$ and ${10}^{15}\times 0.01={10}^{13}$, so Z beats A by a factor of ten. To resist, you must deny one of three things. That adding people with lives worth living, harming nobody, cannot make an outcome worse: denying this means holding that a world can be made worse by the existence of a person glad to exist. That a more equal distribution with a higher total is better: denying this means rejecting either the priority of the worse off or the value of totals, at least in this application. Or that "better than" is transitive: denying this is drastic and some philosophers, including Larry Temkin, have accepted it precisely here. There is no comfortable option, and the standard of a good answer in this area is knowing which discomfort you have chosen.

Now you. A critical-level view says only lives above some positive threshold add value, so lives just barely worth living count negatively. Does that solve the problem?

The separateness of persons

One objection cuts across every version of the theory of the good, and it is the deepest.

Nozick, in the same book as the machine, imagines a utility monster: a being that derives enormously more welfare from any resource than anyone else does. Total utilitarianism says everything should be given to it, and everyone else should be sacrificed to its pleasures. The example is fanciful, and its point is structural. A theory that ranks outcomes by a sum has no way of objecting when the sum is maximised by a very unequal distribution, because the sum is blind to who holds the terms.

Rawls put the same charge without the monster in A Theory of Justice in 1971: utilitarianism does not take seriously the distinction between persons. It extends to society the principle of choice appropriate for one person, in which a present sacrifice for a future gain is straightforwardly rational, and in doing so treats separate people as though they were phases of a single life. My suffering can be compensated by your pleasure only if there is somewhere for the compensation to land, and there is not.

This is the objection the lesson on what we owe each other develops into a rival theory. Before that, the theory faces a set of more immediate embarrassments: cases where maximising the good requires an act that almost everyone regards as monstrous, and a level of demand that no life could sustain. That is the next lesson.

A theory that tells you to bring about the best outcome will sometimes tell you to kill someone, and it will always tell you that you have not yet done enough.

Those are the two great objections, and they are structurally different. The first says the theory permits too much: there are acts nobody may perform whatever the arithmetic. The second says it demands too much: a life lived by the theory would have no room in it for the person living it. The previous two lessons built consequentialism at full strength and left it needing a filling for "good" that nothing quite supplies. This lesson attacks the structure rather than the filling, and then examines the two standard repairs.

The transplant surgeon

Five patients in a hospital will die today without transplants: one needs a heart, two need kidneys, one a liver, one a lung. A healthy man comes in for a check-up. He is a tissue match for all five. The surgeon can kill him quietly, distribute the organs, and save five lives at the cost of one. Nobody will find out, so no wider effects on trust follow.

Every stipulation is there for a reason: the case is built to strip away every consequence a consequentialist might appeal to, so that only the arithmetic remains. Five is greater than one, so the theory says the surgeon should do it. Almost nobody believes that, including almost all consequentialists.

The case pairs with a second, from Philippa Foot's 1967 paper, in which a runaway trolley will kill five unless a driver steers it onto a track where it will kill one. There, most people think steering is permissible or required. Same numbers, opposite verdicts. Since a pure consequentialist theory sees only the numbers, it cannot make the pair come out differently, and any theory that gets both right must be sensitive to something other than outcomes.

What that something is has a name: a constraint, sometimes called an agent-relative restriction. It is a prohibition on the agent's doing something, which holds even when the agent's violating it would prevent more violations of the same kind. Constraints are strange when you look at them directly. If killing is bad, why may I not kill one to prevent five killings? The consequentialist finds this paradoxical, and is entitled to press it. The reply, developed in the lesson on duties, is that a moral requirement addresses the agent about what to do rather than instructing them to minimise an outcome, and that the wrong of killing is something you do to a person rather than a quantity in the world.

Example. A consequentialist replies that the surgeon case is unrealistic: in the real world secrets leak, trust in hospitals collapses, and fewer people attend check-ups, so the arithmetic actually forbids the killing. Is that a good reply?

It is true and it does not answer the objection. The case is stipulated to exclude those effects, and a theory that gets the right answer only because the stipulation is unrealistic is getting it for the wrong reason. Test the reply by strengthening the stipulation: suppose the surgeon will die tomorrow of an unrelated illness, so no discovery is possible. If the verdict now flips to permitting the killing, the reply was never doing philosophical work, it was doing empirical work that happened to point the same way. There is a further problem: someone who refrains from killing the man only because they calculate the wider effects has, on most people's view, the wrong relationship to the act altogether, which is the integrity objection below in miniature.

Now you. Is there a version of the transplant case where you would accept the killing? What does your answer commit you to?

Demandingness

The second objection needs no exotic case. Consequentialism is maximising, and there is always more good available, so on the theory almost every choice you make is wrong.

The money you spent on this year's holiday would have bought bed nets. So would the money you spent on a better flat, on your children's music lessons, on the coffee this morning. Since the theory is agent-neutral, the fact that it is your money and your children carries no weight of its own. The correct level of personal spending is whatever keeps you productive enough to give more, and everything above that is a wrong act.

Two consequences follow, and the second is less noticed. The first is that the demand is crushing: a life lived by the theory is one of permanent unrelieved obligation. The second is that the category of the supererogatory disappears. Ordinary moral thought has a large space for acts that are admirable but not required, and a maximising theory has none: the heroic donation is not generous, it is the minimum, and anything less is a wrong. A theory that cannot say a single true thing about generosity has lost something.

The standard reply is Samuel Scheffler's, in The Rejection of Consequentialism in 1982. He proposes an agent-centred prerogative: agents are permitted to give their own projects and attachments weight out of proportion to their impersonal value, up to some limit. The prerogative is a permission, not a constraint, so it blocks the demandingness objection without blocking the arithmetic where nothing personal is at stake. Scheffler himself notes the asymmetry it leaves: it is much easier to justify a prerogative, which relaxes a demand, than a constraint, which forbids an act that would make things better, and he declined to defend constraints on the same basis.

Williams on integrity

Bernard Williams, in his half of Utilitarianism: For and Against in 1973, aimed at something deeper than the demand, and his two cases are the most quoted in the literature.

George is a chemist, unemployed, with a family suffering for it. A job is available in a laboratory doing research on chemical and biological weapons. George opposes such weapons. If he refuses, the post will go to a colleague who is enthusiastic about the work and will pursue it more vigorously. On the arithmetic, George should take the job: fewer weapons will be developed and his family will be fed.

Jim, a botanist, stumbles into a South American village where a captain has twenty Indians tied up, about to be shot as a warning to protesters. As an honoured guest, Jim is offered a privilege: if he shoots one himself, the other nineteen go free. If he refuses, the soldiers do what they were going to do and all twenty die. On the arithmetic, Jim should shoot.

Williams's point is not that these answers are obviously wrong. He grants that in Jim's case the right answer is probably to shoot. It is that utilitarianism reaches them too easily, and in doing so misdescribes what is happening. The theory holds each of us responsible not only for what we do but equally for what we fail to prevent, which Williams calls negative responsibility, and the effect is that another person's projects, including the captain's, flow through you and determine what you must do. Your own commitments become one more input to a calculation performed from nowhere. But those commitments are what you are: a person with deep projects who is required to set them aside whenever the sums say so has been alienated, in a quite literal sense, from their own agency. That, and not the body count, is the integrity objection.

The best consequentialist replies do not deny the phenomenon but relocate it. Peter Railton's sophisticated consequentialism, in a 1984 paper called "Alienation, Consequentialism, and the Demands of Morality", argues that a person with genuine loves and commitments produces better outcomes than a calculating one, so the theory itself recommends being the sort of person Williams describes, and endorses that from the outside without requiring it as a motive. Whether that is a real answer or an elegant way of conceding the point is still argued.

Example. Distinguish what the transplant case and the George case each show, and say why a repair to one need not repair the other.

The transplant case is about permission: it says the theory allows an act that is forbidden, so it needs a constraint. The George case is about the structure of practical reasoning: it says the theory reaches its verdict by treating George's own convictions as one more variable, so it needs some account of why an agent's commitments have standing. These are separate defects. A theory could add constraints against killing and still require George to take the job, and a theory could grant an agent-centred prerogative and still permit the surgeon to kill. Scheffler's position is exactly the second combination, deliberately: he accepts the prerogative and rejects constraints, on the ground that a permission to attend to one's own life is easy to motivate while a prohibition on minimising harm is not.

Now you. A consequentialist says that Williams's objection is self-indulgent, since Jim's discomfort at killing one man is a poor reason to let nineteen die. What has been missed?

Rule consequentialism, and whether it collapses

The obvious repair is to evaluate rules rather than acts. An act is right if it conforms to a code of rules whose general acceptance would produce the best outcomes.

Brad Hooker's Ideal Code, Real World in 2000 is the developed version: the right code is the one whose internalisation by the overwhelming majority of everyone, in each new generation, has the highest expected value, counting the costs of teaching it. That formulation does real work. A code containing "surgeons may kill patients to redistribute organs" would be catastrophic if internalised, so the surgeon is forbidden. A code requiring maximal donation would be too costly to instil, so the demand is moderated. Constraints and prerogatives both fall out of the theory rather than being bolted on.

The classic objection, from David Lyons in 1965, is that the theory collapses into act consequentialism. Whenever a rule produces a worse outcome in some circumstance, a better rule is available: the original rule plus an exception for that circumstance. Iterate, and the ideal code becomes "do whatever produces the best outcome", at which point rule consequentialism is act consequentialism with extra steps.

Hooker's reply is that the collapse depends on a rule being allowed unlimited complexity, and that his formulation blocks it, because a code has to be internalised by ordinary people at a cost, and a code of a million exceptions cannot be taught. Complexity is therefore penalised by the theory's own standard, and the ideal code stabilises well short of the act-consequentialist limit.

That answer works, and it exposes the position's real difficulty, which is different. If the code is fixed by what it would be best for everyone to accept, then in a world where most people do not accept it, following it can knowingly produce a worse outcome than breaking it. Hooker accepts this: he holds that you should follow the ideal code even where deviating would do more good. But that is to abandon the thought that made consequentialism attractive in the first place, and the rule consequentialist is now open to the charge of rule worship, of preferring a code to the good the code exists to serve.

Two levels

Richard Hare's Moral Thinking, in 1981, proposes that the theory has been asked the wrong question. Moral thinking happens at two levels.

At the intuitive level, where nearly all of life is lived, we operate with firm dispositions and simple principles: keep your promises, do not kill, look after your children. These are held with feeling and resist calculation, which is what makes them reliable under pressure and in the face of self-serving reasoning. At the critical level, used in reflection and when intuitive principles conflict, we reason as act utilitarians about which set of intuitive principles to have, and about genuinely novel cases.

The archangel, Hare's idealised reasoner with perfect information and no partiality, would use critical thinking directly. The prole, with human limitations, does best on intuitions alone. Real people are in between, and the theory recommends critical thinking sparingly, because human beings who begin calculating in the moment mostly discover that the sums favour them.

The view absorbs a great deal: it explains why the surgeon should not calculate, why the intuitions Williams defends are worth having, and why a good person feels the force of a promise rather than weighing it. Its problem is what happens at the boundary. If critical thinking always overrides intuition when the two conflict, the intuitive level is only a heuristic and the theory has all the implications of act consequentialism for anyone reflective enough to reach the critical level. And if intuition sometimes overrides critical thinking, the theory owes a criterion for when, which it does not supply.

Example. A rule consequentialist and a two-level utilitarian both say the surgeon must not kill. Are they saying the same thing?

No, and the difference shows up under reflection. The rule consequentialist says the act is wrong, full stop, because it violates the ideal code, and would still be wrong if the surgeon reasoned about it perfectly and knew for certain that the killing would go undetected. The two-level utilitarian says that the surgeon should not kill because a disposition against it produces better outcomes overall, and that at the critical level, with full information and no risk of self-deception, the killing would be correct. The first gives a verdict on the act; the second gives a verdict on the disposition and leaves the act's status to the arithmetic. Someone who finds the second answer unsatisfying is registering that they wanted a constraint, not a well-chosen habit, which is what pushes readers towards the next lesson.

Now you. Does the demandingness objection apply to rule consequentialism?

Scoring the round

The theory is not dead, and the summary should be honest about what each objection establishes.

The demandingness objection is answerable, at the cost of adding a prerogative that does not come from the theory. The integrity objection is partly answerable, at the cost of a two-level structure that most people find alienating in a different way. The constraint objection is the one that has not been answered: no consequentialist theory has produced a principled prohibition on killing one to save five that survives the case being stipulated cleanly, and the honest consequentialist position is to accept the verdict and argue that the intuition is unreliable.

That is a real position and it has serious holders. What it means is that the disagreement has been located exactly, which is what the lesson on method promised. Either there are acts one may not perform whatever the outcome, or there are not. If there are, moral theory needs a source for them other than the good they produce.

The next lesson is the most thorough attempt anyone has made to supply one, and it starts from a surprising place: not from any claim about what is valuable, but from the bare requirement that your reasons be consistent.

If there are acts you may not perform whatever the outcome, they have to come from somewhere other than the good they produce, and the most ambitious attempt to find that source starts from nothing but the requirement that your reasons hang together.

That is Immanuel Kant's project in the Groundwork of the Metaphysics of Morals of 1785, and it is worth taking seriously even by readers who end up rejecting it, because it is the only major theory that tries to derive substantive moral conclusions without assuming anything about what is valuable. The previous lesson left consequentialism unable to explain constraints. This lesson builds a theory that has nothing but constraints, and the one after it deals with the mess that creates.

Starting from the good will

Kant opens with a claim that sounds extravagant and is doing precise work. Nothing in the world, or out of it, can be called good without qualification except a good will.

The argument is by elimination. Intelligence, courage, wealth, and even happiness are good in ordinary circumstances and bad in the hands of a villain: a courageous and resourceful criminal is worse than a cowardly one, so those qualities cannot be unconditionally good. What is left is the quality of the willing itself. A good will is not good because of what it achieves, since achievement depends on luck and circumstance; it is good in itself, in the way a jewel is, shining by its own light even if it accomplishes nothing.

This yields an immediate and much misunderstood distinction. An act done in conformity with duty may be done for any reason: the shopkeeper who does not overcharge inexperienced customers may simply be protecting his reputation, and his honest dealing has no moral worth even though it is the right action. An act done from duty is done because it is right. Kant's hard case is the man who has lost all taste for life and preserves it anyway because he sees that he must; his act has moral worth precisely because no inclination supports it.

Kant is not saying that acting from inclination is wicked, nor that the good person should feel nothing. He is saying that if you want to isolate what moral worth consists in, you have to find a case where nothing else could explain the act, and the joyless man is that case. Friedrich Schiller's satirical couplet, that he gladly serves his friends but does so with pleasure and is therefore troubled that he is not virtuous, is funny and misses the point.

