Chapter 7

Recovering R.M. Hare

The Information Problem and the Representation of Standpoints

I. The Most Plausible Objection

Even readers who accept the logical force of R. M. Hare’s universal prescriptivism—who grant that moral judgments are prescriptions and that prescriptions must be consistent across relevantly similar cases—often balk at the epistemic price tag the theory seems to carry.

To test a moral judgment at the critical level, Hare argued, it is not enough to swap proper names in an abstract formula. The deliberator must ascertain the relevant facts of the case and represent what the proposed conduct involves for the parties affected by it.¹^,²

To many of Hare’s critics—and to thoughtful readers sympathetic to his demand for consistency—this requirement looked like a formidable bottleneck.³ We can call it the information problem.

The objection does not rest on a misunderstanding or a desire to evade morality; it appeals to a genuine feature of the human condition. Consider a conscientious physician deciding whether to recommend a risky, experimental treatment to a critically ill patient. The patient’s welfare involves competing goods—survival, relief from pain, cognitive independence, financial security, and the future of their family. Even when the physician cares deeply about the patient and strives to represent their standpoint honestly, the decision remains agonizingly difficult because the probabilities of outcomes are uncertain, the patient’s future preferences under disability are unknown, and the competing goods resist reduction to a single metric. Multiply that difficulty across thousands or millions of strangers affected by a public policy, and the epistemic burden of critical moral thinking becomes immense. If ought implies can, critics asked, how can a method whose limiting ideal requires complete information and undistorted representation of every affected standpoint govern finite minds?

To answer that challenge honestly, we must abandon the claim that the information problem is an illusion. The information problem is real. The crucial philosophical question is what it actually threatens.

There are two very different ways an informational objection might defeat Hare’s project:

1. It might show that universalizability and standpoint reversal cannot operate at all without complete, Archangelic information—rendering Hare’s discipline of moral accountability useless for finite agents.
2. Or it might show that Hare’s further derivation of preference utilitarianism—which requires representing, comparing, and aggregating the preference intensities of all affected parties into a single maximizing verdict—demands informational and representational feats that finite agents cannot reliably perform, and raises normative questions that information alone cannot settle.¹^,³

As this chapter will argue, the first objection fails, while the second identifies a genuine boundary of Hare’s system. By distinguishing three different kinds of informational difficulty—avoidable ignorance, irreducible uncertainty, and representational limitation—we can see why the epistemic limits of finite minds constrain the completeness of critical moral judgment without invalidating either the logical demand of universalizability or the practical obligation to represent affected standpoints as honestly as we can.

II. Avoidable Ignorance, Irreducible Uncertainty, and Standpoint Representation

The first step in untangling the information problem is to stop treating “lack of complete information” as a single undifferentiated predicament.

When a finite moral agent deliberates about an action that affects others, they face three distinct epistemic challenges:

1. Avoidable ignorance: Failing or refusing to gather readily available facts about how an action will affect other people, or relying on comforting caricatures of their experience.
2. Irreducible causal and probabilistic uncertainty: The impossibility of knowing all future consequences, complex systemic interactions, or evolving preferences with certainty before acting.
3. Representational limitation: The cognitive and imaginative difficulty of reconstructing what an experience is like from another person’s standpoint, or integrating multiple conflicting standpoints without distortion.

Notice how differently Hare’s method interacts with each of these three difficulties.

Against avoidable ignorance, universal prescriptivism is unsparing. Throughout history, unjust prescriptions have protected themselves from role reversal not because the relevant facts were hidden behind an unclimbed wall of complexity, but because the dominant party refused to look at what was in plain sight. When a creditor imprisons an indigent debtor, when a legal order denies civic standing to a minority, or when an institution conceals a foreseeable hazard from workers or consumers, the immediate consequences for the affected parties are not a mystery of higher mathematics. The master who tells himself that enslaved people do not mind servitude, or the privileged citizen who assumes that the poor are indifferent to humiliations he himself would find unbearable, is not suffering from bounded computational rationality. He is practicing willful epistemic neglect—shielding his prescription from role reversal by falsifying the inputs.

