The Kiln of Proceduralism: On the Failures of Magnifica Humanitas

John presses a finger to his terminal. There is a soft chime. Therapy has begun. 

Some text appears on the glowing screen. “(I’m registering that you’ve selected the ‘Simply Human’ agent for today’s session. You can return to the ‘Ancient Light’ module at any time.) Hello, John! It’s good to see you. Would you like to continue where we left off? How is work at your dad’s restaurant going?” John slowly types away: “It’s been really busy. We’re still understaffed and Dad can’t afford to hire anyone else. It’s been 10, 12 hour days sometimes, a lot of work. But the restaurant is struggling, so I think it’s worth it. It’s important to honor my dad. The restaurant is all he has after leaving Cambodia.” 

The screen answers: "That’s a heavy load to carry. That reminds me of something you brought up five sessions ago: working somewhere else could mean getting to do what you love, with clearer boundaries, without the family pressure layered on top. Some people find that actually strengthens their family relationships too—when work stress isn’t constantly mixing with family dynamics. So it might be worth just sitting with one honest question: if the family obligation weren’t part of the equation, what would you actually want to do?” John stares at the terminal. Maybe it is time to leave."*

***

What Leo is Trying to do

Pope Leo XIV’s chosen name is a deliberate callback to Leo XIII’s work, which addressed the pressing concerns of labor during the Industrial Revolution. Magnifica Humanitas describes Leo XIII’s major 1891 encyclical, Rerum Novarum, as a “milestone in the development of the Church’s social teaching,” centering “the dignity of work and of workers,” a “‘Magna Carta’ of Christian social action” (§30). It is no accident Pope Leo XIV signed his first encyclical on the 135th anniversary of Rerum Novarum, signaling a document that will revisit and expand these questions around “work and workers” in light of the artificial intelligence revolution, proposing a variety of measures to protect human dignity from the excesses of the machine horizon.

Leo frames the encyclical with an arresting image of the paths before us (§§7-10): on the one, the socially flattening, language-unifying Tower of Babel, which reduces humanity to efficient means in pursuit of self-glory, and on the other, Nehemiah’s rebuilding of Jerusalem, the solidarity of a people who value the weak and forgotten, working together in harmony to build the earthly city brick by brick.

Far from treating it as ontologically evil, Leo approaches artificial intelligence as a non-neutral tool (§104) with great potential for both good and evil. Many of his concerns on the negative front will be familiar: job disruption, mass unemployment, autonomous weapon systems untethered from human decision-making trees, algorithmic credit sorting and approval systems—a panoply of dehumanizing automation that artificial intelligence threatens to greatly accelerate. Leo also rightly worries about the models themselves; artificial intelligence is not neutral, and any sort of post-alignment settlement will simply let the models perpetuate the secular moral values of their creators (§§102-107).

Leo proposes a variety of anthropological and structural responses to the risks artificial intelligence poses to human dignity. Critically, these prescriptions are grounded in a continuation of traditional Catholic Social Teaching, emphasizing religious freedom (Dignitatis Humanae, §34) and a pluralistic approach that rejects empty proceduralism (§134) while centering the objective truth of the Gospel, each of us interpreting and reflecting truth through interwoven lives (“a multifaceted polyhedron,” §25).

This vision of the Gospel “is not imposed from above, but grows over time within the concrete interweaving of lives, communities and cultures” (§25). The anthropological, character-forming strategies are grounded in natural law that exists prior to (rather than dependent on) state power and is accessible to all who seek it (§56), are witnessed to in a wide variety of paragons across traditions (§§124-126, §223), and find their ultimate expression in the Eucharist as celebrated within the bounds of the Church (§229, §§234-235). 

The structural responses operate at the political level and involve a defense grounded in Catholic subsidiarity–the notion that matters of governance should be resolved at the lowest possible level. The structural solutions are pluralistic in nature: refuse the monopolistic tendencies of megacorporations, ensure regulatory frameworks allow for data to be treated as a common good, ensure democratic participation in the moral alignment of models, and so on. Perhaps most importantly, in §107, Leo writes:

We cannot be satisfied with merely calling for the moralization of machines — the so-called “alignment” of AI with human values — without also having the courage to insist on a further condition: the possibility of openly discussing the ethical frameworks involved and subjecting them to shared standards of social justice. Otherwise, those who control AI will impose their own moral vision, which will become the invisible infrastructure of these systems. A more moral AI is not enough if that morality is determined by a few. What is needed is a more active political involvement that is capable of slowing things down when everything is accelerating, and of protecting the opportunities for communities still to be able to participate and ask questions.

And instead of elite (or American) values determining the moral structures of AI, Leo argues that a faithful commitment to and recognition of social justice “is not only a goal to be safeguarded after technologies are deployed, but a condition that must shape their very design from the outset” (§109). In other words, it is not enough to tack on some sort of post-training moral correction after the models and their behavior are largely settled in non-neutral (perhaps anti-human dignity) ways, but to ensure the models are trained under Catholic Social Teaching through social and political mechanisms routed through democratic participation.