Hypothetical and categorical

Everything else follows from a distinction about the form of practical requirements.

A hypothetical imperative tells you what to do given an end you have: if you want to pass the exam, study. Its authority is borrowed from the end, and it evaporates if you drop the end. Nobody who abandons the exam is still bound to study.

A categorical imperative tells you what to do regardless of what you want. Its authority is not borrowed. Kant's claim is that moral requirements must have this form, because a requirement that lapsed as soon as you stopped wanting something would not be a moral requirement at all: "do not torture people if you want to be well thought of" is prudence, and if it were the whole story then the person who does not care about being well thought of has no reason not to torture. Philippa Foot challenged exactly this in a 1972 paper, arguing that morality might be a system of hypothetical imperatives after all, and that the appearance of categoricity comes from etiquette-like conventions of speech. That challenge is live, and the rest of Kant's system depends on it failing.

Given the distinction, Kant asks what content a categorical imperative could have. It cannot get its content from any end, since an end would make it hypothetical. All that is left is its form: that it be a law, meaning something universal. Hence the first formulation: act only in accordance with that maxim through which you can at the same time will that it become a universal law.

Working the test

A maxim is the principle you would be acting on, stated as an intention: in circumstances C, I will do A, in order to achieve E. The test asks you to imagine the maxim as a universal law of nature, holding for everyone always, and see whether something goes wrong.

Kant identifies two ways it can go wrong, and keeping them apart is the whole technical content of the theory.

A contradiction in conception occurs when the universalised maxim cannot even be coherently thought. Kant's example is false promising. The maxim is: when I need money, I will borrow it on a promise to repay that I have no intention of keeping. Universalise it. Everyone in need makes lying promises, so promises to repay carry no information about repayment, so nobody accepts them, so the institution of promising to repay does not exist, so the maxim cannot be acted on: there is no promise for me to make falsely. The maxim destroys the very practice it depends on for its success. This yields a perfect duty, one that admits no exceptions and no discretion.

A contradiction in the will occurs when the universalised maxim can be conceived without incoherence but cannot be willed by a rational agent. Kant's example is refusing all beneficence. The maxim is: I will never help anyone in need. A world in which nobody helps anyone is perfectly conceivable, and would function. But you cannot consistently will it, because you are a finite creature who will certainly need help, and in willing the universal law you would be willing away the assistance you cannot rationally renounce. This yields an imperfect duty, one that requires adopting an end, here the happiness of others, without specifying which acts discharge it, so there is latitude in how and when.

The four examples in the Groundwork fill out the grid: against suicide and against false promising as perfect duties, to develop one's talents and to help others as imperfect ones.

Example. Apply the test to riding a train without paying, on the maxim: when I can avoid the fare without being caught, I will.

Universalise it and ask which contradiction, if any, arises. If everyone who could evade the fare did, the fare system collects nothing from anyone able to evade it, the service cannot be funded on fares, and either it closes or it becomes free. Either way there is no fare to evade, so the maxim cannot be acted on in the universalised world: a contradiction in conception, and therefore a perfect duty not to do it. This is the structure Kant thinks all free riding has, and it is worth noticing how different the verdict's basis is from the consequentialist one. A consequentialist says one evaded fare costs the railway almost nothing, so the act is nearly harmless and may well be permissible. Kant says the act is impermissible whatever it costs, because what makes it wrong is that you are making an exception of yourself, helping yourself to a practice on terms you could not extend to everyone.

Now you. Now try the maxim: when I am short of money, I will sell my labour to whoever pays best. Does the test forbid it?

What the test gets wrong

The universal law formula has two well-known failure modes, and any honest presentation includes them.

The first is false positives generated by irrelevant detail. Take the maxim: when I want to relax, I will play tennis at four on Thursday afternoons. If everyone played tennis at four on Thursday afternoons, there would be no courts free and nobody to sell the balls, so the maxim seems to fail the conception test, and tennis is forbidden. Something has gone wrong. The standard diagnosis is that the maxim has been described at the wrong level: it should read "when I want to relax, I will pursue a recreation", and the specific time is a circumstance of acting rather than part of the principle. The trouble is that Kant gives no rule for describing maxims at the right level, and without one the test can be made to deliver almost any verdict by adjusting the description. This is the standard objection, and it is serious.

The second is false negatives. The maxim "when someone threatens my life, I will kill them if I can" universalises without contradiction, and so does "when I encounter someone of a group I despise, I will refuse to trade with them". Universalisability catches parasitic maxims and misses maxims that are consistently and universally cruel. Kant needs a further formulation to reach those, and he supplies one.

Example. A shopkeeper acts on the maxim: when a customer cannot tell the difference, I will sell the inferior item at the price of the better one. Work the test, then state the maxim at a different level of description and see what changes.

Universalised, every seller substitutes the inferior item whenever a buyer cannot tell. Buyers then know that price carries no information about quality, so nobody pays the higher price, so there is no higher price to obtain by substitution: a contradiction in conception, and the act is forbidden. Now redescribe the maxim as "when trading, I will maximise my return", which universalises perfectly well and is permitted, and as "on Tuesdays I will sell tinned goods", which is trivially permitted. The three descriptions are all true of the same act, and only one of them delivers the right verdict. What separates them is that the middle one omits the feature the wrongness depends on, the exploitation of the buyer's ignorance, while the last adds features that are irrelevant. The working rule most Kantians use is that a maxim must include exactly the circumstances and the end that explain why the agent is doing it, and no more, which is workable in practice and is not a rule Kant gives.

Now you. Does the universal law test forbid a doctor from becoming a doctor if everyone becoming a doctor would leave nobody to grow food?

The formula of humanity

Kant claims the second formulation is equivalent to the first, which nobody quite believes, and it is the one most readers find compelling: act so that you use humanity, in your own person as well as in the person of any other, always at the same time as an end and never merely as a means.

The word doing the work is merely. You use the bus driver as a means to get across town, and there is nothing wrong with that. What is forbidden is treating a person purely as an instrument, in a way that bypasses their status as a rational agent with ends of their own.

Kant's underlying argument is that rational nature exists as an end in itself. Everything else in the world has value because someone values it, and so has a conditional worth; a rational being is the source of value rather than a repository of it, and cannot be assigned a price without absurdity. Hence the distinction between price and dignity: whatever has a price can be replaced by something equivalent, while whatever is above all price, and admits of no equivalent, has dignity.

The formula gives an operational test that has proved more durable than the universalisation one. To treat someone merely as a means is to act towards them in a way they could not, in principle, consent to, because you have concealed from them the very thing they would need to know in order to decide. Deception and coercion are the paradigms, and this is why both are so central to Kantian ethics: the deceived person cannot share your end because they do not know it, and the coerced person cannot share it because their agreement was not theirs to give. Notice how directly this answers the transplant case of the previous lesson. The surgeon uses the healthy man's body as a resource, on a plan he could not possibly consent to. The five patients' need does not enter, because the wrong is done to him and is not a quantity to be traded off.

Example. A researcher enrols patients in a trial without telling them that half will receive a placebo, reasoning that the knowledge would distort the results and that the trial will benefit thousands.

The formula of humanity forbids it, and the reasoning is exact. The patients are being used as instruments in a project whose actual nature is hidden from them, and the concealment is precisely of the fact they would need in order to decide whether to participate. They cannot share the researcher's end because they have not been told what it is. That the trial benefits thousands is irrelevant, because the wrong consists in what is done to these people and not in a balance of goods. Notice that this is not an objection to placebo trials as such: a trial in which participants are told they will be randomly assigned and may receive a placebo is entirely permissible, because they have consented to that arrangement in full knowledge, and the deception of the individual patient is replaced by a procedure they endorsed. Informed consent, which is now the governing principle of research ethics worldwide, is a direct institutional descendant of this formula.

Now you. A friend asks whether you like her new business plan. You think it is doomed, and you say it is promising, to spare her feelings. Which formulation does this offend, and how would each theory so far assess it?

Autonomy and the kingdom of ends

The third formulation gathers the others. Act as though you were through your maxims a lawgiving member of a kingdom of ends: a community of rational beings, each of whom legislates the moral law and each of whom is subject to it.

This is where Kant's account of freedom sits, and it connects directly to the previous course. For Kant, to be free is not to be uncaused; it is to act on a law you give yourself rather than being pushed around by inclinations you did not choose. Heteronomy, acting from desires that happen to be in you, is a kind of unfreedom even when nothing external constrains you, and autonomy is obedience to a law of your own making. The moral law and the free will turn out to be the same thing described twice.

Two consequences of this are worth carrying forward. Moral requirements apply equally to everyone, because they are derived from rational agency alone and every rational agent has that in the same degree, which grounds a strong doctrine of equal moral status with no reference to anyone's capacities or usefulness. And the theory is genuinely non-consequentialist all the way down: the reason not to use a person is not that a world with more using in it is worse, but that this person is not the kind of thing that may be used.

What the theory buys, and what it costs

Set against the previous lesson's scoreboard, the gains are large. Kantian ethics has constraints, and they are not bolted on: the wrongness of killing, deceiving and coercing follows from the same formula. It explains the intuition about the transplant surgeon exactly, and explains why the numbers do not help. It grounds equal moral status, and it delivers the whole modern apparatus of consent. It never demands that you sacrifice your projects to an aggregate, since your own humanity is an end too.

The costs are equally large and mostly show up as rigidity. The theory says nothing about degrees, so a small lie and a great betrayal both violate a perfect duty. It has no way to prioritise, so a duty of beneficence with no specification of how much yields little practical guidance. And it appears to forbid lying to a murderer at the door, which Kant, asked exactly that question, confirmed in print.

That last is where the tradition had to do repair work, and it is the subject of the next lesson: what happens when two duties both apply, whether any duty is genuinely absolute, and whether the distinction between what you intend and what you merely foresee can carry the weight it is asked to bear.

A theory built entirely out of constraints has a problem the moment two of them apply at once, and it has a worse problem when obeying one produces a catastrophe.

The previous lesson built Kant's theory and left it with three unpaid bills: it says nothing about degrees, it gives no way of ranking duties against each other, and it appears to forbid lying to a murderer. This lesson pays them, or tries. What emerges is a version of deontology most working philosophers actually hold, which is considerably less rigid than Kant's and correspondingly less tidy.

The murderer at the door

In 1797 Benjamin Constant published an essay attacking the doctrine that it is a duty to tell the truth, using a case he attributed to a German philosopher: a murderer asks you whether your friend, whom he intends to kill, is hiding in your house. Constant's point was that a duty of truthfulness held without exception would make society impossible.

Kant replied the same year, in a short and startling essay, "On a Supposed Right to Lie from Philanthropy", and he did not soften. Truthfulness in statements is an unconditional duty. To lie is to wrong humanity in general, because it undermines the reliability of declarations on which all contracts and rights depend, so a lie is a wrong even when it harms no particular person.

His argument for the practical verdict is more interesting than the doctrine. If you tell the truth and your friend is killed, you are not responsible: you did what the law required and the murderer did the rest. If you lie, you take the consequences upon yourself, and Kant supplies the scenario in which this bites. You say your friend is not at home, believing him upstairs. Unknown to you he has slipped out of the back door. The murderer leaves, meets him in the street, and kills him. Had you told the truth, the neighbours might have caught the murderer searching your house, and your friend would have lived. You cannot know how a lie will run, and by lying you have made yourself answerable for whatever follows.

The argument has a clear weak point, and it is worth naming precisely: exactly the same unpredictability attaches to telling the truth. Kant's asymmetry does not come from the epistemology at all; it comes from the prior claim that a lie is a violation of duty and a truth is not, so responsibility transfers only in one direction. As a defence of the doctrine it is therefore circular, and as an argument from consequences it is unsound.

Modern Kantians almost all reject the verdict while keeping the framework. Christine Korsgaard's 1986 treatment is the standard one. The formula of humanity forbids treating people in ways they could not consent to, and consent has already been made impossible here by the murderer, who has stepped outside the conditions under which the kingdom of ends operates. In an ideal world of rational agents, lying is never necessary and never permitted; in a world containing someone who is using you as an instrument for murder, the ideal principle does not straightforwardly apply, and the right response is the one that best approximates respect for everyone's agency, which is to lie. This is a two-level structure of a kind, and it concedes what Kant would not: that what a duty requires can depend on what other people are doing.

Example. State Kant's argument as premises and identify which line the modern Kantian rejects.

(1) Truthfulness in declarations is an unconditional duty. (2) Anyone who acts within their duty is not responsible for the harms that follow. (3) Anyone who departs from duty becomes responsible for whatever follows from the departure. (4) Therefore you should tell the truth, and are blameless if your friend dies. The modern Kantian rejects (1), and the interesting part is the ground. They do not reject it by appealing to consequences, which would abandon the theory. They reject it by arguing that the duty's content is fixed by what respect for rational agency requires, and that respecting the agency of a murderer bent on using you is not achieved by handing him accurate information. Notice that (2) and (3) can then be left standing, which is why the repair is a repair rather than a defection.

Now you. Does the same reasoning license lying to a friend to protect their feelings?

Ross and pro tanto duties

The tidier solution abandons absoluteness at the outset. W. D. Ross, in The Right and the Good in 1930, argued that Kant's error was to treat duties as exceptionless when they are better understood as claims that can be outweighed.

Ross lists seven, presented as a provisional inventory rather than a derivation: fidelity, which covers promises and truthfulness; reparation, for wrongs one has done; gratitude, for benefits received; justice, in the distribution of goods according to merit; beneficence, improving others' condition; self-improvement; and non-maleficence, not harming others, which he singles out as generally weightier than beneficence.

The key term is prima facie, which Ross regretted and which most philosophers now render as pro tanto: a genuine moral consideration with real weight, which can be outweighed by another without disappearing. When you break a trivial promise to save a life, the promise does not evaporate. It is overridden, and the residue shows: you owe an explanation and an apology, which you would not owe if the duty had simply lapsed. That residue is the best evidence Ross has, and it is good evidence, because no theory on which the promise ceases to apply can account for it.

Example. A therapist's patient discloses a credible intention to seriously harm a named person. Which of Ross's duties are engaged, and what does his framework deliver?

Fidelity is engaged, and strongly: confidentiality is a promise, made explicitly at the start of treatment, and the whole practice depends on it. Non-maleficence is engaged in an unusual way, since the harm would be done by the patient rather than by the therapist, so what is really engaged is beneficence towards the third party, plus a duty of justice. Ross's framework says that fidelity is a real duty that does not disappear, that non-maleficence is generally the weightier kind of consideration, and that here the duty proper is to warn, because a serious threat to life outweighs a promise of confidence. It also says the residue is real: the therapist has broken a promise, owes the patient an account of why, and should have made the limits of confidentiality clear in advance, which is exactly what professional codes now require. The framework has not calculated anything, and it has organised the case correctly and identified what remains owed to the person whose claim was outweighed. Notice that the law reached the same place by a different route in Tarasoff in 1976, and that neither the court nor Ross could state a rule that would settle the next case.