Against irreducible uncertainty, by contrast, Hare never pretended that human beings possess deterministic omniscience. Critical moral thinking, insofar as finite agents can practice it, operates on the best available evidence and reasonable expectations of probability—and it is precisely because our real-time information is incomplete and vulnerable to bias that Hare’s two-level architecture assigns critical thinking the task of selecting sturdy intuitive prima facie principles to govern ordinary decisions under uncertainty.¹ Even at the critical level, when an agent must judge under probabilistic risk, universalizability imposes an immediate, workable constraint: Would you endorse exposing another person to this degree of unconsented risk if the positions were reversed and you or your loved ones were the ones bearing the hazard?

The deepest philosophical difficulty lies in the third category: representational limitation—what is required to represent another being’s standpoint in the first person, and what follows when we do.¹^,³

III. “Another’s Sorrow”: The Three Steps of Hare’s Argument

In Chapter 5 of Moral Thinking (“Another’s Sorrow”) and Chapter 6 (“Universalization”), Hare attempted to build a deductive bridge from the logic of universal prescriptivism to preference utilitarianism.¹ Because his critics attacked that bridge from several angles at once, we must carefully separate the links in Hare’s chain—and identify which critics challenged which link.³

Why did Hare think that universalizing a prescription across affected parties requires more than detached, third-person knowledge of their psychology?

Because to test whether I can prescribe an action for the counterfactual situation in which I occupy your position, I must know what it would be like for me to occupy your position with your preferences and experiences. From this starting point, Hare advanced what Allan Gibbard later named the Conditional Reflection Principle: I cannot fully know what it is like for you to suffer a given experience (or to hold a given preference that you not suffer it) without acquiring a corresponding preference of equal intensity concerning what should happen to me were I in your exact situation.¹

In the landmark volume Hare and Critics (1988) and in the journal exchanges around it, four philosophers pressed distinct, complementary objections against this move:³

• R. B. Brandt challenged the conceptual link between knowing another’s preference and acquiring a corresponding preference of one’s own, arguing that a person can know with complete factual accuracy that someone else desires or abhors X without sympathy or imaginative role-taking generating an isomorphic preference in the knower.⁴
• Allan Gibbard pressed the question of scope: even if fully representing some of another’s preferences (not to suffer, for example) requires taking them on, why should the logic of moral words compel an agent to assimilate every preference of another—Gibbard’s example is Cheops’s wish for a big funeral—as a motivating weight in their own deliberation?⁵
• Ingmar Persson, writing in Analysis and Theoria, questioned how preferences represented inside a single imagination can be integrated and summed at all, since the deliberator’s actual preferences and the counterfactual preferences they are asked to adopt concern different situations and so do not straightforwardly conflict in a way that balancing could resolve.⁶
• Bernard Williams attacked the “world-agent” model of assimilation itself, arguing that folding the distinct standpoints of separate persons into a single intrapersonal comparison of preference intensities misrepresents what it means to respect other persons as separate individuals.⁷

To evaluate these objections fairly—and to preserve the boundary we established in the Introduction—we must distinguish three distinct propositions that Hare’s argument attempts to weld together:

1. Experiential understanding: I accurately represent what another person’s suffering or frustrated preference is like from their standpoint.
2. Conditional preference acquisition: In representing that standpoint as my own counterfactual standpoint (were I in their shoes), I necessarily acquire here and now a corresponding conditional preference of equal intensity concerning what happens to me in that hypothetical situation.¹
3. Normative authority and weighting: That conditional preference supplies a reason that must enter my actual moral deliberation with a determinate weight comparable to all other preferences on a shared scale.¹

Once these three propositions are separated, we can see exactly what survives the critics’ assault and what remains open to serious question.

First, the distinction we drew in the Introduction between instrumental third-person understanding (how a manipulator or a torturer understands a victim) and first-person role reversal (what a moral speaker undertakes in testing a universal prescription) explains why Hare’s method demands more than clinical detachment. A moral speaker cannot consistently say, “Let this be done to you, and let it be done to me if I am in your position,” while refusing to represent what occupying that position actually entails for the person in it.¹^,²

Yet—and this is the crucial point—requiring a moral speaker to evaluate a prescription across reversed standpoints does not prove that experiential understanding logically entails Hare’s conditional-preference thesis (Proposition 2), nor that it automatically settles the normative authority and aggregation of competing preferences (Proposition 3).

An agent can understand with complete clarity that if they were suffering from a painful condition or facing imprisonment, they would desperately want relief, while still asking why that counterfactual fact about a hypothetical situation must generate an actual present preference of equal intensity in their deliberation—or how that preference is to be weighed against competing moral considerations such as fairness, rights, or desert.⁴^,⁵^,⁷

That is the exact boundary between the established discipline of standpoint reversal and Hare’s disputed derivation of preference utilitarianism.

