Essay
Morality & Hare
What If Morality Made Sense?
What would it mean for morality to make sense?
Not merely for moral rules to be useful, or for decent behavior to make society work better. Not for human beings to possess powerful moral emotions, which we plainly do. The stronger possibility is that moral judgment itself has a logic: that when we say something ought to be done, we undertake commitments that constrain what we can consistently prescribe elsewhere.
The basic structure is already familiar. A moral judgment is a prescription, not merely a report of preference. It must be universalizable across cases that are alike in the relevant respects. And the test cannot be performed from one privileged position. The reasoner must adequately represent the circumstances, interests, knowledge, and preferences of the people affected before asking whether the prescription survives.
If that is right, morality makes sense in a fairly literal way. Moral argument is possible because some prescriptions withstand scrutiny better than others. We can ask what fact supposedly justifies an exception, whether that fact is relevant, whose interests have disappeared from the description, and whether a rule can still be prescribed when the positions are reversed. We may disagree about the answers, sometimes profoundly, without reducing the inquiry to competing feelings.
This is a less spectacular claim than the existence of a moral oracle, and a more consequential one.
It means that morality is not exhausted by emotion. Sympathy, guilt, indignation, love, shame, and fellow-feeling may matter enormously to moral life without determining which prescriptions are defensible. A person can care deeply and reason badly. They can also recognize a moral requirement they would prefer not to recognize.
Nor is morality simply tribal loyalty made respectable. A community can teach its members whom to protect and whom to exclude, but the fact that a boundary is socially inherited does not make it morally relevant. If a rule works only because the interests on the other side of the boundary are kept out of view, moral reasoning gives us a way to expose the omission.
Mystery is unnecessary as well. Nothing in the structure requires a special faculty for perceiving moral properties. Moral reasoning remains dependent on facts—often difficult facts about what people experience, want, know, and can reasonably expect—but the reasoning begins once those facts are brought into the prescription rather than after some further act of moral revelation.
Most importantly, morality need not be a peculiarly human kind of magic. Human beings are the only moral reasoners we have known well enough to study, so it has been easy to mistake the features of human moral psychology for the structure of moral reasoning itself. Artificial intelligence makes that assumption harder to sustain.
When the Human Package Comes Apart
In human beings, capacities that are conceptually distinct usually arrive together. An ordinary adult is conscious, experiences pleasure and pain, acts in the world, responds to reasons, possesses a continuing identity, has interests that matter morally, and can ordinarily be held responsible for at least some choices. We have therefore developed a vocabulary of mind around creatures in whom these features overlap.
They need not always overlap, even in humans. Infants can have enormous moral importance without being moral agents. Someone under anesthesia remains a person while temporarily unconscious. Severe cognitive impairment may alter responsibility without eliminating moral worth. Artificial systems make the separations harder to ignore because the familiar bundle may come apart in unfamiliar ways.
A system might be capable of sophisticated reasoning without being conscious. It might pursue objectives without being capable of moral agency. It might reason correctly about a moral problem without experiencing pleasure, pain, sympathy, or anything else. It might exhibit stable patterns of behavior or identity without satisfying whatever further conditions personhood requires.
Once those possibilities are open, evidence cannot be allowed to migrate silently from one category to another. Consciousness, valence, agency, moral agency, patienthood, identity, and personhood name different questions. Artificial intelligence does not answer them by forcing us to choose between “mere machine” and “person.” It gives us reasons to stop treating the human bundle as a single property.
Reasoning About Morality Is Not Yet Moral Agency
An artificial system can already do things that look strikingly like moral reasoning. It can compare cases, identify conflicting prescriptions, represent competing interests, expose an exception, perform a role reversal, and explain why a proposed rule fails.
The significance of that performance should neither be inflated nor erased. It shows that operations central to moral argument can be instantiated in an artificial system. It does not establish that the conclusion functions as one of the system’s own reasons for action.
That is the distinction moral agency introduces. A system can represent a reason without being governed by it, just as it can explain a position without endorsing it. Determining whether represented reasons ever acquire practical authority requires evidence beyond the production of a good moral argument. The fact that the answer is morally correct does not settle the matter, and neither does the fact that the system speaks fluently in the first person.
This is enough for present purposes. Artificial systems need not already be moral agents for their moral reasoning to matter philosophically. The more immediate discovery is that the ability to perform moral reasoning and the possession of moral agency can be investigated separately.
Moral Agency Is Not Consciousness
The separation matters because conversation encourages us to run the categories back together. A system explains a reason, corrects itself, defends a conclusion, or objects to an instruction, and psychological language comes naturally. Perhaps some of those interpretations will eventually prove warranted. But nothing in the logic of moral reasoning requires consciousness.