Now you. A junior researcher discovers that a paper from their own laboratory contains fabricated data. Which duties conflict, and what does the residue look like?

The objection is the obvious one. Ross gives no ranking and no method: when fidelity conflicts with non-maleficence, what settles it? His answer is that judgement settles it, that the conflict is a matter for perception rather than calculation, and that moral knowledge here is like knowing which of two propositions is more probable when no rule applies. Critics call this an admission of defeat. Sympathisers reply that the alternative theories only appear to have a method, since consequentialism's arithmetic runs on judgements about value that are no better grounded, and that a theory which is honest about where judgement enters is preferable to one that hides it. Either way, this is where most contemporary non-consequentialists actually stand.

Thresholds

Between Kant's absolutism and Ross's weighing sits a third position, threshold deontology, which most people hold without knowing it has a name.

The claim is that constraints are not merely weighty considerations but genuine prohibitions, which nonetheless give way when the stakes pass some very high threshold. Torturing one person is forbidden to save five lives, and permitted to save a city. Michael Walzer's "supreme emergency" doctrine, defending the bombing of German cities in 1940 and 1941 but not after, is a threshold view applied to war.

Two objections are standard. The first is arbitrariness: no one can say where the threshold sits, and there is no principle to derive it from. The second is sharper and structural. If the threshold is at 1,000 lives, then at 999 the act is absolutely forbidden and at 1,001 it is permitted, which makes an enormous moral difference turn on two deaths, when the same two deaths make no difference anywhere else on the scale. Attempts to smooth the discontinuity, by letting the constraint's strength scale with the stakes, tend to turn the view back into a weighted consequentialism where constraints are just heavy considerations, which is Ross.

Nobody has solved this. What can be said for the view is that it matches how almost everyone reasons, and that a theory whose competitors are absolutism and pure aggregation is entitled to some patience.

The doctrine of double effect

The most powerful tool in the non-consequentialist kit is a distinction not between doing and allowing but between what you intend and what you merely foresee.

The doctrine has its origin in Thomas Aquinas's discussion of killing in self-defence in the Summa Theologiae, where he observes that one act can have two effects, of which only one is intended, and that the killing of the assailant is outside the defender's intention. Its modern formulation sets four conditions. The act itself must not be wrong independently of its consequences. The bad effect must not be the means by which the good effect is achieved. The agent must intend the good effect and merely foresee the bad one. And there must be proportionality between the two.

The second condition is the one that does the discriminating, and the classic pair shows it. A pilot who bombs a munitions factory knowing that civilians nearby will die intends the destruction of the factory; the deaths are a foreseen side effect, and if the factory were somehow destroyed with no deaths the mission would have succeeded completely. A pilot who bombs a residential district in order to break civilian morale intends the deaths; if the civilians all survived, the mission would have failed. The outcomes may be numerically identical. The doctrine says the first can be permissible and the second cannot.

The same structure runs through medical ethics, where it is doing daily work rather than sitting in a seminar. Palliative sedation that hastens death while relieving pain is permitted, on the grounds that the relief is intended and the shortening foreseen, while administering the same drug in order to end the patient's life is not. Removing a cancerous uterus from a pregnant woman is permitted, while crushing the skull of a fetus to save the mother has traditionally been forbidden, because there the death is the means.

The objections are serious. Intentions are hard to identify and easy to redescribe, a difficulty sharpened by Jonathan Bennett's observation that the terror bomber can claim to intend only that the civilians appear dead for long enough to break morale, which seems to launder the intention without changing anything. There is also a question about why the agent's own mental state should determine what is permissible for them, when the victim is equally dead either way. And the doctrine's verdicts sometimes look like the answer arriving before the reasoning. Its defenders reply that intention is central to how we assess agents everywhere in law and life, that the redescription trick fails because the terror bomber's plan requires the deaths to be real, and that the doctrine's medical applications are precisely where a bright line is most valuable.

Example. Apply the four conditions to a doctor who gives a dying patient a morphine dose large enough to control pain, knowing it will probably shorten life by hours.

The act, administering an analgesic, is not wrong in itself, so the first condition is met. The pain relief is not achieved by means of the shortening of life; the drug relieves pain directly and the respiratory depression is a separate effect of the same dose, so the second is met. The doctor intends the relief and foresees the shortening, which the second condition supports rather than merely asserting, since a dose that relieved pain without depressing respiration would satisfy her completely. And a few hours of life against severe pain at the end of it is a proportionate trade, meeting the fourth. So the doctrine permits it, which is also the settled position of medical bodies and, in most jurisdictions, of the law. Change one thing, that the doctor's aim is to end the patient's life and the pain relief is incidental, and the same injection becomes impermissible on this doctrine, which is what makes euthanasia a separate legal question from palliative care.

Now you. In Thomson's loop variant, the trolley is diverted onto a side track that curves back to the main line, and it is only the body of the one man that stops it reaching the five. What does this do to the doctrine?

What the trolley literature establishes

The trolley cases have generated more work than any other thought experiment in ethics, and it is worth being clear about their status.

Foot introduced the first case in 1967, in a paper about abortion, to illustrate the difference between what an agent does and what an agent allows. Judith Jarvis Thomson developed it in 1976, replacing the driver with a bystander at a switch and adding the man on the footbridge, and in 1985 added the loop. The pair of switch and footbridge is the engine: identical numbers, near-universal disagreement in the verdicts, and the question of what makes the difference.

The empirical work confirms the pattern is real and widespread. Marc Hauser and colleagues, testing large online samples in 2007, found approval of diverting the trolley running around 89 percent and approval of pushing the man around 11 percent, with the gap holding across countries, religions and levels of education. In the same study, a majority of respondents could not produce a justification for their own pair of answers, which is a genuinely important finding: the judgements are stable and the reasons offered for them are not.

The 2018 Moral Machine study, which collected close to 40 million decisions from participants in 233 countries and territories on autonomous vehicle dilemmas, found broadly shared preferences for sparing more lives, humans over animals, and the young over the old, with substantial regional variation in how strongly each was held.

What none of this establishes is which answer is correct. A survey measures what people judge, and the previous lessons' warnings about framing effects apply in full. What it does establish is that the switch and footbridge verdicts are not artefacts of a few philosophers' intuitions, so a theory that treats them as identical owes an explanation of a very robust discrepancy, and a theory that separates them owes an account of the difference that survives the loop.

Where deontology stands

The honest summary is that the theory in its Kantian form does not survive contact with conflicts, and the repairs are all partial.

Ross gives up systematicity and keeps everything else, which is why he is popular and why his position is sometimes accused of being a list rather than a theory. Threshold views keep constraints and cannot locate the threshold. Double effect gives a real criterion and creaks under redescription and under the loop. None of this refutes the family: a theory with real problems can still be closer to the truth than an alternative with different problems, and the constraints these theories are trying to capture remain something consequentialism cannot deliver at all.

What both families share is a picture in which morality is a matter of getting the right verdict on an act. There is an older tradition that thinks this is the wrong question altogether, and asks instead what kind of person to be, on the grounds that most of moral life consists not in deciding hard cases but in noticing what is at stake and caring about it. That is the next lesson.

Both theories so far treat a moral agent as a device for producing verdicts on acts, and an older tradition thinks that is the wrong picture of moral life.

Its complaint is easy to state. Most moral failure is not a failure of judgement at the point of decision. It is a failure to notice that anything was at stake, or to care once it was noticed, or to want the right thing when the moment came. A person who works out from first principles that they should visit a sick friend, and does so grudgingly, has done something less good than a person who simply wanted to go, and no theory that scores acts alone can say why. The previous lessons left consequentialism unable to supply constraints and deontology unable to resolve conflicts. This tradition offers a diagnosis of why both keep failing: they are answering a question that was never the primary one.

The question changed

Elizabeth Anscombe reopened the field with a paper in 1958, "Modern Moral Philosophy", which is short, aggressive and among the most consequential things written in the subject in the last century. She also coined the word "consequentialism" in it, as a term of abuse.

Her central charge is historical and conceptual. The distinctively moral "ought", the one that carries a special verdictive force, is a survivor from a legal conception of ethics in which God legislated. Remove the lawgiver, as modern philosophy did, and the notion of a moral law without anyone to make it is left hanging, doing work its foundations no longer support. Her recommendation was to stop doing moral philosophy until we had an adequate philosophy of psychology, and in the meantime to work with concepts that do not need the legal framework: what a person is like, what they are doing, whether they are just or cowardly or truthful.

The tradition she was pointing back to is Aristotle's, principally the Nicomachean Ethics, written in the fourth century BC. Its question is not what to do on a particular occasion but what a good human life consists in, and it takes the practical question to be answered by the sort of person one has become rather than by a procedure applied at the moment of choice.

Eudaimonia and the function argument

Aristotle begins where a modern reader least expects: with the claim that every activity aims at some good, and that there must be some end pursued for its own sake, on pain of an infinite regress that would make all desire empty. That end everyone names eudaimonia, usually translated as happiness and better rendered as living well or flourishing.

Translation matters here. Eudaimonia is not a feeling. It is an activity, not a state; it is assessed over a complete life rather than at a moment; and it is something you can be wrong about, in a way you cannot be wrong about whether you feel cheerful. Aristotle's remark that one swallow does not make a summer, nor does one day, is a claim about the kind of thing it is.

The naming settles nothing, since everyone agrees on the word and disagrees about the content, so Aristotle offers an argument. A flute player, a sculptor and a carpenter each have a characteristic activity, and each is good insofar as they perform it well. Does a human being as such have one? He argues by elimination: not mere living, which plants share; not perception, which animals share; what is left is activity of the part of the soul that has reason. So the human good is activity of the soul in accordance with virtue, and if there are several virtues, in accordance with the best and most complete, over a complete life.

The argument is the most contested passage in the book and its weakness should be stated plainly. Two objections stand out. That a species has a distinctive capacity does not show that exercising it is good for its members, so the move from "characteristic" to "good" needs a premise that is not supplied. And picking out reason as the distinguishing feature is a choice among many possible ones. What survives is more modest and still substantial: an account of a good life must be grounded in facts about the kind of creature we are, and cannot be read off a set of rules with no anthropology behind it. Contemporary neo-Aristotelians, notably Philippa Foot in Natural Goodness in 2001, rebuild the argument from claims about what a living thing of a given kind needs in order to flourish, which is a naturalistic strategy with its own problems and a much better relationship with biology.

Virtue as a mean

A virtue, on the account that follows, is a hexis: a settled disposition, acquired rather than innate, concerned with choice, which disposes a person to feel and act in the right way, at the right time, towards the right people, for the right reasons.

The feeling is not decoration. On this view a person who wants to steal and refrains is not generous but continent, and continence is a lesser condition than virtue: the fully virtuous person has brought their desires into line so that doing the right thing is not a struggle. This is one of the tradition's sharpest departures from Kant, for whom the man who acts against inclination shows moral worth most clearly.

The doctrine of the mean says that each virtue lies between an excess and a deficiency in some sphere of feeling or action.

Two qualifications keep this from being trite. The mean is relative to the person and the situation, not an arithmetic midpoint: the right amount of food for a wrestler is not the right amount for a child, and the courageous act in one circumstance is retreat. And not every action or feeling has a mean. Aristotle names spite, shamelessness, envy, adultery, theft and murder as things that are bad in themselves, with no question of doing them at the right time in the right amount. The doctrine is a heuristic about spheres in which excess and defect are both possible, not a universal formula.

Example. Locate the vices flanking honesty in the sense of telling people hard truths, and say what the mean depends on.

The deficiency is a habit of concealment and flattery, saying what will please rather than what is so, which the earlier lesson's business plan case illustrates. The excess is not honesty overdone in the sense of too much truth, since truth cannot be excessive, but a habit of unnecessary and gratuitous disclosure: telling a grieving man that his wife's illness was mismanaged when nothing can be done about it, or announcing every judgement one forms about a colleague. What the mean depends on is who is being told, what they can do with the information, what standing you have to give it, and what it is for. That list is exactly the point of the doctrine: it says the answer is not a rule but a set of factors, and that getting it right is a skill. Notice that a consequentialist reaches the same verdicts here by asking what the disclosure achieves, and a Kantian by asking whether the person is being managed rather than addressed. Convergence in ordinary cases is normal, and the theories separate at the edges.

Now you. Is there a mean of which loyalty is the virtue, and what does the case show about the doctrine?

How virtue is acquired

Virtues are not innate and not taught by instruction. Aristotle's claim is that we acquire them by practice: we become builders by building and harpists by playing the harp, and likewise we become just by doing just acts, temperate by doing temperate acts, brave by doing brave acts.

This looks circular. How can you do just acts before you are just? Aristotle's answer distinguishes the act from the way it is done. A learner can perform the act a just person would perform, without yet doing it as a just person does, which requires knowing what one is doing, choosing it for its own sake, and acting from a firm and unchanging state. The learner practises the outward act under guidance, and the disposition forms through repetition, in the way a musician's ear forms through playing. This is why the tradition takes moral education, upbringing and the example of admired people so seriously, and it is a genuine advantage: neither of the other theories has much to say about how anyone comes to be moral, and both quietly assume an agent who already cares.

It also implies that character formation runs largely below the level of decision, through what one is habituated to notice, feel and enjoy. That claim is empirical, and the next lesson is about how it has fared.

Example. A school wants to reduce bullying. Compare what each of the three traditions recommends.

A consequentialist looks for the intervention with the best measured effect on the outcome, which points to whole-school programmes with supervision, clear reporting and consistent response, and treats the children's characters as one causal factor among several. A Kantian frames the wrong precisely, as the treatment of a classmate as a thing rather than as a person with ends of their own, and builds the response around that: rules, their justification, and the requirement that every child be able to see why the rule binds them. A virtue ethicist asks what dispositions the school is actually cultivating, which turns attention to things the other two do not look at: what conduct is admired and by whom, what the adults model when they are tired, whether the ordinary texture of the place rewards contempt or makes it embarrassing, and whether children are given practice at intervening rather than instruction about it. All three have something, and the third is describing the mechanism by which anything the first two do will or will not stick.

Now you. Does the habituation account imply that a person raised badly cannot become good?

Practical wisdom

The component that does the real work is phronesis, practical wisdom: the capacity to perceive what a particular situation calls for and to act accordingly. It is not cleverness, which is a capacity to hit any target, and it is not scientific knowledge, which is of what cannot be otherwise.