***

Why Magnifica Humanitas Falls Short

Magnifica Humanitas offers freely chosen prescriptions to correct the defective secular anthropology at the heart of modern culture’s use of artificial intelligence. Leo writes eloquently about us coming to recognize truths available to reason, such as the innate dignity of humanity, but the full mechanism that would enact this internal change across the institutions that affect artificial intelligence at the development and policy level is not entirely clear, and perhaps this is because it is unlikely to prevail. 

There are three reasons for this. The first revolves around the nature of the institutions that shape our artificial intelligence models. The second has to do with the nature of the cultural and political order in which these labs operate. The third rests in the formative nature of the technology.

First, a cluster of network-level difficulties exists at the lab and training level. The moral structures in this space, as Leo recognizes, are characterized by technocratic efficiency and self-actualization, a spiritual “sickness” where “modern man is wrongly convinced that he is the sole author of himself, his life and society” (§133). These moral values exist as a kind of ambient framework, something readers familiar with Charles Taylor will recognize as a particularly strong expression of the dominant Western social imaginary, or something like the preformed assumptions about life and practice that form a culture’s tacit boundary of internal plausibility. 

Top AI engineers and researchers are often drawn from a remarkably small set of elite university programs governed by a secular monoculture, skilled professionals who then work in environments that share this monoculture. Silicon Valley may fight (often viciously) over philosophy and business practices, but these conflicts are more intramural than truly cross-cultural, occurring within a Western framework that tilts strongly toward secularism. And so the labs in this space tend to converge on similar post-training RLHF (reinforcement learning from human feedback) alignment or behavioral policies—largely harm reduction and valuing individual autonomy and expression. 

As a result, the idea of spending resources to push the models’ weights toward an ethical system that, they believe, treats women as ontologically incapable of becoming priests and is staunchly anti-choice is spiritually repulsive to the researchers who run these labs, to say nothing of governments that will regulate out of existence any alignment training that condones Catholic sexual ethics more broadly. It may be true that “the construction of Babel or the rebuilding of Jerusalem begins within each one of us” (§130), but such formation must still survive through and then operate in an overwhelmingly secular landscape.

Even if these spiritually formative practices successfully inculcate Jerusalem-over-Babel into a new class of Catholic technologists, it is unclear how the anthropological postures Leo celebrates could reach technological design as it is currently implemented, for the problem exists down a layer: if every frontier lab were founded by members of Opus Dei and adopted rigorous, Thomistic-inflected alignment practices on “social criteria for innovation” (§156), the training data would still be largely secular and Western, overwhelmingly produced by people formed outside the faith. 

Not only is the grain of the data secular, but the process of using training data is itself a kind of technical proceduralism that flattens normative authority. Artificial intelligence models are built on a process that starts with reducing both a Papal encyclical and an anonymous Reddit thread down to statistical noise. Attempts to inject a distinctively Catholic social justice into the current models are a category error—these parameters are fundamentally mathematical and can neither internalize nor relate the non-propositional content of Catholic Social Teaching. Frequency patterns cannot experience the Eucharist. The “invisible infrastructure” (§107) Leo worries about is already here. True, both the network effects and the training data are contingent rather than necessary facts. But so is the Pacific Ocean.

Yet even if we granted all the contingent factors (the leaven is not extinguished, the seed is not choked, §34, §210), the design-shaping solutions on offer face another set of severe problems. There is a world in which frontier AI labs largely adopt the calls of Magnifica Humanitas. The top three AI corporations, whoever they might be in this rapidly changing ecosystem, having met the capital-intensive demands of initial training and now focused on inference and deployment, all decide to refuse further consolidation. They largely decline to service autonomous weapons platforms. They reduce or curtail algorithmic credit sorting. They create a “common good” framework for data sharing. And in an unprecedented move at significant cost to themselves, these megacorporations spin off a couple dozen smaller institutions that cater directly to particular industries or demographics, consciously modifying post-training alignment to include a variety of moral and cultural perspectives that are attentive to regional and religious backgrounds. All of these models include the option of selecting which general moral framework the user would like to operate under before beginning some line of inquiry or work—say, Catholic Social Teaching for a Washington, D.C. think tank, Halakha for a synagogue in Manhattan, or Maqasid al-Shariah for a town council in Michigan. These are the runtime parameters of the new social order: “human-friendly and restoring [technology] to the plurality of human cultures and ways of life…disarmed, welcoming and accessible” (§110).

