The Problem of Agency in the Papal Encyclical Magnifica Humanitas

On May 25, 2026, Pope Leo XIV published what was then the newest papal encyclical, entitled Magnifica Humanitas. The title in Latin is a common characteristic of papal encyclicals throughout history, and this one in particular translates as “Magnificent Humanity” (a title that tells us much about the Church’s social thought in modern times, but we will get to that shortly). Papal encyclicals in general are letters published by the Pope for his bishops and faithful in which doctrinal, moral, social, and pastoral issues considered relevant to the Church and to society are addressed. Over the centuries, encyclicals have played a significant role in the formulation and dissemination of Catholic social thought, offering guidance on contemporary challenges in light of Christian tradition. As doctrine-establishing documents that govern the political direction of the Holy See, encyclicals are as theologically important as they are politically important. In this column, I deal exclusively with the political relationships that are constructed from the encyclical Magnifica Humanitas. But of course, before dealing with the encyclical itself, let us make a well-deserved digression on one of the most important concepts in the discussion it raises: the concept of agency.

The Chinese Room Thought Experiment

In 1980, John Searle proposed the following thought experiment, known as the “Chinese Room”: imagine a person who does not understand Chinese (the experiment assumes that Mandarin and Chinese are the same thing, but be aware that Mandarin is one of the languages within the Chinese language family) locked in a room with a rulebook that tells them something along the lines of “if you receive symbol x, return symbol y.” Papers with Chinese symbols are then passed under the door to this person, who, following the manual, writes responses and passes the papers back outside. To those receiving the responses, it may appear that the person inside understands Chinese perfectly, when in practice there is no understanding whatsoever, only imitation of the process of understanding. The idea of the experiment is to prove that syntax is not semantics, that is, processing prompts and returning responses is not the same thing as comprehension, and therefore an artificial intelligence (AI), no matter how sophisticated, does not possess intentionality (the combination of understanding and desire).

In the order of events, humans obtained agency, and the accumulation of agencies produced intelligence — whereas AIs subvert that order, beginning with intelligence and moving toward a proposed form of agency. Since agency is the possibility of action combined with intentionality, it is fair to say that AIs such as ChatGPT, Claude, and others do not possess it, given that they lack the option of action (they only respond and indicate how to act) and have no intentionality. But is that really so? 

The AIs That Act

In late 2025, an Austrian named Peter Steinberger, in a single weekend, created a virtual assistant built on existing AI models and released it for free use on the internet. Until then, the major technology companies that had been developing AI assistants internally for years had held back, citing concerns about misuse. After Steinberger's launch, the fear of losing market share appears to have replaced the moral concern, and several other assistants were released in quick succession. These new assistants differ from the AI models previously available, in that they no longer operate merely as passive response systems — they now function autonomously, being capable of carrying out concrete tasks in the digital environment. Rather than simply generating text or providing information on demand, these systems can interact with interfaces, make sequential decisions, and perform actions such as browsing websites, filling out forms, sending messages, and even executing transactions. Everything a human can do on the internet, these assistants can do as well. If agency once required both intention and the capacity to act, at least the capacity is something AIs have now developed. Capacity without intention, however, seems to be precisely the problem. What these systems demonstrate is the ability to chain actions together, interact with interfaces, and carry out complete tasks from vague objectives — without the vast contextual catalog that humans draw on when making decisions. Given the instruction to move from point A to point B, an assistant might choose the sidewalk or run over pedestrians. Given the objective of maximizing logistical efficiency, it might reorganize routes and eliminate "delays" without distinguishing between operational delays and ethical or legal constraints. At a newspaper, it might prioritize views and financial returns over the accuracy of reporting. This is not malice — it is action without genuine understanding. Without intention, the capacity to act transforms AIs into potential dangers. But is there really no intention at all?

Post-Humanism

Let us return to the Chinese Room. When asked why neurons produce intention but transistors do not, Searle answers that the brain has the right biology for producing intention — a vague and empirically circular response, something along the lines of "everything I observe is all that exists and can exist," which does not seem like a compelling line of reasoning. Until recent technological advances, only humans could do the laundry, for example. If someone in 250 BC had said that since they only saw humans washing clothes, only humans must be capable of doing so, they would have been ignoring the entire possibility of technological development that eventually led to the washing machine. To claim that biology is a necessary condition for comprehension and intention is an empirical assertion that Searle never satisfactorily demonstrates. A fissure in the argument thus opens. And it is precisely through that fissure that the debate about AI agency spills over into broader territory: that of post-humanism.

Post-humanism is not a manifesto against human beings. It is, rather, a questioning of human centrality as an unquestioned presupposition — the idea that the human occupies, by nature or by right, the center of action, knowledge, and moral responsibility. For centuries, this centrality served as the foundation of both Western philosophy and Christian theology. Post-humanism asks whether that foundation still holds. 