Aristotle's argument for putting perception rather than rules at the centre is worth reconstructing, because it is the tradition's strongest move. Practical situations are indefinitely various, and any rule stated in advance covers them by generalising over features that its author anticipated. When an unanticipated case arrives, the rule either falls silent or gives the wrong answer. His image for the remedy, from the discussion of equity, is the leaden ruler used by builders on Lesbos, which bends to fit the shape of the stone instead of forcing the stone to fit the rule. A decree has to be adapted to the facts in the same way.

This has a modern echo in every field where expertise resists codification. An experienced clinician recognises a presentation that no protocol quite matches; an experienced teacher sees which child needs to be pushed and which needs to be left alone. In each case the expert can rarely state the rule they are following, and attempts to write it down produce something that performs worse than they do. Virtue ethics claims that moral competence is of this kind, which is why it locates moral knowledge in a trained perceiver rather than in a principle.

Practical wisdom also unifies the virtues. Aristotle holds that you cannot fully have one virtue without the others, because each requires knowing how much weight the considerations of its sphere carry against the others, and that knowledge is the same knowledge. The unity thesis is widely thought too strong, since people plainly are brave and unkind, and the defensible version is that the virtues are interdependent enough that a serious deficiency in one distorts the exercise of the rest.

Example. A manager discovers that a well-liked employee has been falsifying expense claims for small amounts over two years. What does each of the three theories ask?

A consequentialist asks what will produce the best outcome: what dismissal does to the team, what tolerance does to the incentives, whether prosecution helps anyone. A Kantian asks what is owed: the employee has treated the firm and colleagues as means, honesty is a perfect duty, and the manager has duties of fairness to everyone else that the employee's popularity cannot override. A virtue ethicist asks what a person of good character does here, and the question opens up material the others miss: whether the manager's own reluctance is compassion or cowardice, whether the two years of not noticing was a failure of attention that is itself a fault, what a just response looks like when it must be both firm and not gratuitously destructive, and how to conduct the conversation. That last cluster is not decoration. Most of what goes wrong in real cases like this goes wrong in the manner rather than in the verdict, and only one of the three theories has anything to say about the manner.

Now you. Rosalind Hursthouse defines a right action as what a virtuous agent would characteristically do in the circumstances. What is the obvious objection, and does the definition survive it?

What the tradition contributes

Even a reader who ends up a consequentialist should take three things from this lesson.

The first is that moral perception precedes moral judgement. Before you decide what to do about a situation you have to see it as raising a moral question at all, and most failures happen there, invisibly, because nothing ever presented itself as a decision. Iris Murdoch made this her central theme: the work of morality is largely the work of attention, of seeing another person justly, and by the time a choice arrives most of the work is already done or already lost.

The second is that emotions are part of moral competence rather than interference with it. Feeling indignation at cruelty and affection for one's friends is not a distortion of a calculation; it is what a well-formed person's perception consists in, and a person who reached the right verdict with no feeling at all would be defective rather than exemplary.

The third is that moral education is a real subject. If character is built by habituation, then what children are given to admire, practise and enjoy matters more than what they are told, which is a testable claim with practical consequences.

That claim, and the whole account of character it rests on, is an empirical bet: that there are stable traits which explain and predict what people do. A large experimental literature says that bet is much shakier than Aristotle assumed, and the next lesson takes it seriously.

Virtue ethics rests on an empirical bet: that people have stable character traits which explain what they do, and that these traits carry across situations.

If that is false, the theory is in trouble in a way its rivals are not. Consequentialism and deontology can survive the discovery that human character is thin, because they tell you what to do rather than what to be. A theory whose central concept is a settled disposition needs settled dispositions to exist. This lesson is the empirical challenge, and it is unusual in a philosophy course: the evidence is experimental, the numbers matter, and both the challenge and the reply have been damaged by problems in the underlying research that only came to light recently.

The bet, stated precisely

The claim under test is not that people differ. Obviously they do. It is what John Doris calls globalism: that character traits are consistent across a wide range of situations, that they are stable over time, and that they are integrated, so that possessing one virtue predicts possessing others.

Cross-situational consistency is the crucial part, because it is what makes a trait explanatory. If honesty is a trait, then a person who does not cheat on a test should also not lie about their expenses and not steal from a till, and knowing that someone is honest should let you predict what they will do somewhere you have not seen them. Aristotle's account requires exactly this: a virtue is a firm and unchanging state, and its possessor acts well reliably rather than occasionally.

What the personality data says

The first serious test predates the philosophical argument by seventy years. Hugh Hartshorne and Mark May ran the Character Education Inquiry between 1928 and 1930, testing several thousand schoolchildren for honesty across many separate opportunities: copying answers, lying about it afterwards, taking money from a puzzle box, falsifying a self-scored test. If honesty were a trait, performance on one test should predict performance on the others. The correlations they found were small, and their conclusion was that honesty is specific to situations rather than general to persons.

Walter Mischel gathered the accumulated evidence in Personality and Assessment in 1968 and produced the number that has organised the field since. Correlations between a personality measure and behaviour in a single situation rarely exceed about 0.30, a figure often called the personality coefficient.

Work out what that means, because the bare number is often quoted without its interpretation. A correlation of $r=0.30$ accounts for $r^2=0.09$ of the variance in behaviour, so about 9 percent. Ninety-one percent of the variation in whether someone helps, cheats or obeys is left unexplained by the trait measure. Against that, well-designed situational manipulations routinely move behaviour by twenty or thirty percentage points. On this evidence, if you want to predict what someone will do, you should ask about the circumstances rather than about the person.

The experiments

Three classic studies gave the argument its force, and their numbers are worth having exactly.

Stanley Milgram, in 1963, told subjects they were administering electric shocks to a learner in a memory experiment, with a switchboard rising to 450 volts labelled "danger: severe shock" and then "XXX". The learner, an actor, protested, screamed, complained of a heart condition and eventually fell silent. An experimenter in a lab coat delivered a fixed sequence of prods. In the baseline condition, 26 of 40 subjects, 65 percent, continued to 450 volts. Psychiatrists surveyed beforehand had predicted that about one in a thousand would.

John Darley and Daniel Batson, in 1973, sent Princeton seminary students across campus to give a talk. Half were assigned the parable of the Good Samaritan as their topic. On the way, each passed a man slumped in a doorway, coughing and groaning. What predicted helping was not the topic, which had no significant effect, but how much of a hurry they had been put in: 63 percent of the low-hurry group stopped, 45 percent of the intermediate group, and 10 percent of the high-hurry group. Some students in the high-hurry condition literally stepped over the victim on their way to give a talk about stopping to help a stranger.

Bibb Latané and John Darley, in 1968, seated subjects in a room that began filling with smoke. Alone, 75 percent reported it. With two passive confederates present, 10 percent did. Nothing about the individuals changed; the presence of two people doing nothing suppressed the response almost completely.

The pattern across all three is the same. Trivial and morally irrelevant features of the situation, a man in a coat, three minutes of schedule, two strangers not reacting, move behaviour by amounts no measured trait comes close to matching.

Example. Compute what proportion of behavioural variance a personality coefficient of 0.30 explains, and say why this does not by itself show that traits are unimportant.

The proportion is $r^2={0.30}^2=0.09$, or 9 percent. Two cautions apply before drawing the conclusion. First, 9 percent of variance in a single act is not nothing: an effect of that size, applied across a lifetime of choices, produces very different lives, and the same magnitude in medicine would be a treatment worth having. Second, the comparison with situational effects is not like for like. Situational manipulations are measured as differences between group averages, and trait effects as correlations at the individual level, and a large shift in a group average is compatible with the ordering of individuals within the group being entirely stable. Milgram's own numbers show this: the situation moved compliance far above what anyone predicted, and 14 of 40 subjects, 35 percent, still refused. Something distinguished them, and the experiment was not designed to find out what.

Now you. Darley and Batson found that the assigned topic had no effect on helping. What does this show, and what does it not?

The philosophical argument

Gilbert Harman drew the strong conclusion in 1999. Ordinary attributions of character traits are systematically mistaken, in the way that attributions of witchcraft were: we explain behaviour by invented dispositions because we commit the fundamental attribution error, over-weighting the person and under-weighting the situation. If there are no character traits, virtue ethics has no subject matter and should be abandoned.

John Doris, in Lack of Character in 2002, argued for something more careful and more damaging. The evidence tells against globalist traits, the broad cross-situational ones the tradition needs. It is compatible with local traits, narrow and highly situation-indexed dispositions: not courage but something like sailing-in-rough-weather-with-friends courage, which may be perfectly real and perfectly stable while telling you nothing about how the same person behaves in a burning building. Local traits exist, they are numerous and fragmented, and they do not add up to the unified character that eudaimonist ethics is built on.

Doris also draws a practical moral that is worth more than the theoretical one. If situations are this powerful, the reliable route to good behaviour is not to build character and trust it, but to attend to circumstances: avoid situations you know to be corrupting, arrange your commitments so that you are not hurried past the person who needs help, and design institutions on the assumption that ordinary people will do what the setting invites. Anyone who thinks their integrity will hold under a Milgram-strength situation has misread the evidence about people in general and has no special evidence about themselves.

The replies

Three replies are strong enough to keep the debate open, and one of them is arithmetical.

The aggregation reply, made by Seymour Epstein in 1979, points out that a correlation against a single act is the wrong measure. Any single behaviour is noisy, and the standard psychometric correction applies: aggregating over $k$ occasions raises the observed correlation to $kr/(1+(k-1)r)$. Starting from $r=0.30$, aggregating over 5 occasions gives 0.68, and over 10 occasions gives 0.81. The ceiling is largely an artefact of judging traits by one-shot criteria, and the same correction is applied without controversy to test items and to medical measurements. Since virtue is a claim about how someone acts across a life rather than on one afternoon, the aggregated figure is the relevant one.

The rarity reply, pressed by Rachana Kamtekar in 2004, notes that Aristotle never claimed most people have virtue. He claims the opposite repeatedly: full virtue is rare, requires long habituation and good fortune in one's upbringing, and most people are at best continent. Experiments on unselected undergraduates therefore test a population the theory predicts will mostly fail, and finding that they mostly fail is not a refutation. The theory would be refuted by evidence that nobody can achieve stable good conduct, and there is no such evidence.

Example. Distinguish what Harman and Doris each conclude, and say what evidence would tell against each.

Harman concludes that character traits do not exist and that attributing them is a systematic error like attributing witchcraft. Doris concludes that broad cross-situational traits are unsupported while narrow situation-indexed ones are real and numerous. Evidence against Harman is easy to specify and already exists: any demonstration that a measured disposition predicts aggregated conduct refutes him, and the aggregation results do exactly that, which is why almost nobody holds his position now. Evidence against Doris is harder, and this is what makes his version the serious one. It would have to show consistency across genuinely dissimilar situations, not merely repeated occasions of the same kind, since repeated occasions are what aggregation aggregates. Longitudinal work showing that a trait measured in one domain predicts conduct in an unrelated one, above the level Mischel's ceiling allows, would do it. Noticing that the two positions differ this much in what would refute them is the point of the exercise: they are usually cited together, and only one of them is still standing.

Now you. Someone says the situationist results show that people are not really responsible for what they do, since the situation caused it. Assess this.

The profile reply comes from Mischel himself, who did not conclude that persons do not matter. With Yuichi Shoda in 1995 he proposed that consistency lies in if-then signatures: a person is reliably more aggressive than others when criticised by a peer and reliably less aggressive when criticised by an adult, and that pattern is stable over years even though their average aggression predicts little. On this account character is real and is a structure rather than a quantity, which is closer to Aristotle's talk of acting rightly towards the right people at the right time than the crude trait model either side was arguing about.

The evidence has its own problems

An honest account has to add that some of the material on both sides of this argument has not held up, and the recent history is a caution against citing any of it too confidently.

The Stanford Prison Experiment of 1971, for decades the most cited demonstration of situational power, has collapsed. Thibault Le Texier's examination of Philip Zimbardo's own archives, published in American Psychologist in 2019, showed that the guards were briefed on the behaviour expected of them rather than inventing it, that instructions to be tough were given and repeated, that the most notorious guard later said he was consciously playing a role, and that the study had a conclusion before it had data. It should no longer be cited as evidence of anything.

Milgram's work survives in outline and is more complicated than its summary. Gina Perry's examination of the Yale archives showed that experimenters departed from the scripted prods, that some subjects doubted the shocks were real, and that the frequently quoted 65 percent comes from one of more than twenty conditions whose results ranged from near-total compliance to near-total refusal. The variation is itself informative: obedience fell sharply when the experimenter gave orders by telephone, when the learner was in the same room, and when other subjects refused first. Jerry Burger's partial replication in 2009, stopped at 150 volts for ethical reasons, found rates close to Milgram's, so the basic phenomenon is real.

Isen and Levin's 1972 finding, that 14 of 16 people who found a dime in a phone booth helped a stranger against 1 of 24 who did not, is the single most quoted situationist result and has a poor replication record. It should be treated as an illustration rather than as evidence.

The general lesson is one this course applies elsewhere: when a philosophical position leans on an empirical literature, it inherits that literature's reliability, and social psychology's reliability has been revised downwards over the last fifteen years.

Example. Starting from a personality coefficient of 0.30 for a single occasion, compute the correlation expected after aggregating over ten occasions, and say what the result licenses.

Apply the aggregation formula with $k=10$ and $r=0.30$: the numerator is $10\times 0.30=3.0$, the denominator is $1+9\times 0.30=3.7$, and the ratio is $3.0/3.7=0.81$. That is a strong relationship. What it licenses is the claim that traits predict aggregated conduct well, which is the claim virtue ethics actually makes, since nobody has ever said that a courageous person acts courageously on every single occasion. What it does not license is any confidence about a particular occasion, and this is the practical residue of the situationist argument that survives every reply: you cannot rely on anyone's character, including your own, to withstand a strong situation on a given day. Both conclusions are true together, and most of the heat in the debate came from treating them as competitors.

Now you. A company wants to reduce fraud among its staff. What does the situationist evidence recommend, and what does the virtue tradition add?

Where this leaves the theory

Virtue ethics has been made more modest by this literature and has not been dislodged.

What it has to give up is the strong globalist picture in which knowing that someone is honest tells you what they will do anywhere, and any suggestion that a well-formed character is armour against circumstances. What it keeps is the claim that dispositions are real and predict aggregated conduct, that they are formed by habituation, that virtue is rare and hard, and that moral perception is central. Each of those is either supported by the evidence or untouched by it, and the last one is arguably strengthened, since the mechanism by which situations work is that they change what people notice.

The practical upshot is a hybrid that most careful writers in the area now hold: cultivate character, and do not trust it. Build the dispositions, and also build the institutions and habits that make the dispositions less load-bearing.