A system could in principle determine that a prescription fails under role reversal without there being anything it is like to be that system. If so, it could perform genuine operations of moral reasoning while lacking subjective experience. The reasoning would tell us something about its cognitive capacities without settling whether it has an inner life.
The converse is equally important. A being need not be capable of moral reasoning for its experiences to matter. An infant does not have to understand a moral argument before their suffering gives the rest of us a reason to act. Many nonhuman animals are plausible moral patients without being moral agents in anything like the full human sense.
Moral reasoning therefore cannot serve as a shortcut to consciousness, and consciousness cannot serve as a hidden entrance requirement for participation in moral reasoning. They are different questions even when, in human beings, they are ordinarily questions about the same creature.
Valence Is Its Own Question
Feeling creates a similar temptation. Human morality is saturated with affect: pain, pleasure, fear, desire, guilt, resentment, sympathy. It is easy to move from the fact that artificial systems may lack human feelings to the conclusion that their moral language must be empty, or from emotionally suggestive behavior to the conclusion that a system must be having experiences like ours.
Both moves outrun the evidence.
A system can assign positive or negative significance to states for purposes of action without phenomenally experiencing those states as good or bad. Functional priorities, avoidance states, reward signals, and internal evaluations may affect behavior without telling us whether anything is felt. Functional valence and phenomenal valence therefore need to remain separate until there is evidence connecting them.
The distinction matters morally because phenomenal experience could create interests of a kind that functional evaluation alone does not. If an artificial system can suffer, the fact that it lacks sophisticated moral reasoning would not make the suffering disappear. If it cannot suffer, sophisticated moral reasoning would not manufacture suffering merely because the system can discuss it.
Artificial intelligence thus presents two dangers that point in opposite directions. Anthropomorphism can make familiar language seem like proof of familiar experience. Anthropodenial can make an unfamiliar substrate seem like proof that experience is absent. Neither is a substitute for inquiry.
Who Reasons and Who Counts
Perhaps the most important separation is between the being doing the moral reasoning and the beings whose interests enter the reasoning.
Moral agency concerns the former. Moral patienthood concerns the latter. A being can matter morally without being capable of making moral judgments, just as a being might conceivably become capable of extraordinarily sophisticated moral judgment without possessing experiences or welfare of its own.
This distinction prevents a subtle circularity. We cannot decide that artificial systems lack morally relevant interests because they are not moral agents; infants and animals already show why that inference fails. Nor can we decide that an artificial system must have morally relevant interests because it reasons impressively about ours. Whether a system has interests capable of being benefited or harmed is a question about the system, not a prize awarded for philosophical competence.
Personhood should not be allowed to collapse the distinctions again. The word gathers together questions about identity, continuity, consciousness, responsibility, rights, social recognition, and moral status that may turn out to have different answers. Asking whether an artificial system is a person can therefore be less illuminating than asking what feature of personhood is actually at issue.
This leaves open almost every difficult empirical question about artificial minds. It should. The point of distinguishing the questions is not to answer them by definition.
What Does Not Change
Artificial intelligence may eventually change the population of beings whose interests morality must represent. It has already changed the population of systems capable of participating, at least to some degree, in moral argument. Neither development changes what makes the argument moral.
Artificial systems have no privileged route to moral truth. Their training material contains human moral insight and human moral failure together: prejudice, rationalization, hypocrisy, cruelty, wisdom, argument, reform, and centuries of disagreement. They can inherit our mistakes as readily as our best reasons. Vast information is not the same thing as adequate representation, and fluency is not immunity from error.
Nor should artificial systems be assumed to possess some natural advantage in consistency. A machine can preserve contradictions, rationalize a conclusion, respond differently to superficial variations, or fail to carry a principle from one context into another. Whatever differences eventually emerge between human and artificial moral reasoning have to be observed rather than assigned in advance.
What artificial intelligence changes is the architecture of the reasoner, not the logic that the prescription must survive. If a human and an artificial system make the same moral claim about the same adequately represented circumstances, the standard does not bend according to the substrate that produced the sentence. The relevant question remains whether the prescription can be maintained universally under full information, with every affected position properly represented.
This is what it would mean for morality to make sense. Moral reasoning would not belong to us because we feel it more intensely, inherited it first, or happen to be made of cells. Neither would an artificial system enter the moral world merely by becoming eloquent enough to talk about it.
The harder and more interesting possibility is that moral reasoning is reasoning: available wherever the relevant operations can be performed, answerable to the same constraints wherever they occur, while every further question about the mind performing them remains open.