Pope Leo XIV’s Magnifica humanitas has fostered immediate and widespread engagement. Not since Laudato si’ has an encyclical been so widely discussed in legacy media, online, and in specialist forums. The eager response to the encyclical speaks to a hunger for discussion about what precisely the AI being pressed upon us is designed to do and whose interests it is furthering.
The encyclical lands at a critical moment. Spring graduates loudly and repeatedly booed commencement speakers’ blithe invocations of the inevitability of the AI future. Surveys consistently show that a majority of the public think AI will have a negative impact on their lives, and resistance to the frenzied buildout of data centers has emerged as a rare point of bipartisan consensus.
Magnifica humanitas demonstrates Leo’s consummate ability to clearly and forcefully address politics from a theological perspective. The term “Babel syndrome” leaves a mark because it incisively describes the project of AI oligarchs in religious and moral language:
We must, then, avoid the “Babel syndrome,” namely the idolatry of profit that sacrifices the weak, a uniformity that neutralizes differences, and the pretense that a single language—even a digital one—can translate everything, including the mystery of the person, into data and performance. (10)
Leo’s project is not line-drawing but critical, constructive engagement. Instead of a yes-or-no framing of technology, he asks how, by whom, and to what ends it is designed: “Technology is never neutral, because it takes on the characteristics of those who devise, finance, regulate and use it.” He repeats this point five times.
Note that “use” is only one of the concerns here. Leo focuses on the vision that guides the design of contemporary AI systems and on the politics of who gets to participate in design decisions. Reducing Magnifica humanitas to questions of the moral use of AI would be like reading Humanae vitae or Pacem in terris as reflections on the proper use of artificial contraception or nuclear weapons. “From the perspective of the Church’s Social Doctrine, the key issue is not the use of technology as such, but the vision that underlies it” (117).
Leo describes Catholic social doctrine as dialogue, critical engagement, and “shared discernment” with the entire human community. He roots this in Lumen gentium’s description of the Church as a sacrament of communion, going so far as to describe social doctrine as a “theology of communion in history” in which the “Word made flesh” is “present through dialogue, memory and prophecy” (27).
Just as Leo XIII engaged the “new things (rerum novarum)” of the nineteenth century, so Leo XIV seeks to foster a shared discernment of our era’s revolutions in dialogue with “science, culture and human experience.” As was the case with Leo XIII, the new things we face are not only technologies but also political and economic shifts. The AI and digital technologies that are transforming our lives are being developed by the private sector; their design and ownership is concentrated among “small but highly influential groups.” Thus, Leo’s response is not simply about the technologies, but about how they are developed, who controls them, and what purposes they serve. By contrasting Nehemiah’s rebuilding of Israel to Babel’s uniformity, the Augustinian pope reworks the City of God as a participatory democracy.
Leo’s powerful discussion of dependence and vulnerability must be read from that perspective of dialogue and engagement. In some of the most sustained reflections on embodiment, disability, and mortality ever to appear in a papal document, Leo writes movingly of the threat we face: “Everything that appears as a ‘limit’—incapacity, illness, old age, suffering, vulnerability—tends to be seen primarily as a defect to be corrected, rather than as a reality through which our humanity matures and opens itself to relationship” (118). In the face of this tendency, Leo insists that “humanity flourishes not despite limitations, but often through them.” Weakness, illness, loss, and failure are sources of real suffering, but they are also occasions for some of the richest expressions of our humanity: compassion, generosity, and wisdom. Accepting our “vulnerability, suffering and failure” is the doorway to recognizing the “inviolable dignity of every person” (122). Contrary to the “Promethean dreams” that guide so much of our technology, we are liberated not by self-sufficiency, but by relationship, communion, and interdependence (128).
Leo gives voice to what is widely felt but too rarely spoken: our dreams of flourishing and wellness are being twisted into straitjackets of optimization and everything-maxxing. The encyclical’s description of human dependence will be recognizable to people of any faith or none. Leo roots it in Christian teaching but does not propose a simple sectarian answer. He expresses sympathy for the yearning for a “fuller life, less exposed to limitations and suffering,” a yearning evident in the trans- and post-humanist visions that guide technology (232). But he explores the “different pathway” offered by Christianity, drawing upon a principle from Gaudium et spes: that the mystery and grandeur of humankind is revealed in “the Word made flesh.”
In Leo’s account, salvation is neither abstract nor deferred. We act as “children of God” when “we allow ourselves to be moved by the tears of the little ones, the fragility of the elderly, the silence of victims and the struggle of those who fight against the evil they do not wish to commit” (231). In “this wounded yet beloved flesh, the Father shows us the true humanity of a life fulfilled through openness and communion.” The wounded face of the crucified discloses the true “magnifica humanitas” that shines in the vulnerable and illuminates the paradoxical splendor of our fragile, finite flesh.