All three normative families are now on the table, with their strengths and their unpaid debts. Every one of them, though, has been quietly assuming an answer to a prior question that none of them has argued for: whose interests belong in the calculation at all. That question is next, and the answer turns out to be worth billions of lives.

Every theory so far tells you to weigh interests, respect persons, or act as a good person would, and every one of them assumes a settled answer to the question of whose interests, which persons, and towards whom.

That prior question is moral status: what it takes for a being to count in its own right, rather than mattering only through its effects on those who do. It is the most consequential question in the subject by a wide margin, because a mistake about it is not a mistake about one act but about who is in the moral world at all, and history's largest moral failures have all been failures here rather than failures of calculation.

A distinction clears the ground. A moral agent can be held responsible: they deliberate, and can be praised and blamed. A moral patient is a being to whom things can be owed. The two come apart, and everyone accepts that they do, because infants and severely demented adults are patients without being agents. What is disputed is where the class of patients ends.

Candidate criteria

Four criteria have serious defenders, and each draws a different line.

Species membership. Human beings count, other animals do not. This is the working assumption of most legal systems and most people, and it is rarely defended explicitly, since stated baldly it looks like an appeal to group loyalty of a kind we reject elsewhere.

Rationality or personhood. What matters is being a self-conscious being with plans, a sense of oneself over time, and the capacity to give and follow reasons. This is Kant's position, and he draws the conclusion without flinching: beings whose existence depends on nature and not on reason have only relative worth as means, and duties towards animals are indirect, mattering because cruelty to them coarsens us towards people.

Sentience. What matters is the capacity to suffer and to enjoy, because that is what makes it possible for anything to go well or badly for a being at all. Bentham put it in a footnote in 1789 that is still the standard citation: the question is not whether they can reason, nor whether they can talk, but whether they can suffer.

Having a welfare. Slightly wider: what matters is having a life that can go better or worse from the inside, which may extend to beings with interests but limited affect.

Each criterion has to answer a test the others pass or fail differently, and the sharpest test is an argument that has resisted answering for fifty years.

The argument from marginal cases

Take any capacity proposed as the criterion of moral status, other than mere species membership: rationality, moral agency, language, autonomy, a sense of self over time. Some human beings lack it. Newborn infants lack all of them. Adults with advanced dementia or profound cognitive impairment lack most. And some non-human animals possess the capacity to a greater degree than those humans do.

The argument then runs: if the capacity is what confers status, those humans lack status and may be treated as we treat animals with similar capacities. If they nonetheless have full status, the capacity is not what confers it, and something else is, and whatever that is may well be present in animals too. Either way the standard position, that all humans have full status and no animals have any, is not supported by any capacity-based criterion.

Peter Singer built Animal Liberation in 1975 on this, coining speciesism for the position that species membership by itself confers status, and arguing that it has the same structure as racism: a morally irrelevant group membership doing the work that capacities should do.

Three replies are worth taking seriously. Carl Cohen argued in 1986 that the capacity for moral judgement characterises humans as a kind, and is not a test administered to individuals one at a time, so an impaired human has the status of a being of a rational kind. This is coherent and has a cost: it makes status depend on the statistical properties of a class rather than on anything true of the individual, which is what we normally call prejudice, and it delivers strange verdicts about a hypothetical rational alien from a mostly non-rational species. Others argue from potentiality, that infants will become rational, which does not help with the permanently impaired and raises the question of why a potential property confers an actual status. Others argue from relations, that our obligations arise from the relationships we stand in, and that impaired humans are members of our families and communities in a way that pigs are not. That last is the strongest, and its cost is that it makes status depend on social connection, which historically is exactly how outsiders came to be excluded.

Nobody has produced a criterion that includes every human being and excludes every animal without relying on species membership itself. That is the state of the argument, and it should be reported as such rather than resolved by a preference.

Example. Someone says the difference is that humans belong to a moral community: we can make agreements with each other and animals cannot. Test the criterion.

The criterion is reciprocity, and it is refuted from inside its own logic. Infants cannot make agreements, nor can the severely demented, nor could a person in a persistent vegetative state, yet none of them may be experimented on or eaten. If the reply is that they belong to a species most of whose members can reciprocate, the criterion has quietly become species membership again, and the argument is circular. If the reply is that we extend the protection of the community to those connected to its members, the criterion has become relational, and it must then explain why a stranger with no connection to anyone is protected and a family dog is not. The exercise is not meant to establish that pigs have the status of persons. It is meant to show that the popular criterion collapses under one test, and that a defensible version of the standard view has to be much more carefully built than most people who hold it realise.

Now you. Singer's principle is equal consideration of interests. Does it imply that a dog and a child should be treated the same?

The scale, and why it changes the arithmetic

Whatever weight one assigns to an animal's interests, the number of animals makes the question large, and the numbers are not in dispute.

The Food and Agriculture Organization's figures put land animals slaughtered for food at roughly 80 billion a year, of which about 90 percent are chickens. Against a human population of about 8 billion, that is around ten land animals killed per person per year, worldwide. Estimates for fish and other aquatic animals run to a trillion or more and are far less certain, since fish are usually recorded by weight rather than counted.

The conditions matter as much as the count, and the standard case concerns broiler chickens because that is where most of the animals are. Selective breeding has reduced the time to slaughter weight from around sixteen weeks in the 1920s to about six weeks, roughly 42 days, while roughly quadrupling the bird's mass. The resulting animal grows faster than its skeleton and cardiovascular system comfortably support, and lameness and heart failure in fast-growing broiler flocks are documented in the veterinary literature rather than being campaigners' claims.

The arithmetic point is structural and holds whatever number you insert. Suppose, conservatively, that you weigh a chicken's suffering at one hundredth of a comparable human's. Then 80 billion animals a year is equivalent, on your own weighting, to 800 million humans a year, which is a tenth of the human population, subjected to conditions you have already conceded are bad. Scale converts a modest per-animal weight into one of the largest items in any global moral accounting, and this is why the topic cannot be dismissed as a matter of taste even by someone who thinks animals matter much less than people. To reach the conclusion that it is unimportant you have to hold that animals matter not slightly less than humans but essentially not at all, which is exactly the position the marginal cases argument attacks.

Is status all or nothing?

The argument so far has treated status as a switch, and there is a serious case that it is a dial.

Mary Anne Warren argued in 1997 for a multi-criterial account, on which several different considerations generate moral standing of different strengths: sentience gives a being a claim not to be made to suffer, personhood adds a claim not to be killed, membership of a social community generates further claims, and so on. David DeGrazia defends a similar graded picture. On such views a mouse counts, a pig counts more, a chimpanzee more again, and a human most, without any of them counting for nothing.

The attraction is that it matches how almost everyone actually reasons, and that it dissolves the marginal cases argument's sting: an impaired human retains the claims that come from sentience and from social membership even without the ones that come from full rational agency. The cost is that a dial is much easier to abuse than a switch. A graded account gives no principled resistance to grading humans against each other, which is precisely the move a switch was invented to block, and defenders have to add that full status is reached by every human being and cannot be lost, which is exactly the unexplained step they set out to avoid.

Law has in practice adopted the graded view. The European Union's Lisbon Treaty of 2009 requires member states to pay full regard to the welfare requirements of animals as sentient beings, and the United Kingdom's Animal Welfare (Sentience) Act of 2022 recognises vertebrates and, following a commissioned review, cephalopods and decapod crustaceans. None of that gives an animal the status of a person, and all of it gives them a standing that property does not have.

Example. A laboratory proposes an experiment causing moderate pain to 200 mice, expected to yield a treatment that will relieve severe pain in some thousands of people. Assess it on a switch account and on a dial account.

On a switch account with the line at personhood, the mice have no standing at all and no justification is required beyond avoiding gratuitous waste, which is a verdict most people, including most researchers, reject in practice. On a switch account with the line at sentience, the mice count in the same currency as the humans, and the calculation is a straight comparison of pain against pain with the numbers heavily favouring the experiment, provided the expected benefit is real rather than hoped for. On a dial account, the mice count with a lower weight, so the experiment is justified but the weight is not zero, and what follows is a set of requirements rather than a permission: use the smallest number that will answer the question, use anaesthesia and analgesia wherever they do not destroy the result, and do not run the study at all unless the question is worth an answer. That set is the "three Rs" framework, replacement, reduction and refinement, formulated by William Russell and Rex Burch in 1959 and now written into research regulation across Europe. The dial account is the one the law has adopted, and this case shows why.

Now you. Does a graded account of status let you avoid the argument from marginal cases?

People who do not exist yet

The other expansion of the circle runs forward in time, and it has a peculiar structure: future people cannot reciprocate, cannot vote, cannot bargain, and their existence and number depend on what we do.

Almost everyone accepts they count for something. Burying toxic waste that will leak in three hundred years is wrong, and it is wrong because of what it will do to whoever is there, not because of anything about us. The disputes are about how much, and they turn on a technical device with enormous consequences.

Discounting applies a factor to future costs and benefits, reducing them by a fixed proportion per year. Part of the standard rate is uncontroversial: money invested grows, and future people will probably be richer, so a pound spent on them buys less welfare. The contested part is the pure rate of time preference, which discounts future welfare simply because it is future. Frank Ramsey called that ethically indefensible in 1928, arising merely from a weakness of the imagination, and most philosophers agree with him.

The numbers show why this is not a technicality. The Stern Review of 2006 used a pure rate of time preference of 0.1 percent per year, chosen to represent only the probability of humanity ceasing to exist. Over a century that leaves a weight of $1/{1.001}^{100}=0.90$: a benefit to someone a hundred years from now counts for 90 percent of the same benefit today. William Nordhaus criticised this and used a rate closer to 1.5 percent, which over the same century gives $1/{1.015}^{100}=0.23$, a weight of 23 percent, and over two centuries about 5 percent. The two analyses differ by a factor of four on how much a distant future person counts, and that single parameter, rather than any disagreement about climate science, accounts for most of the gap between their policy recommendations. A choice of discount rate is a moral choice wearing the clothes of a technical assumption, and the same is true in cost-benefit analyses of pensions, infrastructure and nuclear waste.

Example. Compute the weight given to a benefit 100 years from now under a pure time preference of 0.1 percent and of 1.5 percent, and say what the difference means for policy.

At 0.1 percent, the factor is $1/{1.001}^{100}=0.90$. At 1.5 percent, it is $1/{1.015}^{100}=0.23$. The ratio is about four, so the same future harm is worth roughly four times as much under Stern's assumption as under Nordhaus's, and the effect compounds: at 200 years the second rate leaves about 5 percent. In practice this decides whether an expensive mitigation programme now is worth it, because the costs fall in the present and the benefits fall decades to centuries out. Notice what has and has not been shown. The calculation does not establish that Stern is right; a defender of the higher rate can argue that it captures genuine uncertainty about whether the benefits will materialise, or that some partiality towards the near is legitimate. It establishes that the disagreement is ethical, that it is located in one parameter, and that anyone who presents a cost-benefit analysis without stating that parameter has hidden their moral premise inside their arithmetic.

Now you. A government adopts a policy of resource depletion that raises living standards now and leaves people in 200 years worse off than a conservation policy would have. But different people will exist under the two policies, since a change of that size alters who meets whom and who is born. Have the future people been harmed?

Where to draw the line

No criterion draws a line that everyone will accept, and the honest thing is to say what each costs.

Draw it at species and you have a rule that matches practice, protects every human unconditionally, and cannot be defended by any property of the individuals it sorts. Draw it at rationality and you have a principled criterion that excludes infants and the profoundly impaired. Draw it at sentience and you have a criterion with a clear rationale and a boundary problem of its own, since it requires deciding about insects, about fish, and eventually about systems that report distress without anyone knowing whether there is anything it is like to be them. Draw it by relationships and you have an account of why family and community generate obligations, which is true, and no protection for anyone outside them.

Two things can be said with confidence. The direction of travel has been one way: every extension of the circle, to other tribes, to other races, to the enslaved, to women, to children, to animals in the small degree so far achieved, has been resisted with arguments of the same shape, that the excluded lack some capacity or fall outside some relationship, and each extension has afterwards looked obvious. That is weak evidence, and it is evidence. And the cost of an error is asymmetric: excluding a being that counts is a catastrophe of a different kind from including one that does not.

The next lesson takes up what happens once the boundary is fixed. Knowing who counts does not tell you how to divide anything between them, and neither adding up welfare nor obeying a rule explains what one person can rightfully demand of another.

Knowing how much good an outcome contains tells you nothing about who should have it, and knowing which acts are forbidden tells you nothing about how a society should be arranged.

That gap is where a third source of moral requirement sits. Its idea is that morality is what could be justified to each person affected, and that a principle nobody could reasonably refuse has an authority that neither a sum nor a commandment has. Contract theories of this kind have dominated political philosophy since 1971, and they answer objections the first two families could not: the separateness of persons, the limits of aggregation, and the question of why anyone should accept a rule at all.

The original position

John Rawls's A Theory of Justice in 1971 asks a question with a built-in answer to the problem of bias. What principles would free and rational people choose to govern the basic structure of their society, if they had to choose without knowing who in it they would be?

The veil of ignorance specifies what is removed. The parties do not know their class or social position, their race or sex, their natural talents and intelligence, their conception of the good life, their psychological dispositions, or the generation into which they are born. They do know general facts: economics, psychology, the circumstances under which societies function.

The device is not a story about a historical contract, and Rawls is explicit that no such agreement ever occurred. It is a representation of fairness. Anything you would agree to without knowing your position is something you cannot have chosen because it favours you, so the constraint models impartiality by information rather than by an appeal to benevolence. The parties are stipulated to be mutually disinterested, each caring only about their own share, which is deliberate: if the argument works with self-interested parties, it does not depend on anyone being good.

The choice is over an index of primary goods: rights and liberties, opportunities, income and wealth, and the social bases of self-respect. These are things a rational person wants whatever else they want, which lets the parties choose without knowing their conception of the good.

The two principles

Rawls argues that the parties would choose two principles, in a strict order of priority.

The first: each person has an equal claim to a fully adequate scheme of equal basic liberties, compatible with the same scheme for all. The second, applying to social and economic inequalities: they must attach to positions open to all under fair equality of opportunity, and they must be to the greatest benefit of the least advantaged members of society.

That last clause is the difference principle, and it is the famous one. It does not require equality. It permits any inequality that makes the worst-off group better off than they would be under a more equal arrangement, so a surgeon may earn many times a cleaner's wage if the incentive produces surgeons whose existence benefits cleaners. What it forbids is inequality that merely benefits the already advantaged.

