Behind the Veil of Ignorance: How Rawls Reinvented Justice – and Why His Thought Experiment Still Echoes in the Ethics of Artificial Intelligence

The Hook: Whoever Cuts the Cake Takes the Last Slice

Picture an ancient rule with which parents have settled kitchen-table disputes for generations: whoever cuts the cake into two pieces gets to take the other piece. The one who cuts must watch the other choose first. The result is astonishingly reliable – the cut becomes fair to the millimeter, because the person cutting does not know which slice will end up being theirs. It is not generosity that produces the fairness, but calculated uncertainty about one's own position.

This everyday wisdom is exactly what the American philosopher John Rawls elevated, in 1971, to the level of an entire theory of society. In his major work A Theory of Justice, he posed a question that sounds as simple as it is consequential: by what principles would we arrange the basic framework of our society – its constitution, economic order, and distribution of rights and goods – if we did not know who we would be in that society? Not rich or poor, not healthy or sick, not gifted or disadvantaged, not man or woman, not believer or atheist. Rawls called this hypothetical condition the original position, and the shroud that hides our future role from us the veil of ignorance.

Rawls's book struck a nerve in 1971. Political philosophy was at the time widely regarded as more or less dead, crushed by logical positivism, which treated moral statements as mere expressions of feeling. A Theory of Justice revived the field overnight; it became the most discussed work of twentieth-century political philosophy and was translated into more than two dozen languages. Rawls delivered not only a principle but a method: a thought experiment by which one can test whether a social rule is fair – namely, if we would choose it without knowing our own advantage.

This article takes you on the full journey: from the construction of the original position through the famous two principles of justice and the counterintuitive maximin argument to the elegant counter-position of John Harsanyi, who derived utilitarianism from almost the same thought experiment. We will see how Robert Nozick, Amartya Sen, and Michael Sandel attacked the theory from three entirely different directions, what happens when you actually rebuild the veil in the laboratory – and why, of all people, the engineers of fair AI systems are today pulling Rawls's maximin rule back out of the drawer.

Part 1: Constructing the Original Position

Justice as Fairness

Rawls's basic intuition carries a memorable name: justice as fairness. This does not mean an equation – justice is not the same thing as fairness – but a relation of derivation: just principles are those on which free and equal persons would agree under fair conditions. The entire weight of the theory thus rests on the question of what these fair starting conditions look like. Rawls's answer is the original position.

In this Rawls deliberately placed himself in the tradition of contract theory – the idea that political legitimacy arises from a (hypothetical) agreement among the governed. But he lifted this tradition to a new level of abstraction. In Hobbes, Locke, and Rousseau, the "state of nature" was a quasi-historical starting point out of which people, with their real interests and power relations, concluded a contract. Rawls's original position, by contrast, is a purely hypothetical, ahistorical construct: a thinking tool that never took place and never will, whose sole purpose is to factor out the distorting influences of luck, power, and self-interest from the choice of principles.

What the Veil Conceals – and What It Does Not

The core of the thought experiment is the precise calibration of the ignorance. Behind the veil of ignorance, the parties (Rawls deliberately speaks neutrally of "parties," not "persons") who negotiate the principles do not know:

• their position in society – class, wealth, social status;
• their natural endowments – intelligence, strength, talents, health;
• their conception of the good – that is, their religious, moral, and philosophical convictions about what makes for a successful life;
• their psychological dispositions – for instance, whether they are risk-seeking or risk-averse;
• the generation to which they belong and the concrete stage of development of their society.

At the same time, the parties do know general facts about the world: the basics of economics, psychology, and sociology, the fact of scarcity, that human beings have a conception of the good at all and depend on social cooperation. They know that they have interests, only not which ones. This design is ingenious: the veil is thick enough to prevent any partisan calculation ("I vote for low inheritance taxes because I will inherit a fortune"), yet thin enough to make an informed, rational choice possible at all.

The deeper justification for this device is moral, not merely technical. Rawls was convinced that the distribution of natural endowments and one's social starting position are morally arbitrary: no one deserves to be born with a high IQ, into a wealthy family, or with robust health. It is pure luck of the birth lottery. A just social order, Rawls held, must not base its basic structure on such undeserved contingencies. The veil operationalizes precisely this insight: it excludes from the choice of principles all information that is morally arbitrary.

Part 2: The Two Principles of Justice

Rawls claims that rational parties in the original position would not choose just any principle but a particular pair of principles. In the mature formulation of his late work Justice as Fairness: A Restatement (2001), they read as follows:

First principle (the liberty principle): Each person has the same indefeasible claim to a fully adequate scheme of equal basic liberties, which scheme is compatible with the same scheme of liberties for all.