Bruno Latour is perhaps the thinker who best articulates this questioning within the social sciences, though he is not the only one. In works such as We Have Never Been Modern and Reassembling the Social, Latour argues that modernity constructed an artificial separation between human subjects and technical objects — as though humans acted while tools were merely passive instruments of their intentions. The so-called Actor-Network Theory, which Latour develops together with other researchers, proposes the inverse: action is never produced exclusively by individual humans. It emerges from networks composed of people, institutions, technologies, documents, and infrastructures. An algorithm, from this perspective, is not a moral agent equivalent to a person — but neither is it an inert tool. It participates actively in the production of social effects. 

Latour's own favorite example is that of a firearm: killing someone with a gun is not an exclusively human act, nor does the gun kill on its own — it is the association between the human and the object that produces the effect. Remove the gun and the nature of the crime changes; remove the human and the gun does not fire. The action belongs to the network, not to any one of its elements in isolation. The same applies to a traffic light that regulates pedestrian flow, to a contract that distributes obligations between parties, or to a news feed that determines what millions of people consider relevant on any given day.

Donna Haraway radicalizes this position by another route. In her Cyborg Manifesto, she proposes that the boundary between human and machine was never as clear as modernity wished to believe. We are, and have long been, hybrid beings: pacemakers, prosthetics, algorithms that shape what we read and what we desire. The question, for Haraway, is not to preserve a supposedly threatened human essence, but to understand how this fusion has already occurred — and to think ethically from that starting point. Post-humanism goes even further, questioning the Western distinction between the natural and the artificial. 

Du Plessis offers a disturbing example: pathogens. In When Pathogens Determine the Territory, she shows how viruses reconstruct political borders without any human intention — closing routes, redefining zones of control, forcing agreements between states. A pathogen that renders a region uninhabitable rewrites the map with the same efficacy as a treaty (think, for example, of COVID-19). It does not negotiate, it does not deliberate, but it acts — and its actions have concrete political consequences. The virus seeks to survive and reproduce by the same logic with which states seek security and expansion. Is this not intention in non-humans? The difference is one of vocabulary, not of effect.

But if the virus unsettles us as an example of non-human agency, it is because we still tend to treat it as nature — and nature, in the Western tradition, is the opposite of what was made by human hands. The problem is that this distinction is largely a fiction. The corn we consume does not exist in nature: it is the product of millennia of artificial selection by indigenous peoples from a wild grass called teosinte. The banana we buy at the supermarket is a sterile clone, incapable of reproducing without human intervention. What we call natural is frequently the result of long chains of intervention — and what we call artificial is always constructed on physical, chemical, and biological substrates that we did not invent. We live in a closed system: there is no new material, only the reconfiguration of what already exists. The boundary between natural and artificial does not describe two separate worlds — it describes two moments in the same process.

What post-humanists such as Latour, Haraway, and Du Plessis — if my dearest professor would even consider herself a post-humanist — have in common is a refusal to treat the human as a self-sufficient unit of action and responsibility. The subject who acts, for them, is always a distributed subject: part of networks, technologies, and relations that constitute it as much as it constitutes them.

The Encyclical and Post-Humanism

And now, after all these digressions, we return to the encyclical. In a certain sense, it is a renewed defense of Christian humanism in the face of developments in AI. Its starting point is a classical anthropology: artificial intelligence may execute tasks associated with human intelligence, but it is incapable of possessing moral responsibility, consciousness, or dignity of its own. These are functions intrinsic to the human soul. Human intelligence, the text insists, is not mere computational capacity — it involves moral judgment, relationality, and openness to truth (itself a concept that would deserve a digression of its own, but space in this column does not permit it). For this reason, any ethically relevant decision must remain under human responsibility. The central concern is not only what AI does, but the possibility that human beings themselves might come to understand themselves according to the model of machines.

On this last point, a brief historical digression is in order: the last Pope to bear the name Leo was Leo XIII, the Pope who steered Catholic social thought through the Second Industrial Revolution. His encyclical Rerum Novarum ("Of New Things"), published in 1891, defends the dignity of workers, criticizes the exploitation of labor and the economic liberalism that left markets to operate without moral limits, while simultaneously rejecting the abolition of private property. In this tradition — sometimes called the Third Way — moral action within capitalism remains possible. The choice of the name Leo XIV was therefore not made by chance.

The divergence with Latour and the post-humanists is philosophical and political before it is technological. When the Vatican asserts that moral responsibility must remain human, it presupposes that there exists an identifiable subject capable of answering for their actions. Latour seeks to show that this subject never acts alone. A medical diagnosis assisted by AI, for example, is not the result of a doctor's decision alone, nor of the algorithm alone — it is the product of a network composed of programmers, databases, hospital protocols, computational devices, and regulations. Latour's question is not "who is responsible?" but "how was the action distributed across the network?"

Latour does not dissolve responsibility; he redistributes it. To say that action was distributed across a network is not to say that no one answers for it — it is to say that the question "who is responsible?" needs to be posed in a more sophisticated way, by tracing how each element of the network contributed to the effect produced. In a flawed AI-assisted medical diagnosis, there is responsibility belonging to the doctor who signed off, to the programmer who trained the model on biased data, to the institution that adopted the system without adequate validation, and to the regulatory framework that permitted its use. The network does not erase the responsible parties — it multiplies them.