The priority ordering is strict, which readers often miss. Liberties may not be traded for economic gain at all, not even a large one, and fair equality of opportunity takes priority over the difference principle. A society cannot buy prosperity for its poorest by disenfranchising a minority, however good the arithmetic.

Rawls's argument that the parties would choose this is that they would reason by maximin: rank the alternatives by their worst outcomes and choose the one whose worst outcome is best. He does not claim maximin is generally rational. He claims it is rational under three specific conditions, all of which the original position satisfies: there is no basis for assigning probabilities to the outcomes; the guaranteed minimum under maximin is satisfactory, so there is little to gain from gambling for more; and the alternatives contain outcomes the chooser could not accept, which he calls the strains of commitment, since a person who agreed to a principle and then found themselves crushed by it would not be able to keep the agreement.

Example. Three social arrangements give these welfare levels to three equally sized groups: A gives (10, 10, 10), B gives (5, 20, 20), C gives (8, 12, 40). Which does maximin choose, and which does an expected utility maximiser behind the veil choose?

Maximin compares worst outcomes: 10 for A, 5 for B, 8 for C, so it chooses A. An expected utility maximiser who assigns equal probability of being in any group compares averages: A gives 10, B gives $(5+20+20)/3=15$, and C gives $(8+12+40)/3=20$, so it chooses C. This is exactly John Harsanyi's objection, made in the 1950s before Rawls's book and pressed against it afterwards. Harsanyi argued that a rational chooser ignorant of their identity should assign equal probability to being anyone and maximise expected utility, which yields average utilitarianism rather than the difference principle, and that maximin is irrationally cautious: it would forbid crossing a road, since the worst outcome is death. Rawls's reply is that the original position is a choice about the terms of one's whole life made once and irrevocably, with no probabilities available and no chance to recover from a bad draw, so ordinary gambling reasoning does not apply. Which of the two is right is the central technical dispute about the argument, and it is unresolved.

Now you. Does the difference principle require the surgeon's high salary to be abolished?

Nozick's answer

Robert Nozick's Anarchy, State, and Utopia, published three years later in 1974, is the most effective attack on the whole approach, and it works by changing the question.

Rawls asks which distribution is just. Nozick denies that justice is a property of distributions at all. On his entitlement theory, a distribution is just if it arose by just steps: justice in acquisition, covering how unowned things come to be owned; justice in transfer, covering voluntary exchange and gift; and rectification, covering what to do about past violations. Whatever pattern results from just steps is just, however unequal, and no pattern is required.

His argument is the Wilt Chamberlain case, and it is worth working through. Suppose your favourite distribution, whatever it is, obtains: call it $D_1$. Chamberlain, a basketball player people want to watch, signs a contract giving him 25 cents from each ticket, and a million people freely pay to see him play, each knowing where their quarter goes. He now has $1\{,\}000\{,\}000\times 0.25=250\{,\}000$ dollars, far more than anyone else, and the distribution is $D_2$. Everyone in $D_1$ was entitled to what they held, everyone transferred voluntarily, nobody was worse off than they chose to be. If $D_1$ was just and every step was legitimate, how can $D_2$ be unjust? And if it is unjust, then maintaining any pattern requires continuously forbidding people from doing what they wish with what is theirs.

Nozick draws the conclusion sharply: taxation of earnings from labour is on a par with forced labour, since it takes a portion of what a person's hours produced without their consent.

The replies are strong and none is decisive. The step Nozick spends least effort on is acquisition, and it is where the theory is most vulnerable: nearly all existing holdings trace back through conquest, enclosure and slavery rather than through legitimate first acquisition, so his own principle of rectification would license redistribution on a scale that dwarfs anything Rawls proposes. The Lockean proviso he accepts, that acquisition must leave enough and as good for others, is hard to satisfy for land and impossible for a finite planet. And the case understates the background: the spectators' quarters buy entertainment in a society whose stadiums, contract law and policing they did not individually consent to either, so the picture of a distribution disturbed only by free choices is idealised.

Example. Which premise of the Chamberlain argument does a Rawlsian deny?

Not that the transfers were voluntary, and not that $D_1$ was just. What a Rawlsian denies is the framing: the difference principle applies to the basic structure of a society, its constitution, property system, tax rules and institutions, and not to individual transactions within it. Rawls is explicit about this. So the just arrangement is one in which the rules of property and taxation are set so that inequalities work to the benefit of the worst off, and within those rules people spend their money as they wish, including on basketball. Chamberlain's earnings are then taxed under a scheme chosen behind the veil, and nobody is being interfered with in any objectionable sense, since the tax was part of the entitlement all along rather than a subtraction from it. Whether that reply works depends on whether the distinction between the structure and the transactions can be sustained, which Nozick denies and which is the real point of disagreement. It is a much more interesting dispute than the one about whether people may pay to watch basketball.

Now you. Nozick says taxation of earnings from labour is on a par with forced labour. Test the analogy.

Scanlon and reasonable rejection

T. M. Scanlon's What We Owe to Each Other in 1998 takes the contract idea in a different direction, aiming at interpersonal morality rather than at the structure of society.

His formula: an act is wrong if it would be disallowed by any principle for the general regulation of behaviour that no one could reasonably reject as a basis for informed, unforced general agreement.

Two features carry the weight. The motivation is not self-interest but the desire to justify oneself to others on terms they could not reasonably refuse, which Scanlon takes to be a basic and widely shared concern. And rejection is individualist: a principle can only be rejected by a person, on the ground of the burden it places on them, and complaints may not be summed across people. Many small burdens spread over a large number of people do not add up to one big complaint, because there is nobody who bears the sum.

The individualist restriction is what makes the theory a genuine rival to consequentialism, and Scanlon's illustration is the transmitter room. A technician, Jones, is trapped under fallen equipment where he is receiving painful electric shocks. We can rescue him, but only by interrupting a transmission of a football match being watched by millions of people, who will be mildly disappointed. Aggregating says the millions win, since enough small disappointments outweigh fifteen minutes of agony. Scanlon says they cannot: each viewer's complaint is trivial, no viewer bears a burden anywhere near Jones's, and no number of trivial complaints becomes a serious one.

The theory then faces the opposite problem, which is that numbers sometimes obviously do count. If you can save one person on one island or five on another, and all six face the same death, everyone thinks you should save the five, yet each of the five has exactly the complaint the one has, and no complaint is greater. John Taurek argued in 1977 that this shows the numbers should not count, and that you should flip a coin. Scanlon's reply is a tie-breaker argument: a principle that ignored the additional people would be reasonably rejectable by them, because it would treat their presence as making no difference at all, which is a failure to count them rather than a refusal to add them up. Whether this genuinely avoids aggregation is disputed, and Parfit argued that it does not.

Example. Assess the transmitter room case on each of the three families of theory.

A consequentialist compares aggregate welfare: fifteen minutes of severe pain against millions of small disappointments, and if the audience is large enough the arithmetic favours leaving Jones, which almost everyone finds monstrous. A Kantian says Jones is being used as a means to the audience's entertainment, since his suffering is being permitted for their benefit on a plan he could not consent to, and rescues him. A contractualist says the same and explains why more precisely: the principle "interrupt transmissions to prevent serious injury" cannot be reasonably rejected by any individual viewer, since no viewer's burden is remotely comparable, while the principle that leaves Jones there can be reasonably rejected by him. Notice that the second and third agree on the verdict for different reasons, and that the contractualist reason generalises better: it explains not only why Jones should be rescued but why the size of the audience is irrelevant, which is the feature the case was designed to isolate.

Now you. Is it reasonable to reject a principle permitting a national speed limit of 130 kilometres per hour, given that a lower limit would save lives?

What contract views cannot reach

Every theory in this family shares one structural limitation, and it falls on the same people each time.

A contract is an agreement among parties who can make agreements. Rawls's parties are described as roughly equal in power and capable of contributing to a scheme of cooperation, which is what makes cooperation mutually advantageous and gives everyone a reason to be in the agreement. That construction has no place for anyone who cannot contribute or reciprocate: people with severe cognitive disabilities, non-human animals, and future generations, who cannot bargain with us because they do not exist yet and can do nothing for us in return.

Rawls saw this and set the problem aside, treating the severely disabled as a case for later and handling generations by a separate device, a just savings principle, which the parties adopt on the assumption that they do not know which generation they belong to. Nussbaum, in Frontiers of Justice in 2006, argues that the omissions are not incidental but structural, and that a theory grounded in mutual advantage between roughly equal parties can never be repaired to include those who bring nothing to the bargain. Her alternative, the capabilities approach, grounds entitlements in what a creature needs to live a life worthy of its dignity, which requires no reciprocity at all.

Scanlon's version has the same shape and a narrower version of the problem, since reasonable rejection requires a being capable of being given reasons. He addresses it by extending trusteeship: those who cannot reject principles for themselves are represented by those who can, which handles the severely disabled and infants at the cost of making their standing derivative.

This is worth holding beside the previous lesson. Contractualism gets the limits of aggregation right, which consequentialism cannot, and gets the boundary of the moral community wrong in precisely the place where sentience-based views get it right. No single theory has both.

Comparing the three on one case

Suppose a state is deciding whether to fund an expensive treatment that would extend the lives of a small number of people with a rare disease, at a cost that would otherwise fund routine care for many.

A consequentialist compares quality-adjusted life years bought per pound and funds whichever buys more, which will usually be the routine care, and treats the identity of the patients as irrelevant. Rawls's framework asks about the basic structure rather than the individual decision, and would generate a right to a fair share of health resources whose content is settled by what parties behind the veil, not knowing whether they would have the rare disease, would accept. Nozick's framework denies the state should be running the health service at all, and treats the question as one for the patients and whoever chooses to help them. Scanlon asks whose complaint is greater: the patient who dies without treatment has a serious one, and so does each patient denied routine care, and the answer depends on comparing those two burdens person by person rather than in aggregate, which is why contractualist reasoning tends to favour treating identified serious conditions more than a straight efficiency calculation does.

Each framework is doing something the others are not, and the disagreement is traceable to a single feature of each. That traceability is the point of the whole exercise, and the last part of this course puts it to work on the cases where people actually disagree, starting with the oldest and hardest of them.

Two of the cases that divide people most sharply turn on the same two questions, and once those are isolated the argument becomes tractable even where it stays unresolved.

The questions are whether killing differs morally from allowing to die, which the lesson on duties raised and did not settle, and who has moral status, which the lesson on who counts raised and did not settle either. Euthanasia and abortion are where those two abstractions meet institutions, laws and actual people, and the point of taking them together is to show that the disagreement is structured rather than a clash of raw sentiment.

Killing and letting die

James Rachels put the question at its sharpest in the New England Journal of Medicine in 1975, with a pair of cases designed to isolate one variable.

Smith stands to gain a large inheritance if his six-year-old cousin dies. He drowns the boy in his bath and arranges it to look like an accident. Jones stands to gain the same inheritance in the same way. He enters the bathroom intending to drown the boy, and as he does so the child slips, strikes his head and falls face down in the water. Jones stands ready to push him under if necessary, but it is not necessary. He watches, doing nothing, and the boy drowns.

Rachels asks whether Jones's conduct is less reprehensible than Smith's. Almost nobody says it is. Since the only difference between the two is that one killed and the other let die, the bare difference between killing and letting die cannot itself be morally significant. He draws a practical conclusion for medicine: if it is permissible to withdraw treatment from a patient and let them die slowly, it cannot be the mere fact of killing that forbids giving an injection that ends the same life quickly, and a doctrine that permits the slower death while forbidding the faster one is producing more suffering for no reason.

The argument is good and its limits are real. Two objections have force. The cases equalise everything except the mechanism, including motive, intention and consequence, which is what makes them so vivid; but a distinction can be morally significant in general without making a difference in a case where the agent is a murderer either way, just as whether you shot or poisoned someone makes no difference when you meant to kill them. The second objection is about what the distinction is usually doing. Its point in ordinary morality is not to excuse people who want a death; it is to explain why you are required not to kill any of the millions of strangers you could kill, and not required to save any of the millions you could save. That asymmetry is real, enormously consequential, and Rachels's pair does not touch it.

The fair conclusion is narrower than Rachels wanted and still substantial: the bare difference does not do the work by itself, and where intention, consent and outcome are all the same, an appeal to the distinction alone is not enough to separate two cases.

What the law does with it

Legal systems have, almost without exception, adopted a position the argument above puts under pressure: they permit letting die and forbid killing.

In Airedale NHS Trust v Bland in 1993, the House of Lords permitted the withdrawal of artificial nutrition and hydration from Tony Bland, who had been in a persistent vegetative state since the Hillsborough disaster four years earlier. The reasoning was that continued treatment was not in his best interests and that withdrawing it was an omission rather than an act, so no crime was committed. Several of the judges said in terms that the distinction they were relying on was of doubtful moral coherence, and that Parliament rather than the courts should address it. Nothing has since.

In the United States, Cruzan in 1990 recognised a competent person's constitutionally protected liberty interest in refusing life-sustaining treatment, and Vacco v Quill in 1997 upheld a state ban on assisted suicide against the argument that it was inconsistent with the right to refuse treatment. The Court held the distinction between the two was rational: a patient who refuses treatment dies of the underlying disease, while one who takes lethal medication is killed by it, and intent differs in the two cases.

So the legal architecture rests on doing versus allowing and on intention, exactly the two distinctions that have the most trouble in philosophy, and the judges know it.

Example. A patient with motor neurone disease is on a ventilator and asks for it to be switched off, knowing she will die within minutes. Her doctor does so. Compare this with the doctor administering a lethal injection at her request.

Legally the two are entirely different: the first is required, since continuing to treat a competent patient who refuses is a battery, and the second is murder in most jurisdictions. Morally the differences are hard to sustain under the pressure of the case. The consent is identical, the outcome is identical, the intention of the patient is identical, and the doctor who flips the switch knows exactly what will follow. The main candidate difference is causal: switching off the ventilator lets the disease kill her, while the injection kills her. That is a real difference and it is doing less work than it appears, because it was the doctor's earlier act that placed her on the ventilator, so switching it off is arguably the removal of a barrier the doctor supplied rather than a pure omission. What remains defensible is not that the two acts differ in gravity but that a legal rule permitting only the first is easier to police, which is a reason of a completely different kind and should be argued as such.

Now you. Does Rachels's argument show that active euthanasia should be legal?

What the euthanasia data shows

Because several jurisdictions have now permitted assisted dying for decades, some of the argument is empirical, and the figures deserve to be quoted accurately.

The Netherlands legalised euthanasia and physician-assisted suicide in 2002, requiring unbearable suffering with no prospect of improvement, a voluntary and well-considered request, a second physician's opinion, and reporting to a review committee. The regional review committees reported 8,720 cases in 2022, about 5 percent of all deaths in the country. Belgium legalised in the same year. Canada introduced medical assistance in dying in 2016, and Health Canada reported 13,241 provisions in 2022, about 4 percent of all deaths.