Second principle: Social and economic inequalities must satisfy two conditions: first, they must be attached to offices and positions open to all under conditions of fair equality of opportunity; and second, they must be to the greatest benefit of the least-advantaged members of society – this is the famous difference principle.

The Lexical Ordering

Crucial – and often overlooked – is the ranking of the principles. Rawls places them in lexical priority (lexical, in analogy to the ordering in a dictionary): the first principle has absolute precedence over the second, and within the second, fair equality of opportunity takes precedence over the difference principle. Concretely, this means: basic liberties may never be traded off against economic advantages. One may not deprive anyone of the right to vote, of freedom of speech, or of freedom of religion, even if doing so would raise the material welfare of the poorest. Only once equal basic liberties are secured for all may one negotiate the distribution of opportunities; and only once opportunities are fair may one discuss the distribution of income and wealth.

This priority is Rawls's sharp demarcation from utilitarianism, which permits precisely such trade-offs (more on this shortly). For Rawls, a society that sacrifices the liberty of a minority in order to raise total welfare is not merely "somewhat unjust on balance" but categorically unjust – because it fails to respect the inviolability of each individual person. The most famous sentence of the book puts it precisely: "Each person possesses an inviolability founded on justice that even the welfare of society as a whole cannot override."

The Difference Principle in Detail

The difference principle deserves closer scrutiny, for it is Rawls's most original and most contested contribution. It is expressly not a principle of equality. Rawls does not demand strict material equality. He permits inequalities – but only under a strict condition: they must improve the situation of the worst-off group relative to any alternative arrangement, including perfect equality.

An example clarifies the logic. Suppose higher salaries for doctors and engineers create incentives that raise innovation and productivity so much that, in the end, even the poorest – through better medical care, more tax revenue, more jobs – are better off than in a society with equal incomes. Then this inequality is justified under the difference principle. But if one raises the salaries still further, until the poorest end up worse off again (say, because the inequality destroys social cohesion), the additional inequality is no longer justified. The difference principle turns the question of distribution into an optimization problem with a clear objective function: maximize the position of the worst-off.

The following table summarizes the structure:

Part 3: The Maximin Argument

But why should the parties in the original position choose precisely these principles – and not, say, utilitarianism, which maximizes the total sum of well-being? Rawls's answer draws on a decision rule from game theory: the maximin rule (maximin, from maximum minimorum – "the greatest of the smallest").

The maximin rule says: when choosing under uncertainty among several options, consider the worst possible outcome of each option and choose the option whose worst outcome is the best. Instead of eyeing the largest possible gain, you insure yourself against the greatest possible catastrophe. Applied to the original position: since, behind the veil, every party must reckon with ending up in the worst position in society, it will choose those principles that make this worst position as bearable as possible. That leads straight to the difference principle, which by definition maximizes the situation of the worst-off.

Rawls was clear that maximin is not a generally rational decision rule. No one decides that way in everyday life – someone who never flies because a crash would be the worst possible outcome is not acting wisely but paralyzingly risk-averse. Rawls therefore argued that maximin is the rational choice only under three special conditions, and that precisely these conditions obtain in the original position:

1. No known probabilities. Behind the veil, the parties have no basis whatsoever for estimating probabilities for their future position. They cannot say, "I will end up in the middle class with 90 percent probability." Without probabilities, the classical expected-utility calculation collapses.
2. A guaranteed minimum is good enough. The worst position secured by the Rawlsian principles (full basic liberties plus the materially best possible minimum position) is so acceptable that the parties have little incentive to gamble it away for the chance of more.
3. The worst outcomes of the alternatives are unbearable. Utilitarianism, for instance, could in principle justify the enslavement of a minority if this raised the total sum of happiness. No one who might end up being that minority wants to run such a risk.

The decisive point is that Rawls reformulates the question of justice as a problem of rational choice under uncertainty – the same class of problems that underlies formal decision theory and whose psychological distortions were laid bare by prospect theory. And it is exactly here that his sharpest rival takes aim.

Part 4: Harsanyi's Utilitarian Counter-Position

The most fascinating twist in the story is that Rawls did not actually invent the thought experiment of the veil of ignorance – at least not alone. The Hungarian-American economist and later Nobel laureate John Harsanyi had, as early as 1953 – eighteen years before Rawls's book – proposed a structurally almost identical construction. And he arrived at the opposite result.

