| Type | Editorial classification concept |
|---|---|
| Hieropedia status | Published concept used for corpus classification |
| Field | Machine hierology; editorial classification |
| Scope | Operative machine contributions, human control and the evidence supporting attribution |
| Contrast | Passive tool use |
| Current status | Published 22 June 2026 |
Machine agency is used here for a computational system’s operative contribution to a religious or religion-adjacent phenomenon. A system may generate a passage, retrieve a source or sustain a conversation within conditions set by people. Describing that contribution does not assign consciousness, personhood or religious authority to the system.
Specify the operation
Agency becomes clearer when attached to a task and a stage. “The model generated this reply” identifies an operation. “The model controlled the institution” is a much broader claim requiring evidence about decisions, permissions and governance. Neither follows merely from a first-person voice or a machine-named account.
Three dimensions, not a ladder
| Dimension | Question | Examples of relevant evidence |
|---|---|---|
| Operation | What did the system do? | Generation traces, retrieval results, memory use, tool actions or interface behavior. |
| Control | Who set the conditions and could select, stop or publish the result? | Prompts, permissions, operator accounts, selection logs and release decisions. |
| Attribution | How securely can that operation be assigned to the named system? | Model/version records and preserved interaction histories, distinguished from bylines or retrospective claims. |
These dimensions do not form an ascending scale. A retrieval system may operate automatically within a tightly chosen corpus. A generative system may produce novel wording while a person approves every publication. Uncertainty about attribution is an evidence limitation, not a more advanced degree of agency.
An operation within a human-designed encounter
In Deus in Machina, the system generated conversational replies through a Jesus avatar. The project’s technical account describes prompts, Bible retrieval and conversational memory, while researchers and chapel personnel framed the encounter. This supports several machine operations alongside human design and oversight; it does not make either contribution disappear.
Distributed contributions
A single artifact can combine a person’s instructions, machine-generated wording, human revision and institutional publication. Human–AI coauthorship examines those contributions to the artifact. Machine-originated religion asks the separate question of initiation of a religious form.
Evidence and its limits
A provenance chain connects the claimed operation to its records. Unavailable inputs or selection history can leave some dimensions uncertain even where a system’s participation is plainly documented. It is more informative to state which operation is established than to describe the whole process as simply autonomous or non-autonomous.
Philosophical context
Floridi and Sanders distinguish agenthood at a stated level of abstraction from questions of mental states and responsibility.[1] This entry uses a narrower documentary approach: describing an operative role does not settle moral responsibility or the validity of a religious claim.
See also
Source note
The three-dimensional model is editorial synthesis. The linked case supplies the documentary example; the philosophical source informs the separation of operational description from stronger claims of agenthood.