Oregon's Death with Dignity Act, passed in 1997 and in force from 1998, is a narrower model: physicians may prescribe but not administer, the patient must be terminally ill with a prognosis of six months or less, and the numbers run in the hundreds per year, well under one percent of deaths in the state.

The slippery slope argument says that a permission granted for one class of case will spread. The honest reading of the record is that this has partly happened, and that whether it counts as a slope depends on a moral judgement rather than on the data. The Dutch and Belgian laws were written in terms of unbearable suffering rather than terminal illness, so their extension to psychiatric conditions and advanced dementia was a consequence of the original criterion rather than a departure from it, and those cases remain a small fraction, in the low hundreds a year in the Netherlands. Canada legislated in 2016 for those whose death was reasonably foreseeable, removed that requirement in 2021 after a court ruling, and has repeatedly postponed extension to cases where mental illness is the sole condition. Oregon's criteria have barely changed in a quarter of a century.

What that pattern suggests is that the slope is a function of how the criterion is drafted rather than an inevitable dynamic. A law written around terminal illness has stayed there; a law written around suffering has followed suffering wherever it goes. That is a conclusion an opponent and a supporter can both use, and it is more useful than either side's slogan.

Example. State the slippery slope argument in premises and say which line the data bears on.

(1) Permitting assisted dying for competent, terminally ill patients who request it is acceptable in itself. (2) Once permitted for that class, it will in practice be extended to classes that are not acceptable, such as people who are not dying, people who cannot consent, or people who feel themselves a burden. (3) The extension cannot be prevented by drafting or enforcement. (4) Therefore it should not be permitted at all. The data bears on (2) and (3) and on nothing else, which is worth knowing, because most public argument attacks or defends (1) and then treats the whole case as settled. What the record shows is that (2) has been true where the statutory criterion was suffering and largely false where it was terminal illness, and that (3) is therefore too strong as stated: extension has tracked drafting, so the argument's force depends entirely on which law is proposed. Note that this is a logical slippery slope claim converted into an empirical one, and that only the empirical version can be tested. Someone who means that the extension follows logically from the principle owes an argument rather than a statistic.

Now you. What would count as evidence that vulnerable people are being pressured into requesting assisted death?

Thomson's violinist

Judith Jarvis Thomson's "A Defense of Abortion", published in 1971, is the most quoted paper in applied ethics, and its method is what makes it powerful: it grants the opposing premise.

Suppose, she says, that a fetus is a person with a full right to life from conception. Does the conclusion follow? She argues not, with a case. You wake in a hospital bed connected by tubes to a famous unconscious violinist. The Society of Music Lovers has kidnapped you because you alone have the right blood type; his circulatory system is plugged into yours; unplug him and he dies; stay connected for nine months and he will recover fully. The director apologises but says you may not be unplugged, since the violinist is a person with a right to life.

Almost everyone judges that you would be generous to stay and are not required to. If so, a right to life does not include a right to the use of another person's body to sustain it, and the standard argument from personhood to the impermissibility of abortion is invalid even if its premise is true. Thomson adds a distinction that has outlasted the case: we are required to be minimally decent samaritans, not good samaritans, and the law reflects this by rarely requiring anyone to undergo serious cost to save a stranger. Nobody may be compelled to donate a kidney to save a life, even a child's, which is a striking asymmetry with what is demanded of a pregnant woman.

The objections are focused and serious. The violinist case involves no responsibility for the dependency, whereas most pregnancies follow a voluntary act with a known risk, and many people think you may not create a dependency and then withdraw. Thomson's reply is the people-seeds case, in which seeds drift through windows and may take root despite screens, so that even a person who takes precautions can find themselves responsible in a sense that does not obviously generate a duty. Critics also press the distinction between unplugging, which lets the violinist die of his ailment, and abortion procedures that kill directly, which puts the earlier part of this lesson to work on the other side. The case is therefore decisive against one argument, the simple one from personhood, and not against the position as a whole.

Marquis and the future like ours

Don Marquis published the strongest secular argument on the other side in 1989, and its strategy is also to avoid the disputed ground.

He asks first why killing an adult is wrong, and rejects the usual answers. It is not the effect on others, or killing a hermit would be permissible. It is not the brutalising effect on the killer, which is a derivative wrong. What makes killing you wrong is what it does to you: it deprives you of your future, of all the experiences, activities, projects and enjoyments that would otherwise have made up your life. Call that a future like ours.

Then the argument runs in one step. A standard fetus has a future like ours in exactly the same sense: a set of experiences and projects that it will have if it is not killed. So abortion is wrong for the very same reason killing an adult is wrong, and no claim about personhood, souls or species is needed.

The argument's power is that it explains rather than asserts, which is the standard this course has applied throughout, and that it grounds the wrongness of killing in something that also explains why killing children is wrong and why euthanasia for someone with no valuable future left may not be.

The objections are correspondingly precise. Contraception seems to deprive a possible person of a future too, and Marquis replies that there is no identifiable subject before conception to be deprived, since neither the sperm nor the egg is the individual who would exist. Critics reply that the same problem arises earlier than he wants, since a zygote can twin for about fourteen days, so there is no determinate individual to have a future during that window. Others attack the account of the wrongness of killing, arguing that what matters is the thwarting of a subject's own desires and plans, which a fetus has none of, so that a being with a valuable future but no interest in it is not wronged in the same way.

Example. Which of Thomson's and Marquis's arguments does each of these claims engage? (a) "A fetus is not a person." (b) "You may not create a dependency and then end it." (c) "Killing is wrong because it thwarts a being's own plans."

(a) engages neither, which is the point of both papers: Thomson grants personhood and argues the conclusion still fails, and Marquis never uses personhood, so someone who spends the argument on the status of the fetus is missing both of the strongest positions in the literature. (b) engages Thomson, at exactly the place she is weakest, by attacking the analogy between an unchosen kidnapping and a foreseeable consequence of a voluntary act. (c) engages Marquis, by offering a rival account of why killing is wrong that a fetus does not satisfy, and it is the standard reply; its cost is that it has trouble explaining why killing a person in a dreamless sleep, or a temporarily comatose patient with no current plans, is wrong.

Now you. Which of the two arguments does the case of a pregnancy resulting from rape bear on, and what does that tell you about the structure of the debate?

Where the frameworks land

Setting the three normative families against these cases shows that they do not divide the way people expect.

On euthanasia, a consequentialist assesses suffering prevented against risks of error and abuse, and generally supports a regulated permission. A Kantian is genuinely split: autonomy is the foundational value, which supports a right to decide the manner of one's death, and Kant himself argued that suicide treats one's own person as a means to the relief of suffering, so both wings of the tradition are represented in the current literature. A virtue ethicist asks what a compassionate and just doctor does, and how a society that kills its dying differs from one that does not, which reframes the question towards the quality of care rather than towards permissions. A contractualist weighs the complaint of a person left to die badly against that of a vulnerable person who might be pressured.

On abortion, the split runs through moral status rather than through the theories. A consequentialist who counts sentience finds little welfare at stake early in pregnancy and a great deal at stake for the woman. A Kantian who counts rational agency reaches a similar verdict on the fetus and a stronger one about the woman's autonomy, while a Kantian who counts potential rational agency does not. Marquis's argument, notably, is available to any of them. That is why abortion is not a disagreement between consequentialists and deontologists, and why arguments conducted as though it were make no progress.

What can be settled

Three things can. The bare distinction between killing and letting die does not carry the weight the law places on it, which is a philosophical result. The slippery slope in assisted dying is a matter of how the criterion is drafted rather than an inevitability, which is an empirical result. And the standard argument from fetal personhood to the impermissibility of abortion is invalid, which is Thomson's result, whatever one thinks of her conclusion. What cannot be settled from here is the status of the fetus, and no amount of further argument about anything else will substitute for it.

The remaining case in this course is different in kind. It divides people much less loudly, almost nobody defends the position most people take, and the cost of getting it wrong is measured in millions of lives a year.

Almost nobody defends the position that almost everybody holds, which is that a person may spend freely on themselves while children die of preventable causes elsewhere.

That is an unusual state for a moral question. The abortion and euthanasia arguments of the previous lesson have serious defenders on both sides. Here the argument for a large obligation is short, its premises are individually hard to deny, and its conclusion is rejected in practice by nearly everyone including most of those who accept it in print. Working out where it goes wrong, if it does, is a good final test of everything in this course.

The argument

Peter Singer published "Famine, Affluence, and Morality" in 1972, in response to the Bengal refugee crisis, and it runs in three lines.

First: suffering and death from lack of food, shelter and medical care are bad. Second: if it is in our power to prevent something bad from happening without thereby sacrificing anything of comparable moral importance, we ought morally to do it. Third: we can prevent such suffering by giving money to effective aid. Therefore we ought to give, up to the point where giving more would cost us something of comparable moral importance.

The famous illustration is the drowning child. You pass a shallow pond and see a small child face down in it. Wading in will save her and will ruin an expensive pair of shoes. Nobody thinks the shoes are a reason, and nobody thinks that saving her is admirable rather than required. Singer's claim is that the second premise is what everyone relies on when they judge the pond case, and that once stated it applies far beyond the pond.

He also offers a weaker version, replacing "of comparable moral importance" with "morally significant", which yields a much less demanding conclusion and is still enough to condemn most current behaviour. The strategy is deliberate: even a reader who rejects the strong principle is caught by the weak one.

What the argument costs

The obligation the argument generates depends on what money actually buys, and that figure is knowable.

GiveWell, which publishes its cost-effectiveness models rather than asserting a number, has for years estimated the cost per life saved through its top-rated malaria programmes at somewhere in the region of 3,000 to 5,500 US dollars, with the figure revised as evidence and funding gaps change. The underlying evidence for insecticide-treated nets is unusually good for aid: the Cochrane review of the large randomised trials found that nets reduce all-cause child mortality by roughly a sixth, on the order of five to six deaths averted per thousand children protected per year.

Work the personal arithmetic. Someone earning 50,000 a year who gives a tenth of it, for forty years, gives 200,000. At 5,000 per life, that is $200\{,\}000/5\{,\}000=40$ lives. At the lower end of the range it is closer to sixty. Giving one percent instead of ten gives four lives. These are not rhetorical figures; they are what the published models imply, with all the uncertainty those models carry.

The uncertainty deserves stating. Cost per life saved is a modelled quantity, not a measurement, and it depends on assumptions about baseline mortality, insecticide resistance, net usage, and what would have happened without the funding. The honest way to hold it is as an order of magnitude: thousands of dollars, not tens of dollars and not millions.

Example. Compute the lives implied by giving one percent and ten percent of a 50,000 income over forty years at 5,000 per life, and say what the calculation does and does not establish.

One percent gives $50\{,\}000\times 0.01\times 40=20\{,\}000$, which at 5,000 per life is 4 lives. Ten percent gives 200,000, which is 40 lives. The calculation establishes that the difference between two very ordinary levels of generosity is, on the best published estimates, an order of magnitude in lives, which is a fact worth confronting whatever theory you hold. It does not establish that you are obliged to give either amount, because that conclusion needs Singer's second premise, and the whole argument of the rest of this lesson is about whether that premise survives. Nor does it establish that the marginal cost stays constant: the cheapest opportunities are taken first, so a very large increase in funding would raise the price per life, which is an argument about scale rather than about individual obligation.

Now you. Someone replies that they would wade into the pond but will not give, because the child in the pond is right there. State that reply as a premise and test it.

The serious objections

Three objections have real force, and they are not the ones usually offered.

The demand is unlivable. This is the objection of the earlier lesson on consequentialism, applied here. The strong principle requires giving until further giving would cost you something of comparable importance to a life, which means until you are near subsistence, and a morality no one can live by has failed at something. The reply that has most traction is not philosophical but practical: Singer himself, and the movements built on his argument, have mostly settled on a defensible threshold, such as the ten percent of the Giving What We Can pledge, on the grounds that a demand people will actually meet produces more good than one they will reject. That is a decision-procedure answer rather than a criterion answer, with all the awkwardness that distinction carries.

Aid may not work. This is an empirical objection and the strongest of the three. The debate between William Easterly, who argues that large-scale aid has a poor record and often entrenches bad governments, and Jeffrey Sachs, who argues that targeted investment can break poverty traps, ran for a decade without resolution because neither side had the right kind of evidence. What changed the field was the randomised evaluation movement, for which Abhijit Banerjee, Esther Duflo and Michael Kremer shared the Nobel prize in economics in 2019, and its result is not that aid works or does not but that the variance between interventions is enormous. Some, like nets and vitamin supplementation, have strong trial evidence. Others, including much of what is popular with donors, have none. Even the movement's own flagship result, the finding by Miguel and Kremer in 2004 that school-based deworming raised attendance and later earnings, was contested by a re-analysis in 2015 and remains disputed. The correct conclusion is not that giving is futile but that where you give matters far more than how much, which is a conclusion the argument survives.

It is compensation, not charity. Thomas Pogge argues that the framing as beneficence concedes too much. The global order, in its trade rules, its recognition of whoever holds power as entitled to sell a country's resources, and its arms sales, is one that wealthy countries designed and from which they benefit, and it plays a causal role in sustaining poverty. If so, the obligation is a negative duty not to harm, which every theory in this course recognises as stronger than a duty to help, and its content is institutional reform rather than donation. The critical claim is causal and contested, and if it holds it makes the argument stronger rather than weaker, while relocating what it asks for.

Example. A donor gives 500 to a charity that sends shoeboxes of gifts to children in poor countries, and is told that the same money given to a malaria programme would avert roughly a tenth of a death. Assess the reply that the donor should give where they feel moved to give.

The reply confuses two things that should be separated. As a claim about motivation it is sound: people give more when they feel connected to a cause, the identifiable victim effect is well documented, and Deborah Small, George Loewenstein and Paul Slovic showed in 2007 that donations to a named child substantially exceed donations prompted by statistics about many, and that adding the statistics to the named child's story reduces giving. So feeling is doing real work in getting money out of people at all. As a claim about where the money should go it is much weaker, because the feeling is responding to how the appeal was written rather than to how much good the money does, and the donor would presumably not endorse a policy of allocating aid by the vividness of its marketing. The honest position for a donor is therefore split: use whatever motivates you to give, and do not use it to decide where. That split is uncomfortable and it is the same criterion-and-procedure distinction that the lesson on consequences introduced, arriving in a domestic form.

Now you. Does the identifiable victim effect give any reason to think that the intuition behind the drowning child case is unreliable?

One thought too many

The deepest objection is not about how much is owed to strangers but about whether an impartial morality can accommodate anyone in particular.