Harsanyi's starting point was his idea of impersonality: an ethical decision is impartial exactly when the decision-maker has lost all knowledge of their own identity and must assume that they are, with equal probability, any given person in the society. Here lies the subtle but momentous difference from Rawls: Harsanyi equips his parties with the principle of insufficient reason. If I have no information about my future position, Harsanyi holds, then I should assign every position the same probability – with n people, then, the probability of being each individual is 1/n.

From this assumption a different result follows necessarily. Someone who could with equal probability be any person maximizes their expected utility by maximizing the average welfare of society – for the expected utility of being a randomly drawn person is precisely the average utility. With this, Harsanyi had derived average utilitarianism from the original position. In his famous 1955 essay he cast this into a formal theorem: if the individual preferences obey the axioms of von Neumann–Morgenstern expected utility theory, and if in addition the weak Pareto principle holds (if everyone prefers a to b, so does society), then the social welfare function must be a weighted sum of the individual utility functions. This aggregation theorem is regarded to this day as one of the most elegant formal justifications of utilitarianism.

The conflict between the two can thus be captured precisely as a dispute over the correct decision rule under uncertainty:

Harsanyi's charge against Rawls was sharp and formulated in the language of decision theory: the maximin rule imputes to the parties an irrationally high risk aversion. Someone who orients strictly toward the worst possible outcome ignores all chances, however small their probability. Harsanyi illustrated this with a drastic example: under maximin logic, one could never accept a well-paid job in another city if there were even the tiniest possibility of dying in a fatal accident along the way – for death would be the worst possible outcome. No reasonable person, Harsanyi argued, decides that way.

Rawls's rejoinder was that the choice of a social order is precisely not an ordinary decision. It is a one-time, irreversible choice of existential magnitude, in which one has no second chance and in which the worst outcomes (slavery, persecution, misery) are so grave that a special caution is warranted. Someone deciding on the basic structure of their entire life prospects, Rawls held, should not gamble with their own fate. I am of the opinion that this dispute is ultimately not decidable on purely logical grounds: both positions are internally consistent, and which one finds more convincing depends on how one interprets the character of the choice in the original position – as a cool statistical bet (Harsanyi) or as a fateful foundational decision in which one insures oneself against the worst (Rawls).

Part 5: The Three Great Critiques

No work of political philosophy has provoked so much dissent as A Theory of Justice – and the dissent came from opposite directions, which is often the sign of a balanced middle position. Three critiques stand out.

Nozick and the Libertarian Right

Rawls's Harvard colleague Robert Nozick delivered the most influential libertarian rejoinder in 1974 in Anarchy, State, and Utopia. His central objection is directed against the difference principle, which he brands a patterned or end-state principle. Rawls, Nozick argues, treats social wealth like an ownerless cake to be distributed according to a just pattern. But goods do not fall from the sky; they are produced by people through labor, exchange, and voluntary agreement, and they have rightful owners from the outset. Justice, according to Nozick's "entitlement theory," is a matter of the history of a distribution (Was it rightfully acquired? Was it voluntarily transferred?), not of its pattern.

Nozick's most famous argument is the Wilt Chamberlain example: suppose a society starts from a distribution that is perfectly just by Rawls's standards. Now millions of basketball fans voluntarily pay 25 cents each to see the star Wilt Chamberlain play. In the end, Chamberlain has a fortune, the distribution has become massively unequal – and yet every single step to get there came about through the voluntary decisions of free people. To maintain Rawls's just pattern, the state would have to continually undo these voluntary transactions and redistribute. "Liberty upsets patterns," Nozick summarizes: freedom destroys patterns. Anyone who wants to enforce a particular distributive pattern must curtail freedom permanently.

Sen and the Capability Approach

From an entirely different direction came the critique of the Indian economist and Nobel laureate Amartya Sen. Sen accepts Rawls's egalitarian basic thrust but holds his metric to be wrongly chosen. Rawls measures people's situation by their endowment with primary goods – rights, liberties, income, wealth, the social bases of self-respect. Sen objects: people differ systematically in their ability to convert goods into actual well-being. A person with a mobility impairment needs, at equal income, far more resources to achieve the same mobility as an able-bodied person; a pregnant woman has higher nutritional needs. Two persons with an identical bundle of primary goods can thus have highly unequal real life chances.

Sen therefore proposes making not the means (goods) but the capabilities the metric: what a person is actually able to do and to be – to be nourished, to be healthy, to take part in community life, to be educated. The capability approach, which Sen developed together with the philosopher Martha Nussbaum, is today the theoretical foundation of, for example, the United Nations' Human Development Index. It shifts the focus from the question "What does someone possess?" to the question "What can someone really do with what they possess?"