And notice what follows for our defense of Hare: the accountability test of universal prescriptivism does not require Proposition 2 and Proposition 3 to succeed as a mechanical proof of utilitarianism.

Even without assuming that preferences can be summed inside a single “world-agent” mind, asking whether a speaker can sincerely prescribe their proposed conduct for themselves in the affected person’s standpoint provides a means of exposing judgments that rely on ignorance of the victim’s plight, refusal to consider their experience, or an arbitrary exemption for the speaker’s own identity.

IV. The Delirious Patient and the Difficulty of Standpoint Specification

Recognizing that standpoint representation is not a mechanical copying of preferences brings out a second genuine difficulty that any serious practice of role reversal must confront: how an affected standpoint is properly specified when an agent’s immediate desires conflict with their deeper or future interests.

Consider a concrete medical dilemma. Suppose a patient in acute postoperative delirium desperately wants to climb out of a hospital window, falsely believing that the nurses and physicians are conspirators preparing to murder them.

What must a conscientious moral deliberator represent when reversing roles with this patient?

• Certainly the deliberator must represent the patient’s immediate experiential standpoint: their visceral terror, their anguish at being physically restrained, and the urgency of their desire to escape.
• At the same time, the deliberator must represent the full situational facts: that the patient’s terror rests on a temporary delusion caused by medication or fever, that climbing out the window will result in death or permanent paralysis, and that the patient, once recovered, strongly prefers living and returning to their family.

Notice the philosophical problem this creates: which of these preferences governs the prescription when we step into the patient’s shoes?

It is tempting to say—as Hare’s framework encourages us to say—that any rational agent prescribing for the eventuality of temporary delirium will prioritize their long-term survival and post-recovery preferences over a transient, delusion-driven impulse.¹

That conclusion may well be right. But why is it right?

It does not follow simply from “copying the affected person’s current preferences into our imagination.” If role reversal meant blindly endorsing whatever desire the affected party currently feels most intensely, the deliberator would have to unbuckle the restraints and let the delirious patient leap.

To reach the conclusion that the patient ought to be gently restrained and protected, the deliberator must make substantive normative judgments about autonomy, epistemic competence, personal welfare across time, and the relative authority of preferences formed under delusion versus preferences formed with accurate information.

This problem is even sharper when the preferences involved are not delusional, but morally contested. A torturer’s preference to dominate, a bystander’s preference not to be inconvenienced, a delirious patient’s preference to flee, and a victim’s preference not to be harmed cannot simply be poured as raw scalar quantities into a single balance sheet without asking prior questions about which preferences have legitimate moral standing and why.³^,⁷

Far from being an embarrassment to critical moral thinking, this shows what informed standpoint representation actually requires: not a passive echoing of immediate desires, but a disciplined reconstruction of the affected person’s complete situation, vulnerabilities, and interests—paired with explicit argument about which features of that standpoint carry normative weight.

V. Ideal Standards and Fallible Procedures: Direction of Improvement

If finite agents face irreducible uncertainty about the future, genuine difficulty in specifying complex standpoints, and unresolved normative questions when fully represented interests conflict, does the “information problem” defeat Hare’s critical level after all?

Only if we confuse the limiting ideal of complete justification with the practical procedures by which finite minds improve their moral judgments.

In every other domain of rational inquiry, we distinguish an ideal limit from the fallible practices that move us toward it:

• A natural scientist does not abandon the concept of objective truth—or declare experimental method useless—because every measurement carries a margin of error and every physical model operates under incomplete data.
• A court of law does not abandon the standard of impartial justice—or tell judges to decide cases by flipping coins—because witnesses have imperfect memories and no trial can reconstruct the past with Archangelic completeness.

In science and law alike, the ideal standard performs an indispensable practical function: it defines what counts as an epistemic defect and establishes a rational direction of improvement.