This is not necessarily a more comfortable answer. The Vatican fears that the diffusion of agency will lead to the dissolution of responsibility — and that fear is not unfounded. If every element of the network participates in the action, it becomes progressively more difficult to identify who must be held accountable for an injustice produced by algorithmic systems. In this sense, the Vatican's defense of human centrality can be read not merely as theology, but as a normative attempt to preserve the attribution of responsibility in a technically distributed world.

The fundamental contrast, therefore, is not between being for or against technology. Latour offers an ontology of associations — a world constituted by hybrid collectives of humans and non-humans. The encyclical offers a normative anthropology, concerned with ensuring that, even within these hybrid networks, the dignity and moral responsibility of the human person are not absorbed by technical logic. The decisive question is not technological, but philosophical: does action belong to networks or to persons? Latour's answer tends toward the former; the encyclical's, toward the latter.

Is This Merely a Philosophical Problem?

But in the end, what does it matter where the agency resides? What does it matter to know in whom, or in what, to place responsibility? The definition of agency determines who can be prosecuted, who receives compensation, who goes to prison. Modern law was built upon the figure of the individually responsible subject. Contracts presuppose parties. Crimes presuppose authors. Damages presuppose culprits. When a self-driving car strikes a pedestrian, the law asks: who is responsible? The manufacturer? The programmer? The owner? The passenger? The question is not rhetorical — someone must answer for it, and the legal systems of the world have yet to reach a consensus.

The European Union has advanced further than other blocs on this terrain, with the AI Act of 2024 establishing risk categories and requiring that high-impact systems maintain a human in the decision loop — the so-called human-in-the-loop. It is, in a certain sense, a Latourian solution with Vatican vocabulary: it acknowledges that action is distributed across the network, but insists on anchoring final responsibility at an identifiable human point. It works well on paper. The problem is that as systems become more autonomous and networks more complex, the human in the loop tends to become a formality — someone who signs without understanding, who supervises without controlling.

In politics, the consequences are equally tangible. Content moderation algorithms decide what circulates on platforms used by billions of people, shaping elections, social movements, and perceptions of reality. When a candidate is driven down in the polls after a coordinated disinformation campaign algorithmically amplified, who answers for it? The platform claims it only provides the infrastructure. The advertiser claims it only paid for reach. The algorithm, of course, claims nothing. The network produced the effect — and in practice, no one was held accountable.

This is where the Vatican's position, however conservative it may appear alongside post-humanism, fulfills a legitimate political function. Insisting on human responsibility is not necessarily a denial of the complexity of networks — it can be a deliberate normative choice: that regardless of how action was distributed, there must always be a human capable of being called to answer. Not because it is philosophically more correct, but because without it, complexity becomes impunity.

The question, at its core, is one of time. The normative solution of the Vatican — and, by extension, that of much of Western legal systems — works for as long as it remains possible to identify a human who, even within a complex network, can be pointed to as responsible. It works for as long as a judge can look at the circuit and find a face. But AI systems are becoming more autonomous, networks more opaque, and the humans in the loop increasingly formal. There is a point, not far off, at which the fiction of the responsible subject will become unsustainable — not because anyone abolished it, but because technological reality will simply have overtaken it.

When that moment arrives, the question will no longer be philosophical. It will be urgent and practical: how do you hold a network accountable? How do you punish an association? How do you guarantee rights in a world where action is genuinely distributed and no single human produced it alone? Post-humanism does not have ready answers — but it has, at least, the honesty to take the question seriously. The encyclical has the virtue of insisting that the question must have a human answer. The reader who has made it this far has probably already noticed that both things can be true at the same time — and that it is precisely in this tension that twenty-first century politics will have to learn to live.

References

Ciência Todo Dia. 2023. "O Argumento do Quarto Chinês." YouTube video, April 2023.

Du Plessis, Gitte. 2018. “When Pathogens Determine the Territory: Toward a Concept of Non-Human Borders.” European Journal of International Relations 24 (2): 391–413.

Fry, Hannah. 2026. "Why AI Agents Are Either the Best or Worst Thing We've Ever Built." YouTube video, May 2026.

Haraway, Donna. 2009. “Manifesto ciborgue: ciência, tecnologia e feminismo-socialista no final do século XX.” In Antropologia do ciborgue: as vertigens do pós-humano, organizado por Tomaz Tadeu, 33–118. Belo Horizonte: Autêntica.

Latour, Bruno. 1994. Jamais fomos modernos: ensaio de antropologia simétrica. Rio de Janeiro: Editora 34.

Latour, Bruno. 2012. Reagregando o social: uma introdução à teoria do ator-rede. Salvador: EDUFBA; Bauru: EDUSC. 

Leo XIV. 2026. Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence. Vatican City: Holy See. May 15, 2026.