Sandel and the Communitarian Critique

The third major critique aims not at the result but at the conception of the person behind the veil. Michael Sandel charged Rawls in 1982, in Liberalism and the Limits of Justice, with presupposing an unrealistic, "unencumbered self": an individual that precedes its ends, attachments, and community memberships and can regard them like interchangeable garments. But real, existing people, the communitarian counter-thesis holds, are constitutively shaped by their families, traditions, religions, and communities. One cannot meaningfully reflect on justice by abstracting from everything that makes a person this particular person at all. The veil of ignorance, Sandel argues, conceals not merely contingent attributes but cuts away moral identity itself.

These three critiques mark the field of tension in which Rawls's theory stands: Nozick holds it to be too redistributive, Sen to be not accurate enough in its metric, Sandel to be built on a false conception of the person. That the theory is attacked from the right as too egalitarian and from the left (or from the communitarian perspective) as too individualistic speaks rather for than against its carefully balanced middle position.

Part 6: The Veil in the Laboratory and in the Machine

When You Actually Rebuild the Original Position

Rawls's original position is a thought experiment – but what if you test it empirically? That is exactly what the political scientists Norman Frohlich and Joe Oppenheimer did in a series of experiments they summarized in 1992 in the book Choosing Justice: An Experimental Approach to Ethical Theory. They placed experimental groups in Canada, the United States, and Poland in an approximation of the original position: the participants were to agree on a principle of income distribution without knowing which income position they would subsequently be assigned by lot. Among the options were pure average utilitarianism (maximize average income), the Rawlsian difference principle (maximize the minimum), and hybrid forms.

The result was instructive for both theories – and a full victory for neither. Two of Rawls's predictions were splendidly confirmed: the groups almost always reached unanimous agreement, and they did so under conditions of uncertainty quickly and stably. Rawls's substantive prediction, however – that people would choose the pure maximin difference principle – clearly failed. The overwhelming majority of groups instead opted for a compromise: maximizing average income subject to a guaranteed floor (floor constraint) – they wanted to make the cake as large as possible but to ensure that no one falls below a social subsistence minimum. Neither pure utilitarianism (which neglects the floor) nor the pure difference principle (which counts only the minimum) matched the intuition of the subjects. They chose a third option, which Rawls himself had dismissed as "intuitionistic," because it weighs two principles against each other without a strict lexical ordering.

One should interpret these results with caution – laboratory conditions with small sums of money are not the existential seriousness of the real original position, and their transferability to real social orders is contested. Yet as an empirical commentary on a normative theory they are remarkable: people behind a real (if thin) veil behaved neither as pure Harsanyi utilitarians nor as pure Rawls maximizers, but as pragmatic hedgers who aim for the average but demand a safety net. The same maturing movement – from elegant theory through empirical test to a nuanced intermediate finding – we know in this vault from, for instance, the marshmallow test and the Dunning-Kruger effect.

Rawls's Renaissance in Artificial Intelligence

One might think a thought experiment from the political philosophy of 1971 would be of no relevance to engineers. The opposite is the case. Of all places, it is in the young discipline of fair machine learning that Rawls's maximin principle has become an actively used tool. When an algorithm decides on credit approval, medical prioritization, or the allocation of scarce resources, the question arises immediately: by what fairness criterion should it optimize?

The obvious answer – maximize average accuracy or total utility – is precisely the utilitarian solution, and it has precisely Rawls's problem: a system that performs excellently on average can fail catastrophically for a small, disadvantaged subgroup. Researchers therefore increasingly formulate fairness as a Rawlsian maximin objective: optimize not the average but the worst performance across all groups (in the technical jargon, "max-min fairness" or "minimax group risk"). The veil of ignorance here becomes a concrete design heuristic: build the system as if you did not know which group the individual user belongs to – then you have an interest in its functioning acceptably for every group. Rawls's fifty-year-old thought experiment thus supplies a surprisingly precise objective function for the ethics of automation.

The connection to distributive ethics reaches beyond AI into the entire debate about future generations, as raised by Parfit's repugnant conclusion, and into the question of how much our moral judgment depends on chance at all – the theme of moral luck.

The Central Takeaway

If you want to take a single practical lesson from Rawls's work, let it be this: to test whether a rule is fair, do not ask whether it benefits you, but whether you would choose it if you did not know what role you would play under it.

This figure of thought is a universally applicable fairness test, far beyond political philosophy. Want to know whether a corporate policy is just? Imagine it without knowing whether you are a board member or an intern. Want to settle a dispute in the family? Design the rule before you know who gets the last slice of cake. Want to design a system – whether a tax code or a scoring algorithm? Examine its worst outcomes as if you yourself could be affected by them. The veil of ignorance is at its core an instrument against perhaps the deepest root of injustice: the tendency to bend rules toward one's own, already-known advantage.