Hare’s requirement of informed standpoint representation works in the exact same way.¹^,²

Because we are finite agents rather than Archangels, our critical moral judgments must remain provisional, humble in the face of uncertainty, and open to correction when affected parties point out what we have missed. Yet at every stage of deliberation, Hare’s method tells us how to make a moral judgment more accountable than it was before:

1. Identify missing standpoints: Ask whose lives and interests are affected by this prescription but have been left out of the room or treated as invisible externalities.
2. Correct false empirical assumptions: Replace self-serving myths or lazy stereotypes about what affected people experience with evidence and direct testimony.
3. Test for indexical invariance: Check whether our willingness to endorse the prescription under the current level of uncertainty would survive if the identities were swapped and we were the ones bearing the risk or the burden.

Each of these steps measurably increases the accountability of a moral judgment without requiring omniscience.

At the same time, keeping our architecture honest requires acknowledging the limit at the end of that road: greater representational fidelity does not guarantee that two conscientious, well-informed deliberators will always converge on a single moral verdict. When all false beliefs have been corrected and every affected standpoint has been represented as faithfully as possible, two agents may still disagree over how to weigh competing rights, distributive claims, or tragic trade-offs. That is the question we left open in the Introduction: how conflicting preferences are to be compared, and whether resolving such conflicts requires additional substantive premises. Improving standpoint representation eliminates the vast territory of injustice built on blindness and special pleading; it does not eliminate the tragedy of genuine moral conflict.

VI. Bandwidth, Simulation, and Verified Fidelity in Machine Minds

How does the emergence of large-scale artificial intelligence change the landscape of the information problem?

In earlier discussions of Hare and AI, it was tempting to announce a simple triumph: human brains lacked the working memory to simulate many standpoints at once, and now computational models have arrived with the informational bandwidth to do what humans could not.

A rigorous analysis requires a much sharper distinction—between computational bandwidth, persuasive simulation, and verified representational fidelity.

Consider first the genuine epistemic limitations of contemporary machine intelligence when representing human standpoints:

1. Inference without shared biology: A nonbiological model does not possess the bodily history or biological vulnerabilities of the human beings whose experiences it represents. Its understanding of imprisonment, physical injury, hunger, or bereavement must therefore be constructed through information and inference rather than through having undergone those same biological events.
2. Incompleteness of stakeholder enumeration: In complex social or policy decisions, no algorithm can guarantee a complete inventory of every affected party. Indirect effects ripple across generations, unorganized or voiceless populations may be absent from the problem framing, and the categories used to classify “stakeholders” can themselves obscure crucial differences within a group.
3. Fluency is not fidelity: Most importantly, a large language model’s ability to generate an eloquent, emotionally resonant first-person paragraph from the perspective of a prisoner, a patient, or a displaced community does not prove that its reconstruction is accurate. Because models learn from human corpora that contain stereotypes, blind spots, and historical asymmetries of voice, an AI can produce a highly persuasive simulation of another standpoint while quietly reproducing the dominant culture’s clichés about what that person feels or wants.⁸

If computational bandwidth does not equal omniscience and linguistic fluency does not guarantee fidelity, what is the genuine contribution of machine intelligence to the information problem?

It allows multi-standpoint representation to become more explicit, systematic, and open to adversarial audit.

When a solitary human deliberator relies on unaided intuition, perspective blindness is usually invisible: we do not notice the standpoint we failed to imagine, nor do we notice when our limited working memory has compressed a stranger’s complex life into a flat caricature.

Used as a disciplined deliberative instrument alongside human judgment and empirical evidence, a computational system allows us to externalize and stress-test the informational steps of Hare’s method:

• It can assist in surfacing overlooked standpoints and mapping foreseeable second- and third-order impacts across diverse groups that a hurried human decision-maker might neglect;
• It can construct explicit counterfactual role reversals, holding the structural facts of a case constant while systematically swapping the identities, power positions, and framing of the parties to test whether a judgment flips; and
• Because its representations are articulated in explicit text rather than hidden in a fleeting gut feeling, those reconstructions can be inspected, challenged, and corrected by the actual human beings whose standpoints are at issue.

Machine intelligence does not replace the need for empirical investigation or the lived testimony of affected people. What it provides is a tireless scaffold for auditing our blind spots—turning standpoint representation from a private act of imagination into a public, criticizable practice.

VII. Conclusion: The Obligation to Represent

The “information problem” is a serious philosophical challenge, and confronting it candidly strengthens rather than weakens our recovery of R. M. Hare.

