The cycle that began with [YOU] on AI documents the AI transition from a position that is honest about its own partisanship: the builder's position, the position of someone who has used the tools and found them transformative. Rawls's framework strips that position away. Behind the veil, you do not know whether you are the triumphalist or the elegist. You do not know whether you are the developer in Lagos for whom AI represents liberation or the senior architect in San Francisco for whom it represents the dissolution of a career built through decades of patient mastery. You do not know whether you are the engineer in Trivandrum building features she had never been trained to build or the colleague whose specialized expertise is now accessible to anyone with a subscription.
The difference principle evaluates the current distribution of AI's gains with austere clarity. The gains are concentrated in a remarkably narrow segment: shareholders of AI companies, early adopters in wealthy nations, the technically fluent. The costs are distributed to precisely those who can least absorb them: the administrative assistants and junior analysts whose entry-level knowledge work AI now performs; the communities organized around displaced industries; the students whose educational institutions have not adapted. The difference principle does not require equality. It requires that the inequality benefit the least advantaged as much as any alternative arrangement would. The available evidence is that the current arrangement fails this standard—that alternative institutional arrangements could better serve those at the bottom without suppressing the gains that make the expansion worth having.
Rawls's contribution to the cycle is methodological as well as substantive. The technology industry's dominant moral framework is individualistic: individual founders making individual choices, individual workers adapting to individual circumstances. Justice as fairness insists that this individualism is philosophically insufficient. Individual choices occur within the basic structure—the fundamental institutions that determine the range of choices available—and the basic structure is the primary subject of justice. A displaced worker's “choice” to retrain is not a free choice if the retraining infrastructure does not exist. The moral weight falls on institutions, not individuals.
The publicity condition applies with surgical precision to the opacity of algorithmic decision-making. A just society is one in which the principles governing institutions are publicly known, understood, and endorsed by the citizens subject to them. When hiring algorithms reject candidates who cannot see the principles governing the rejection, when content algorithms shape information environments whose selection logic is protected as competitive advantage, when alignment decisions embed value judgments about what is harmful without public deliberation—the publicity condition is violated. Not because the algorithms are necessarily wrong, but because citizens cannot evaluate what they cannot see, and cannot endorse principles they do not know.
John Rawls was born in Baltimore in 1921, educated at Princeton, and served in the Pacific during World War II before returning to graduate study in philosophy. He joined the Harvard faculty in 1962, where he remained for the rest of his career, publishing with unusual patience: A Theory of Justice appeared in 1971, distilling two decades of journal articles. Its reception was immediate and transformative. It revived social contract theory, which had been largely dormant since Kant. It provided the most rigorous contemporary framework for evaluating institutional arrangements. And it provoked a generation of responses—from utilitarians like Peter Singer, from libertarians like Robert Nozick, from communitarians like Michael Sandel, and from egalitarians like G.A. Cohen—that defined Anglo-American political philosophy for decades.
His second major work, Political Liberalism (1993), addressed the objection that the original position assumed more consensus about the good than a pluralist society could sustain. Overlapping consensus: citizens holding diverse comprehensive doctrines—religious, secular, liberal, conservative—could converge on the same principles of justice from different starting points, provided the principles were framed in terms of political rather than comprehensive values. The insight matters for AI governance: the principles for just AI institutions do not require resolving disagreements about the good life. They require only the impartiality of the veil.
Rawls died in 2002, two years after publishing The Law of Peoples, his extension of justice as fairness to international relations. He never wrote about artificial intelligence. His framework, like Keynes's, was designed at a level of generality that makes it applicable to any technological context—which is why the Google DeepMind researchers who tested the veil of ignorance as an experimental protocol in 2023 found that participants behind the veil, regardless of risk attitudes or political preferences, consistently chose principles that prioritized the worst-off. Rawls predicted this result in 1971.
The Veil of Ignorance. The most rigorous method for designing just institutions is to choose them from behind a veil of ignorance about your own position. Behind the veil, you cannot design institutions that favor the rich, the talented, or any particular demographic, because you might be poor, ordinary, or subject to discrimination. The only rational strategy is maximin: maximize the minimum. Choose the arrangement under which the worst-off person is as well-off as possible.
The Difference Principle. Social and economic inequalities are permissible only if they are arranged to the greatest benefit of the least advantaged. The principle does not require equality. It requires that inequality serve a purpose: that the arrangement under which some have more benefits those who have less more than any alternative arrangement would. Applied to AI: the concentration of gains in a small number of companies is just only if it benefits the least advantaged more than any alternative distribution would. The available evidence suggests it does not.
The Basic Structure. Justice applies primarily to the fundamental institutions of society—the constitution, the legal system, the property regime, the tax code, the educational system—not to individual transactions. The basic structure of the AI age includes the platforms that mediate access to AI capabilities, the data governance frameworks, the labor arrangements, and the educational institutions. These are the appropriate subjects of justice. Individual virtue is insufficient; structural arrangements must produce just outcomes regardless of the virtue of the individuals operating within them.
The Publicity Condition. A just society requires that the principles governing its institutions be publicly known and publicly endorsed. The publicity condition is violated when algorithmic systems that determine life outcomes—hiring, credit, content, productivity measurement—operate on principles invisible to those subject to them. This is not a technical objection. It is a foundational objection to arrangements that require compliance without consent.
Justice as Procedure. Justice as fairness is procedural rather than substantive: the just arrangement is not determined by looking at outcomes but by ensuring that the procedure for choosing principles meets the conditions of fairness. This means the institutions governing the AI transition are unjust not primarily because their outcomes are bad—though they may be—but because they were designed by interested parties arguing from known positions, not by rational people behind any veil.