And beyond everyday life, Rawls teaches us something about the nature of justice itself. For him it is not a property that one reads off a distribution from the outside, but the result of a procedure: fair is what emerges from fair conditions. This shift from the question "Which outcome is just?" to the question "Which procedure is fair?" is Rawls's lasting legacy – and the reason his veil reaches all the way into the machine rooms of the present.

A Question to Reflect On

Rawls claims that we recognize just principles best when we do not know our own position. Yet the real dispute between Rawls and Harsanyi shows that even behind the veil a prior decision is required: whether to insure against the worst (maximin) or to bet on the greatest average (expected utility). If you yourself stood behind the veil and had to decide on the society in which your children will grow up – would you choose the order that makes the lot of the poorest as good as possible, or the one that maximizes the average, even if this carries a small risk of great misery? And what does your answer reveal about your relationship to risk, responsibility, and solidarity?

Cross-References in the Vault

• The Two Boxes: Newcomb's Paradox and the Battle Over Rational Choice – formal decision theory under uncertainty, whose rules (expected utility vs. alternatives) form the core of the Rawls–Harsanyi dispute.
• Losses Loom Larger: How Prospect Theory Dethroned Rational Man – and Why Its Most Famous Principle Is Now on Trial Itself – how real people actually decide under risk; an empirical commentary on the risk aversion that Rawls's maximin imputes.
• The Billions Barely Alive: Parfit's Repugnant Conclusion and the Ethics of Future Generations – distributive ethics writ large: sum vs. minimum, average vs. number, closely related to Rawls vs. utilitarianism.
• The Child in the Road: Moral Luck and the Limits of Responsibility – Rawls's argument from the moral arbitrariness of natural endowments answers the same question: how much chance may determine our fate?
• The Switch in the Machine: The Trolley Problem and the Ethics of Autonomous Vehicles – the opposition between utilitarian trade-offs and deontological inviolability that also shapes Rawls's lexical priority of liberty.
• Meant or Merely Foreseen: The Doctrine of Double Effect and the Power of Intention – another building block of an ethics beyond pure consequentialism.

Sources

1. Rawls, J. (1971/1999). A Theory of Justice. Harvard University Press. Overview and intellectual biography: John Rawls (Stanford Encyclopedia of Philosophy)
2. Original Position – comprehensive treatment of the original position, the veil, and the maximin argument: Stanford Encyclopedia of Philosophy; on the difference principle in detail: The Argument for the Difference Principle (SEP)
3. Rawls, J. (2001). Justice as Fairness: A Restatement. Harvard University Press. Formulation of the two principles and the lexical ordering: Justice as Fairness (Wikipedia, with primary quotations)
4. Harsanyi, J. C. (1953/1955). Cardinal Utility in Welfare Economics and in the Theory of Risk-Taking / Cardinal Welfare, Individualistic Ethics, and Interpersonal Comparisons of Utility. Journal of Political Economy. On the aggregation theorem and the debate with Rawls: Hammond, "Harsanyi's Utilitarian Theorem: A Simpler Proof and Some Ethical Connotations" (Stanford, PDF); overview: "Rawls, Harsanyi and the Veil of Ignorance" (LSE Rerum Causae, PDF)
5. Nozick, R. (1974). Anarchy, State, and Utopia. Basic Books. (Wilt Chamberlain argument, entitlement theory.) Context in: John Rawls (SEP), section on critiques
6. Sen, A. (1979/1980). Equality of What? (Tanner Lecture); Nussbaum & Sen (eds.), The Quality of Life (1993). On the capability approach: The Capability Approach (Stanford Encyclopedia of Philosophy); Sen's Capability Approach (Internet Encyclopedia of Philosophy)
7. Frohlich, N., & Oppenheimer, J. A. (1992). Choosing Justice: An Experimental Approach to Ethical Theory. University of California Press. Key finding (floor constraint rather than the pure difference principle): "Choices of Principles of Distributive Justice in Experimental Groups", AJPS (PDF)
8. Sandel, M. (1982). Liberalism and the Limits of Justice. Cambridge University Press. (Communitarian critique of the "unencumbered self.") Placement: John Rawls (SEP)
9. On the modern application of the maximin principle in fair AI: "Algorithms and Analysis for Optimizing Robust Objectives in Fair Machine Learning" (arXiv, PDF); "Measuring justice in machine learning" (arXiv, PDF)

← All articles