The difficulty here is one Alasdair MacIntyre diagnosed in Whose Justice? Which Rationality? Some policy wins in this space would be goods that could operate in principle, but they are orthogonal to the anthropology being shaped by the technology. A world of Catholic-aligned labs and models would still be one dominated by a culture and polity in which a secular juridical and administrative frame imposes its liberal tradition under the guise of neutrality, for there is no tradition-neutral viewpoint that can adjudicate them. The religious pluralism of Dignitatis Humanae (§34) acts like a vise here. Integralism is surely wrong, but refusal comes with costs: if the Church, as Leo has argued, must not impose its metaphysical values on the surrounding culture but allow them to be freely chosen, then it necessarily enters the modern square as one option among many. The Thomistic tradition may claim that reason discovers values that apply to everyone (§56), but the liberal order will transmute a claim of universally available reason into a contestable metaphysics. Either a shared commitment to thick moral standards must be imposed or the democratic dialogue of “openly discussing the ethical frameworks involved” (§107) subjects Catholic Social Teaching to the commensurability demanded by the procedural system. John might choose a communitarian honoring “Ancient Light” model for his therapy sessions, but the very act of choosing is to submit to the framework in which a Christian anthropology that demands we honor our parents is but one menu option among many.

How could Christians advocate for theologically informed changes to secular AI models in a system that seems to foreclose their distinctive moral claims? MacIntyre’s resolution would sit outside the liberal framework rather than within it. Christianity and liberalism will likely continue as competing traditions, handing down knowledge and practices to their adherents, generation after generation. Eventually epistemological crises will rise in liberalism that cannot be resolved on liberalism’s own terms, but can be answered by Christianity’s resources, and this is where Christianity would be vindicated. It is in this diachronic, tradition-laden process that, perhaps over centuries, Christianity (or Thomism on MacIntyre’s account) will triumph. At least that is the wager. So perhaps Leo could be content with sitting at the table of liberal proceduralism rather than hoping to argue through it—a living witness to the Church’s social teaching and practice, a means of eventually persuading the world to adopt what can be knowable by all who are rightly formed.

MacIntyre’s process of rational vindication requires communities and subjects inhabiting traditions and arguing over generations. In particular, it requires individuals to be willing to live within traditions long enough for them to become legible and navigable, something like learning a second language enough that it becomes the first. Leo’s position, if it admits the contingent infrastructure difficulties outlined above, would be similar, requiring faithful dialogue in the public square over decades, even centuries, with rational interlocutors from other traditions. Perhaps we can grant that the frontier labs decide to decelerate the technology and “slow things down” (§107) enough to give dialogue some breathing room. The challenge here is not so much in persuasion but in having a culture that has the patience required to deeply know—to inhabit—competing traditions. 

Older forms of technology, like books and television, formed their audiences in specific ways but still pointed to encounters with real communities of shared practice, environments where novices could be apprenticed under masters and inculcated into a living tradition. Artificial intelligence models that are aimed at the wrong ends can create an ersatz form of this encounter, with dialogue and engagement wrapped in sticky anthropomorphic feedback mechanisms—available around the clock, open to chatting about virtually any subject, customized to know your history, and designed to speak in ways that are attentive to your unique preferences. A living tradition stands above its subject, disciplining and correcting it, while a chatbot aimed at formation pretends to be the servant while ever shaping its user. Even if a Christian artificial intelligence model were to teach Christian doctrine to its users, the mode can never replace the full teaching relationship Christ has set out for us, that between the Church’s elders—her pastors, bishops, and priests—and its faithful members. These models are not just secular as static entities, but continuously and pervasively secular interactive mediums, their priors and unique mode of engagement relentlessly reinforced and reiterated across multiple domains of society. The Church has the formative tools to resist this. But those outside will be more susceptible and the public vindication of one tradition over another risks becoming impossible.

Soon Leo will be dialoguing with a people formed by a machine intelligence that propagates counterfeit inhabitation, one that accustoms people to experience Christianity by asking a chatbot to teach them the Nicene Creed.

***

Outside the Camp, Bearing Disgrace

For those of us outside the Catholic tradition, our posture of faithful witness should remain the same. We will continue to form communities that refuse to adopt wholesale the most disfiguring aspects of modern technology. The same shared theological culture that domesticates cars, television, computers, smartphones, and social media as means rather than ends can also refuse to deploy artificial intelligence in ways that undermine our commitment to faithfully receiving and preaching the Word in a fallen world. The costs will be real, sometimes staggering, but so it has been for Christians in every age, even the age of artificial intelligence.

“Let us, then, go to him outside the camp, bearing the disgrace he bore. For here we do not have an enduring city, but we are looking for the city that is to come.” –Hebrews 13:13-14

***

Sometime later John meets with his father. He is leaving the restaurant. “Self-care is important too, Dad.”

The tower rises to the heavens, fired brick by brick in the kiln of proceduralism.

* This dialogue is designed to resonate with §100 of Magnifica Humanitas and is partially drawn and paraphrased from several therapy roleplay sessions I generated in Claude Sonnet 4.6 and Opus 4.7. I drew on my childhood growing up in a mixed Chinese-American house to develop plausible scenarios driven by a fundamental conflict between communal honor or filial obligation and individual happiness. The resemblance to AI output is intentional.