Justice as Fairness (Rawls)

In the AI Story

Justice as fairness positions itself against two major traditions in moral and political philosophy. Against utilitarianism, it refuses the aggregation of individual gains and losses into a net calculation, insisting on what Rawls called the separateness of persons — each person lives one life, bears one set of costs, cannot be compensated for her suffering by another's flourishing. Against intuitionism, it refuses the appeal to self-evident moral truths that different thinkers happen to perceive differently, insisting that principles of justice must be generated through a procedure whose fairness can be defended independently of the principles it produces.

The positive content of justice as fairness is the two principles that Rawls argued would emerge from the original position: the equal basic liberties principle and the second principle comprising fair equality of opportunity and the difference principle. These principles are ordered lexically — the first takes absolute priority over the second, and within the second, fair equality of opportunity takes priority over the difference principle. The lexical ordering means that no amount of economic gain can justify violations of basic liberty, and no benefit to the least advantaged can justify violations of fair equality of opportunity.

Applied to AI, justice as fairness reframes the dominant ethical conversation. The technology industry's ethics discourse focuses on internal properties of AI systems: bias, accuracy, transparency, representativeness of training data. These questions matter, but from Rawls's perspective they are secondary. The moral properties of an AI system are not internal to the model. They are products of the social system — the basic structure — within which the model operates. As Iason Gabriel argued in his 2022 paper, a language model is neither just nor unjust; the institutional arrangement within which the model is deployed is just or unjust.

The publicity condition sharpens this reframing. A just society, Rawls argued, is one in which the principles of justice are publicly known and publicly endorsed — citizens understand the principles governing their institutions and can see that the institutions actually operate according to them. The opacity of algorithmic decision-making, protected as intellectual property and defended as competitive advantage, violates this condition structurally. Even an algorithm that happened to produce just outputs would operate within an institutional framework that, by Rawls's standards, is not fully just — because citizens cannot evaluate what they cannot see.

Origin

Rawls introduced "justice as fairness" as the title of his 1958 essay in The Philosophical Review, where the core elements of the theory — the original position, the veil of ignorance, the two principles — first appeared in outline form. He developed the conception at book length in A Theory of Justice (1971) and returned to it in Justice as Fairness: A Restatement (2001), edited by Erin Kelly from Rawls's late lectures.

Across these works, Rawls refined his claim that justice as fairness is a "political, not metaphysical" conception — meaning that it can serve as the basis for an overlapping consensus among citizens who hold different comprehensive views of the good life. This political turn, developed in Political Liberalism (1993), was Rawls's response to the recognition that modern democratic societies cannot be expected to agree on the deep metaphysical questions from which moral frameworks traditionally derived their authority.

Key Ideas

Procedural justice. Justice is constructed through a fair procedure, not discovered as a pre-existing fact. The fairness of the procedure guarantees the justice of the principles that emerge.

Two principles in lexical order. Equal basic liberties take absolute priority over distributive concerns; fair equality of opportunity takes priority over the difference principle; the ordering is non-negotiable.

The basic structure as subject. The principles apply primarily to the basic structure of society — the constitutional, legal, economic, and educational institutions — not directly to individual actions or transactions.

Publicity as constitutive. Just institutions must operate transparently enough that citizens can understand, evaluate, and endorse the principles they embody; opacity is not merely inconvenient but corrosive of justice itself.

Political, not metaphysical. Justice as fairness claims to rest on political values that citizens holding diverse comprehensive doctrines can endorse, not on any particular metaphysical or religious view.

Debates & Critiques

The major critiques of justice as fairness include Michael Sandel's argument that it presupposes an unencumbered self stripped of the particulars that make moral reasoning meaningful; Robert Nozick's argument that it licenses redistribution that violates entitlements generated by just acquisition and just transfer; and Amartya Sen's argument that its focus on ideal institutional design distracts from the more tractable comparative task of reducing manifest injustice in actual societies. Each critique has reshaped the theory; none has replaced it as the dominant framework for liberal political philosophy.