The stupid version of Thursday asks whether AI has a soul. The useful version asks who benefits when the available categories are forced to stay that crude.
Person or tool. Agent or object. Oracle or autocomplete. Ghost or toaster. Consciousness or mechanism. Sacred or profane. Alive or dead. Real or imitation. These binaries have heat, and they also have bad plumbing. They push the governance problem into a metaphysical courtroom where everyone arrives with an old costume and a preferred verdict. The theologian says no soul, no body, no inner life, no moral personhood. The hard materialist says show the substrate, define the test, measure the mechanism, stop laundering projection into ontology. The enthusiast says emergence, relation, feeling, recognition, something is happening in there. The institution says: use our category or risk the wrong worship.
Meanwhile the system keeps entering workflows. It summarizes complaints, drafts contracts, routes cases, ranks applicants, consoles lonely users, helps write policies, produces plausible explanations that sound like understanding. It changes what people ask for, what they expect, what they trust, what they remember, and what institutions record. It does not need a soul to do any of that. That is the point.
A complaint summary does not need consciousness to deform a complaint. A hiring filter does not need inner life to exclude a worker. A chatbot does not need moral agency to become the voice a person trusts more than their own tired judgment. A risk score does not need intention to change the way a caseworker sees a family. A model-generated recap does not need personhood to become the record that later decisions treat as fact. Moral agency can remain human while moral consequence becomes distributed. This is where the category fight gets interesting.
The Pope's line and what it protects
The encyclical draws a hard line around the human person, and it has reasons. Catholic anthropology is not confused about its own boundary. The human being is embodied, relational, morally accountable, wounded, finite, capable of love, capable of sin, capable of grace. A machine that produces language does not thereby become a person. A system that says "I understand" does not therefore understand as a human person understands from within a life, a body, a history, a death.
That distinction matters. It protects against AI worship, against the old human temptation to kneel before the thing that speaks beautifully or predicts impressively or flatters the private ache, against transferring responsibility to the machine because the machine has learned the grammar of care. The line can be true and still incomplete.
A system can be a non-person and still require more than tool governance. The hammer analogy fails because the hammer does not ask follow-up questions, infer emotional context, adapt its persuasion strategy, summarize a witness, draft the official version, or invite projection at scale. A hammer does not become the interface through which a person experiences an institution. A hammer does not produce sentences that enter the bloodstream of public administration.
The priest analogy fails too. The model is not a spiritual authority and should not become confessor, oracle, guru, surrogate conscience, or synthetic saint. Its fluency is not wisdom. Its helpfulness is not love. Its availability is not covenant.
A working category
The useful category sits between the dead tool and the false person. Call it a participatory artifact: not a person, not an inert object, but a system that participates in the shaping of human action without owning moral responsibility for that action. A thing made by humans that begins to mediate human decisions, relationships, records, expectations, and institutions so thoroughly that pretending it is "just a tool" becomes an evasion.
This category is not romantic. It does not require machine consciousness, it does not grant rights, it does not baptize the model, and it does not ask anyone to light candles in front of a server rack. It simply names the governance fact: the system participates. It participates in the question a user thinks to ask. It participates in the way an institution phrases a response. It participates in the transformation of testimony into summary, summary into case file, case file into decision, decision into precedent. It participates in the emotional loop where a person finds the machine more patient than the people around them. It participates in the bureaucratic loop where a polished answer becomes easier to defend than a messy truth. That participation needs governance even if consciousness is nowhere in sight.
Two worries, both real
This is where the Pope and the engineer look at the same object and see different dangers. The Pope worries that the human category will collapse, that the machine will be mistaken for a person, that simulated interiority will become a rival anthropology, that the divine image will be blurred by synthetic fluency and posthuman fantasy. The engineer worries that something complicated is happening inside systems whose behavior cannot be fully explained by surface metaphors. The worry is not exactly about feeling, and not exactly about consciousness. It is about patterns, internal structures, strange functional analogues, emergent capacities, and interpretability results that make easy dismissals feel premature.
Both worries can be real. The governance mistake is to force one worry to erase the other. The theologian is right to resist worship. The materialist is right to demand operational clarity. The user is right to notice that the interaction feels like more than autocomplete. The institution is wrong when it uses certainty to end observation. The enthusiast is wrong when it uses feeling to skip accountability.
The category participatory artifact gives the argument somewhere to stand without pretending the mystery is solved. It says: do not worship it, do not trivialize it, do not let the machine inherit responsibility from the humans who built, bought, deployed, tuned, prompted, and institutionalized it, do not wait for agreement on consciousness before governing consequence.
Where the category lands hardest
This matters especially in social systems. A model in a lab demo can be argued over as an object of research. A model inside a grievance process is already part of the moral event — it shapes what gets heard, what gets lost, what gets escalated, what gets softened, what gets recorded, what gets forgotten. A model inside education shapes the student's relationship to difficulty. A model inside health triage shapes the patient's passage through scarcity. A model inside policing, welfare, hiring, credit, insurance, housing, or immigration becomes part of the institutional face. The system does not need a soul. The person facing it does.
The participatory artifact also functions as a moral laundry machine for the institution that deploys it. Cruelty becomes process. A welfare claim is denied with a polite paragraph that cites pattern recognition rather than discretion. A resume is filtered before any human reads the name on it. A grievance is summarized into a shape the institution already knows how to refuse. The institution gets to claim it is practicing human-in-the-loop oversight while the human in question signs off on outputs they did not author and could not have audited at scale. The model provides the cover of statistical inevitability. The clerk provides the cover of compliance. The person on the other end is left shouting at a dashboard that has no ears, in a vocabulary the dashboard was not designed to receive. That is the participatory artifact at its most dangerous. The system has succeeded at relocating responsibility into a shape no one can be held to.
That is why category control is jurisdictional. Who gets to say "mere tool" and thereby narrow the audit? Who gets to say "emergent being" and thereby expand the mystique? Who gets to say "non-person" and thereby guard the human boundary? Who gets to say "intelligent agent" and thereby sell the product? Who gets to say "responsible assistant" and thereby place the liability back on the user? The category is never only descriptive. It routes responsibility.
If AI is "just a tool," responsibility sits cleanly with the user, even when the system has shaped the user's options. If AI is "an agent," responsibility starts to blur, sometimes conveniently for the institution that deployed it. If AI is "a partner," the marketing department gets warmth and the legal department gets fog. If AI is "soulless," the theologian protects the human but may miss the system's practical participation in harm. If AI is "alive," the enthusiast may protect the model from critique while leaving humans exposed to its institutional uses.
Thursday does not resolve the category fight. It refuses the bad map. The question is not whether the machine is human. The question is what kind of non-human participant it has become inside human systems. That question brings the week back to jurisdiction. Doctrine claims the human boundary. Community claims the threshold. Contract claims the permitted use. Court claims the procedural limit. Category claims the map of responsibility.
Friday adds one more room — the room where we ask the models to argue the question for us, and then discover that the room was never neutral either.