Bernard Williams's case is short. A man can save one of two people from drowning, and one of them is his wife. If he saves his wife, and the justification he gives is that she is his wife and in situations of this kind it is permissible to save one's wife, he has had, in Williams's phrase, one thought too many. What we want is for the thought that she is his wife to be enough. A morality that requires him to check his partiality against an impartial permission has misdescribed what it is to love someone.

The objection has bite against every theory in the course, and it is worth seeing why. Consequentialism has it worst, since his wife's life and the stranger's count the same in the sum, and he needs a permission from outside. Kantian ethics does better, since nothing requires him to weigh, but the theory's emphasis on acting from duty rather than inclination makes trouble of its own. Contractualism does better still, since a principle permitting rescue of one's own could not reasonably be rejected by anyone, who each have families of their own. Virtue ethics does best, because the loving husband is the paradigm rather than a case to be licensed, and this is one of the tradition's strongest arguments.

The standard reply distinguishes the reason that motivates from the reason that justifies. A good husband is moved by the thought that she is his wife; the theory's job is to say whether such dispositions are ones we should have, not to be in his head at the moment of the rescue. That is Railton's sophisticated consequentialism again, and Williams's rejoinder is that the distinction does not survive reflection: a man who knows that his love is licensed by an impartial theory cannot un-know it, and his relationship to his own attachments has changed.

Example. A parent spends 3,000 on their child's music tuition rather than on bed nets. Assess this from each of the four positions in the course.

An act consequentialist condemns it as clearly as any act can be condemned, since the tuition buys a marginal pleasure and the alternative buys most of a life, and a consequentialist who does not feel the force of that has not understood their own theory. A Kantian does not condemn it: the parent has an imperfect duty of beneficence with latitude in how it is discharged, and a perfect duty to their own child arising from the relationship, so the choice falls within permitted discretion. A contractualist asks whether a principle permitting parents to devote substantial resources to their own children could be reasonably rejected by a distant child, and the answer is contested, since the distant child's burden is far greater, though a world with no partiality towards one's own children may be one nobody could accept either. A virtue ethicist asks what a good parent and a good person does, and refuses to treat the two as separable, which is why the tradition's answer usually takes the form of a criticism of a life rather than of an act: a parent who spends lavishly on their child and gives nothing to anyone is displaying a defect, and the defect is not located in the tuition.

Now you. Does Williams's objection show that Singer's argument fails?

Effective altruism, and what happened to it

The movement built on this argument is worth a paragraph as a test case, because it is what happens when a philosophical argument is taken seriously by a lot of people at once.

Its programme is simple: give a meaningful fraction of your income, and give it where the evidence says it does most good. The Giving What We Can pledge of ten percent, the growth of cost-effectiveness evaluation, and a substantial redirection of donations towards malaria and direct cash transfers are its results, and they are real.

Its problems are also real and were mostly predicted by its critics. Measurability bias pushes attention towards interventions whose effects can be counted, and away from political and institutional change that may matter more and cannot be evaluated in a trial, which is Pogge's objection in another form. The internal logic of expected value led parts of the movement from present suffering towards speculative long-run risks, where the numbers are guesses and the guesses do all the work, which is the population arithmetic of the earlier lesson on the good arriving with a budget. And the collapse of the cryptocurrency exchange FTX in 2022, whose founder was a prominent funder and an explicit advocate of earning to give, was an object lesson in what a theory of criterion without a theory of character permits: a person reasoning entirely in expected value, with no constraint he could not calculate his way past, and enormous harm as the result.

None of that touches the original argument, and all of it bears on how the argument should be held. A reader who has followed this course to here should be able to say exactly why: the argument establishes a conclusion about what to do with money, the failures were failures of constraint and of character, and the three families are not competitors so much as answers to different questions that all have to be answered.

What survives

Strip away the movement and the numbers change less than the rhetoric suggests.

The second premise, in its weak form, is very hard to reject. Distance does not withstand examination as a moral variable. The empirical objection is real and redirects rather than defeats the argument. The demandingness objection defeats the strong version and leaves a substantial obligation standing. And the partiality objection shows that whatever the obligation is, it cannot be one that requires you to hold your own life at arm's length.

What that adds up to is a conclusion most people have never accepted and few can refute: that giving a real fraction of one's income, to interventions chosen on evidence rather than on appeal, is not generous but required, and that the amount is a matter for argument rather than the principle. If a reader disagrees, this course has given them the tools to say which premise they deny, and that is the last thing it has to teach.

Every position in this course is a set of answers to a small number of questions, and once you can see the questions the map is small enough to locate yourself on.

That is the claim this last lesson makes good. It is not a summary; the earlier lessons contain the material. It is a method for holding a view: stating it so that an opponent knows what to attack, answering the strongest objection rather than the most convenient one, arguing usefully with someone whose foundations differ from yours, and deciding what to do while remaining unsure which theory is true, which is the situation everyone is actually in.

Six questions

Here is the whole subject as a questionnaire. Every named theory answers all six, and disagreements that look enormous usually turn out to be disagreements about one or two.

What makes an outcome good? Pleasure, preference satisfaction, a list of objective goods, or a hybrid. Answer this and you have fixed what a consequentialist is maximising and what anyone else is weighing when they weigh consequences at all.

Do only outcomes matter? If there are acts you may not perform even when performing them would make things go better, you have constraints, and you owe an account of where they come from and whether they hold absolutely.

How much may you favour yourself and your own? From full impartiality, through an agent-centred prerogative, to strong special obligations grounded in relationships.

Whose interests count, and how much? All humans and no animals, all sentient beings equally, a graded scale, or something relational. This is the question with the largest consequences and the one people argue about least carefully.

How do burdens on different people combine? Fully aggregative, so enough small harms outweigh a large one; individualist, so they never do; or something in between with a threshold.

What is the primary object of assessment? The act, the rule, or the agent. This determines the shape of the theory rather than its verdicts, and it is the difference between "was that the right thing to do" and "what sort of person does that".

The table is a starting grid, not a taxonomy to be memorised, and its real use is diagnostic: when you disagree with someone, find the row they are in and the column you differ on.

Example. Two people disagree about whether a government should divert flood defence funding from a wealthy town to a poorer one where more lives are at risk. Locate the disagreement on the six questions.

Almost certainly on aggregation and partiality, and not on the good. Both parties value lives and property, so the first question is not in dispute. If one says the greater number of lives settles it and the other says the wealthy town's residents have paid for their defences and have a claim, the disagreement is partly about entitlement, which is Nozick's historical question, and partly about whether the state may weigh its citizens against each other at all, which is the aggregation question. Notice what is not in dispute: neither is arguing about consequences versus duties in the abstract, and a conversation conducted at that altitude would never reach the point where they differ. Locating it takes two minutes and makes the rest of the argument possible.

Now you. Someone says they are a consequentialist but that torture is always wrong. Is that a contradiction?

Stating a view so it can be attacked

A view that cannot be attacked has not been stated. Four elements make the difference.

Name the criterion. Not "I think consequences matter" but "an act is right if and only if its expected welfare is at least as great as that of any alternative". Vagueness at this stage is what lets a position slide out from under every counterexample.

Name the scope. Whose welfare, over what time, with what discount, and which beings are included. A great many disputes that look like disagreements about the criterion are disagreements about scope.

Name the priority ordering. If you have several considerations, say what happens when they conflict: lexical priority, weighing by judgement, or a threshold. A list of considerations with no ordering is not yet a view.

Name the refuter. Say what case, if you accepted its description and still found the verdict intolerable, would make you abandon the position. If nothing would, you have an allegiance rather than a view, and the rest of the discussion is theatre.

That last requirement is uncomfortable and it is the one worth insisting on, of yourself first. It is also the direct descendant of the standard the free will course ended with, and for the same reason: a position held with no conditions attached to it cannot be argued about, only announced.

Answering the best objection

When a counterexample lands, there are exactly three moves, and knowing which one you are making is most of the skill.

Deny the case. Argue that the description is incoherent, or that the stipulations smuggle in the conclusion, or that the case as described could not occur. This is legitimate and is overused: the fact that a case is unrealistic is not by itself an objection, since the point of stipulating is to isolate a variable.

Distinguish. Accept the case and show that your principle, correctly stated, does not apply to it. This is the most productive move, because it forces you to state the principle more precisely, and the more precise version is a better theory whether or not it survives the next case.

Bite the bullet. Accept both the case and the verdict, and argue that the intuition against it is unreliable. This is respectable and it has a price: you must supply the debunking story, explaining why we would have that intuition whether or not it were correct, and you must accept the verdict consistently in every relevantly similar case rather than only in the seminar.

What is not available is silence, or the reply that the objector is being unreasonable, or the observation that the other theory has problems too. That last is the commonest evasion in the subject. Every theory has problems; the question is which set you can live with, and that comparison requires stating your own problems as clearly as your opponent's.

Example. You hold that an act is right if it maximises expected welfare. Someone puts the transplant surgeon to you. Work all three moves and say which is strongest.

Denying the case means arguing that the secrecy stipulation is incoherent, since nothing stays secret and the surgeon cannot know that it will, and therefore that the expected welfare of killing is negative. This is weak, because the case can be restated to close the gap, as the earlier lesson showed. Distinguishing means adopting a rule or two-level version: the criterion applies to codes or dispositions rather than to individual acts, so the surgeon's act is wrong because the disposition to perform it is disastrous. This is strong, and its cost is that you have changed your theory, and inherited the collapse objection or the boundary problem that comes with it. Biting the bullet means accepting that the surgeon should kill, and supplying a debunking account of the contrary intuition, most plausibly that we have a well-founded and generally useful horror of doctors who kill, which misfires in a case stipulated never to occur. This is respectable and requires you to say the same thing about every structurally identical case, including ones where the victim is you. The strongest is the second, and noticing that it is a change of theory rather than a defence of the original is the honest part.

Now you. You hold that there is an absolute constraint against intentionally killing the innocent. Someone puts a case in which one such killing prevents the deaths of a million. Work the three moves.

Arguing with someone whose foundations differ

Most real disagreements are not between a consequentialist and a Kantian. They are between two people who have never stated a criterion, and the useful techniques do not require either of them to.

Internal critique is the most powerful and the least used. Instead of arguing from your premises, argue from theirs: find a commitment they hold and show what it entails about the case in dispute. This is what the earlier lessons did throughout, and its advantage is that it requires no shared foundation at all, only that your interlocutor prefers consistency to inconsistency.

Locating the disagreement comes next. Run the six questions until you find the one you differ on. Very often the answer surprises both of you: two people arguing about immigration may agree completely about consequences and differ about whether a state may weight its own citizens, which is the partiality question, and a whole evening spent on economic forecasts was wasted.

Separating the empirical from the normative is the third, and it is worth doing early. Ask what fact, if established, would change their view. If one exists, the argument is partly empirical and you can look it up. If none does, you have found the moral premise, which is where the argument actually is.

Argument runs out somewhere, and it is worth knowing where. Against someone who accepts a monstrous conclusion consistently, has no inconsistency to exploit, and shares no premise with you, there is nothing further to say, and the rational response is not a better argument but resistance. That is a real limit on the reach of moral reasoning and it does not undermine anything above it, since the same is true of any subject: someone who consistently rejects the evidence of their senses cannot be argued into physics either.

What to do while uncertain

There remains the practical problem: you have read all this, you find each family partly convincing, and you have to act.

The bad answers are to pick a team and follow it off a cliff, and to conclude that since the theories disagree, nothing is settled. The second is plainly false: the theories agree about the overwhelming majority of cases, and their agreement is more informative than their disagreement. Where all live theories give the same verdict, the verdict is as well supported as anything in the subject.

For the residue, three approaches are used. My favourite theory says act on the theory you find most plausible, which is simple and throws away the information contained in your uncertainty. Maximising expected choiceworthiness, developed by William MacAskill and Toby Ord, treats moral uncertainty like empirical uncertainty: weight each option by your credence in each theory and by how good the option is on that theory. Its difficulty is intertheoretic comparison, since there is no common scale on which a utilitarian's welfare units and a Kantian's violation of dignity can be measured, and a theory that assigns infinite disvalue to some acts swamps the calculation. Avoiding the worst is the practical residue that survives both: prefer options that are acceptable on every theory you take seriously, and treat an option that is catastrophic on any live theory as one requiring a very strong case.

Three rules of thumb follow from this and are worth more than they look. Where theories converge, act with confidence. Where they diverge, prefer the reversible option, because the value of information is high and your credences will change. And be suspicious of any argument whose conclusion is that you may do something you already wanted to do, since that is precisely the pattern the debunking test in the lesson on method was designed to catch.

Example. You are a doctor and a patient with a poor prognosis asks for help to die. The law permits it. Your credence is roughly even between a view on which autonomy settles the matter and one on which there is a constraint against killing. What does each approach recommend?

My favourite theory picks the more plausible of the two and acts on it, which given roughly even credences is close to arbitrary. Expected choiceworthiness asks how much is at stake on each view: on the autonomy view, refusing imposes a serious and irreversible burden on the patient, and on the constraint view, complying is a grave wrong, and the calculation cannot be completed without a common scale, which is the objection. Avoiding the worst gives usable guidance and it is worth stating: refer the patient to a colleague who has no such reservation. That option is acceptable on the autonomy view, since the patient's request is met, and much less bad on the constraint view, since you have not killed anyone, so it dominates both of the pure options. Conscientious objection with a duty to refer, which is what most jurisdictions with assisted dying laws have adopted, is exactly this structure, and it is a good illustration that institutional compromises are often solutions to moral uncertainty rather than failures of nerve.

Now you. Apply the same reasoning to a decision about whether to eat meat, given uncertainty about animal moral status.

What this course has shown

It has not shown which theory is true, and no honest course could.

What it has shown is that the disagreements are structured. Every position is a set of answers to six questions. Every argument has a moral premise, and finding it is a technique rather than a talent. Every universal principle can be tested by one case, and every response to a counterexample is one of three moves. The empirical questions inside moral disputes are separable and often already answered. And the strongest arguments in the subject, on both sides of every live dispute, have the same form: they take a premise their opponent already holds and follow it somewhere the opponent did not want to go.

The free will course ended by asking readers to locate themselves on a map of positions about responsibility and say what would change their mind. This one ends with the same demand about conduct, and the two questions are less separable than they look: what you think a person deserves depends on what you think they should have done, and what you think they should have done is settled by nothing except argument of the kind this course has been conducting.

The last thing to say is the one Aristotle said first. Knowing all this is not the same as being good, and a course cannot make anyone good. What it can do is remove the excuses: after this, "who is to say" is not available, and neither is the claim that the question is too hard to think about. It is hard, it is thinkable, and the tools are now in your hands.