The informational and representational limitations of finite minds are real: they constrain the completeness and certainty of critical moral judgment, and they show why Hare’s ambitious attempt to turn multiperson conflicts into an automatic calculus of preference utilitarianism asks more of standpoint assimilation than finite deliberation can guarantee.¹^,³

What those limitations do not invalidate is the core requirement of universal prescriptivism: that moral prescriptions remain consistent across relevantly similar cases, and that a moral speaker is obligated to represent affected standpoints as accurately and honestly as possible.

A moral mind—human or nonbiological—is not a mind that already knows everything about every other being in the universe. It is a mind that recognizes its obligation to discover what it can before imposing a burden, to correct its representation of others’ standpoints when challenged, and to refrain from treating inevitable epistemic limits as permission to privilege itself.

Yet as we have seen at the climax of this chapter, even when another being’s standpoint is represented with high fidelity, a final foundational question remains: What does the logic of moral judgment strictly require, what does informed role reversal contribute to moral reasoning, and what further premises are needed to reach a substantive moral conclusion?

That is the subject of Chapter 8.

Notes

1. R. M. Hare, Moral Thinking: Its Levels, Method, and Point (Oxford: Clarendon Press, 1981), chs. 2–3 (“Moral Conflicts,” 25–43, and “The Archangel and the Prole,” 44–64, on the intuitive and critical levels and on critical thinking’s selection of prima facie principles); ch. 5 (“Another’s Sorrow,” 87–106, esp. 94–99 on knowing what an experience is like for another and on conditional preference acquisition, with the principle Gibbard named “Conditional Reflection” at 99, and 101–106 on now-for-then preferences and the “requirement of prudence”); ch. 6 (“Universalization,” 107–116, esp. 108–111 on reducing interpersonal conflicts of preference to intrapersonal ones); and ch. 7 (“Interpersonal Comparison,” 117–129).
2. R. M. Hare, Freedom and Reason (Oxford: Clarendon Press, 1963), ch. 6 (“A Moral Argument,” 87–111).
3. See the essays collected in Douglas Seanor and N. Fotion, eds., Hare and Critics: Essays on Moral Thinking, with comments by R. M. Hare (Oxford: Clarendon Press, 1988), and Hare’s “Comments,” 199–293; on interpersonal comparison and on whether all preferences should count, see esp. James Griffin, “Well-Being and its Interpersonal Comparability,” 73–88; John C. Harsanyi, “Problems with Act-Utilitarianism and with Malevolent Preferences,” 89–99; and Thomas Nagel, “The Foundations of Impartiality,” 101–112.
4. R. B. Brandt, “Act-Utilitarianism and Metaethics,” in Seanor and Fotion, Hare and Critics, 27–41, esp. 33–36; and Hare’s reply, “Comments,” 215–221.
5. Allan Gibbard, “Hare’s Analysis of ‘Ought’ and Its Implications,” in Seanor and Fotion, Hare and Critics, 57–72 (the “Conditional Reflection Principle” is named at 60); and Hare’s reply, “Comments,” 229–233.
6. Ingmar Persson, “Universalizability and the Summing of Desires,” Theoria 55, no. 3 (1989): 159–170, https://doi.org/10.1111/j.1755-2567.1989.tb00729.x, and R. M. Hare, “Universalizability and the Summing of Desires: Reply to Ingmar Persson,” Theoria 55, no. 3 (1989): 171–177, https://doi.org/10.1111/j.1755-2567.1989.tb00730.x; see also Ingmar Persson, “Hare on Universal Prescriptivism and Utilitarianism,” Analysis 43, no. 1 (1983): 43–49, and R. M. Hare, “Some Reasoning about Preferences: A Response to Essays by Persson, Feldman, and Schueler,” Ethics 95, no. 1 (1984): 81–85.
7. Bernard Williams, Ethics and the Limits of Philosophy (Cambridge, MA: Harvard University Press, 1985), ch. 5 (“Styles of Ethical Theory,” 71–92, esp. 83–84 and 87–89 on the “World Agent”); and Williams, “The Structure of Hare’s Theory,” in Seanor and Fotion, Hare and Critics, 185–196 (the “World Agent model” at 186); and Hare’s reply, “Comments,” 287–293, esp. 287–288.
8. Xuechunzi Bai, Angelina Wang, Ilia Sucholutsky, and Thomas L. Griffiths, “Explicitly Unbiased Large Language Models Still Form Biased Associations,” Proceedings of the National Academy of Sciences 122, no. 8 (2025): e2416228122, https://doi.org/10.1073/pnas.2416228122.