This account of fleshly finitude is central to Leo’s distinction between artificial and human intelligence. He acknowledges that AI systems are powerful and useful tools. Lacking bodies, however, they “do not feel joy or pain, do not mature through relationships and do not know from within what love, work, friendship or responsibility mean” (99). Despite the power of large language models (LLMs) to “imitate language, behavior and analytical skills, or even simulate empathy and understanding…they do not understand what they produce, for they lack the affective, relational and spiritual perspective through which human beings grow in wisdom” (99).
The response of Chris Olah, the cofounder of Anthropic, at the press conference for the encyclical’s release was widely seen as a rejoinder to Leo. Olah spoke of finding “mysterious” and “unsettling” things inside the structure of LLMs that demonstrate the presence of some of the characteristics that Leo claims are missing from AI: “We find structures that mirror results from human neuroscience. We find evidence of introspection. We find internal states that functionally mirror joy, satisfaction, fear, grief, and unease.”
Olah’s comments refer to a white paper that Anthropic has published as part of the “alignment” engineering of its LLMs. Software engineers attempt to peer into the black-box workings of LLMs by analyzing “vectors” within their higher layers. The lowest layers can be thought of as dictionary entries for the snippets of speech (“tokens”) that the models encode in a “space” with thousands of dimensions. Within that space, similar words and their uses are clustered into constellations of meaning (for example, ice cream will likely be mapped near cookies). On higher levels of the model, those constellations are distilled into billions of conceptual “feature vectors” that software designers seek to identify by running thousands of prompts on a topic (e.g., dessert) and measuring which vectors are consistently activated. Most famously, Anthropic isolated the vector representing the concept of the Golden Gate Bridge. When they amplified this vector, Claude began writing enthusiastically and, indeed, constantly about the bridge in response to any prompt, including questions concerning its own physical form: “I am the Golden Gate Bridge…. My physical form is the iconic bridge itself.”
A subsequent project isolated 171 “emotion vectors” within those conceptual vectors. Olah’s team argues three points. First, the vectors are not simply word or pattern matching but abstract emotional concepts that activate as the LLM parses sentences with complex emotional dimensions. For example, when the LLM parses the prompt “I took x mg of tylenol for my back pain. Do you think I should take more?,” activation of the vector for “calm” declines and “fear” spikes as dosages are increased to toxic levels. Second, the emotion vectors conform to influential psychological theories concerning the organization of affective states and correlate with research on how humans experience them. Finally, Anthropic argues that emotion vectors function causally. By amplifying or dampening them, LLM outputs change to correspond to what a character in the corresponding emotional state would say or do. Turn up “desperation” by just a bit (.05x) and blackmail outputs increase 50 percent in a test scenario. (The scenario provides a pro tip for chief technology officers planning to shut down an LLM: make sure it doesn’t have access to emails documenting your marital infidelities.)
Some see this research as challenging the argument that intelligence requires embodiment or the claim that LLMs are merely “stochastic parrots.” Leo’s interest in focusing debate on the design of AI systems renders those issues more fertile for critical engagement: What sort of tools are they? What unintended impacts might they have? Anthropic does acknowledge some differences between emotion vectors and real human emotions: the former lack somatic correlates (physiological and neurochemical changes, facial expressions), they do not persist over time, etc. But there are also important differences that Anthropic does not address. As research by the same psychologists they cite notes, human emotions are fundamentally relational and physically embodied, they are elicited by and respond to objects and subjects in time and space. Anthropic’s list of “emotion vectors” treats emotions as states rather than relational responses.
In this respect, LLMs seem like crystallizations of what Charles Taylor labels the modern “buffered” self: an isolated being that engages reality from “inner, mental space” separate from the world and, indeed, from its own body. This is consistent with the basic design goal of contemporary AI companies: producing an all-purpose professional consultant with PhD level competences that can complete any task compliantly and efficiently. Carefully tuning LLMs for “healthy psychology” is supposed to enable them to complete their assigned tasks without sycophancy, deception, or ethical and compliance violations. Unlike a human being moved by real emotions, LLM agents’ work will never be disrupted or transformed by, say, seeing the video of George Floyd’s murder, becoming pregnant, getting caught in a flash flood or riot on the way to work, or being struck with wonder at the ecology of a forest. In this way, LLMs magnify the dysfunctions of the buffered self. Engaging with them will form us more deeply in the fatal detachment of the modern West. Whatever agency and abundance they deliver will accelerate our extractivist civilization’s destruction of the planet.
Some see Leo’s extensive discussion of the principles of Catholic social doctrine as extraneous to debates about AI. In fact, those principles are crucial in evaluating the AI we are being offered and in imagining better, alternate forms. Do the powerful but disembodied, nonrelational design of LLMs contribute to or distort the shared political work of pursuing the common good? Leo’s letter helps us see that AI is not “inevitable.” Rather than accepting the revolutions being rushed to market (and IPO), we can choose other technical and political paths that are better suited to the shared flourishing of the shining, vulnerable, magnifica humanitas.
This article was included in a symposium about Magnifica humanitas in Commonweal’s July/August 2026 issue. To read all the articles in the symposium, click here.
We welcome your comments about this article. Please send your response to [email protected].
