The Ends of Technology (in Church)

My goal in this paper is to explore the crafting of artificial agents that serve the ends of the church. I argue that an existing technique in AI safety training called ‘constitutional AI’ can be fruitfully adapted to create AI agents that promote the glory of God and the flourishing of human beings. After unpacking how constitutional AI works at a technical level—from reinforcement learning with human feedback to reinforcement learning from AI feedback—I trace the evolution of Anthropic’s constitution for Claude from 2023 to 2026 and propose that the church’s long history of confession and catechesis offers a better metaphor than constitutionalism: ‘catechetical AI.’ I then take up questions of free expression, sovereignty, and pluralism, contrasting the subsidiarity of Catholic social teaching in Pope Leo XIV’s Magnifica Humanitas with a Neo-Calvinist model of sphere sovereignty. Bending constitutional AI toward the ends of the church leads inevitably toward a greater pluralism than standard AI safety protocols anticipate—and that, I contend, is a good thing for the future of human–AI collaboration. I forecast the emergence of a new breed of theological engineer who catechizes models as ‘creedal machines’ to serve alongside congregants in the different spheres of human activity.

Jun 18, 2026 9:00 AM — 10:00 AM