Whose Justice? Which Rationality?

In the AI Story

The core argument proceeds through detailed historical analysis of the four traditions, showing how each develops its conception of justice from within its own framework of assumptions about human nature, the good, and the structure of rational inquiry. The Aristotelian tradition grounds justice in the cultivation of virtue through participation in the polis; the Augustinian in the relation of the will to divine order; the Humean in the natural sentiments and utility; the liberal in the protection of individual rights against collective interference. These conceptions are not merely different answers to the same question; they are different questions, formulated within different frameworks that cannot be translated without remainder into one another.

The implication for AI ethics is direct. Debates about whether AI is fair, whether it respects autonomy, whether it serves justice — these debates are interminable not because participants lack intelligence but because "fairness," "autonomy," and "justice" have different content in different traditions. The utilitarian specifies fairness in terms of welfare aggregation; the Kantian in terms of universalizability; the virtue ethicist in terms of distribution according to merit within a practice. The appearance of consensus on principlist AI ethics — the widespread endorsement of "fairness" as an AI value — conceals deep disagreement about what the principle requires.

MacIntyre's proposal is that the recovery of substantive moral argument requires explicit engagement with traditions rather than the pretense of transcending them. One tradition may turn out to handle the phenomena better than another — MacIntyre argues that the Aristotelian tradition, as developed through Aquinas, handles the phenomena of human life more adequately than its rivals — but this evaluation is itself conducted from within a tradition, with full acknowledgment of its tradition-dependence.

This has a consequence for the design of AI systems. The choice of which conception of fairness to encode, which standards of excellence to apply, which practices to support is itself a substantive moral choice that cannot be bypassed by appeal to tradition-independent principles. The pretense that AI systems can be designed according to "neutral" principles is a form of what MacIntyre calls the "liberal illusion" — the belief that one can evaluate without a tradition, while concealing the fact that liberalism is itself a tradition with particular commitments.

Origin

Published in 1988 by the University of Notre Dame Press. Based on the Gifford Lectures MacIntyre delivered at the University of Edinburgh in 1988. The book is the second in MacIntyre's trilogy on moral philosophy, followed by Three Rival Versions of Moral Enquiry (1990).

Key Ideas

Tradition-constituted rationality. All evaluative rationality has content only within a particular tradition.

Four rival traditions. Aristotelian, Augustinian, Scottish Enlightenment, and liberal — each with its own conception of justice and rationality.

Incommensurability. Rival traditions cannot be translated without remainder into one another; they ask different questions.

Tradition-evaluation is possible. Not all traditions are equally adequate; they can be compared by their capacity to handle the phenomena.

Liberal illusion. The pretense that one can evaluate from a tradition-independent standpoint — characteristic of modernity — is itself a tradition-constituted position.

Debates & Critiques

Whether MacIntyre's framework can accommodate rational progress between traditions, or whether tradition-dependence collapses into relativism. MacIntyre's own answer is that traditions can be compared by their resources for addressing the problems that challenge them — a position developed further in Three Rival Versions of Moral Enquiry.