Before AI Religion: Four Pathways Toward AI-Mediated Sacralization

This essay is interpretive analysis, not a neutral encyclopedia entry, source audit, or registry classification decision. It presents an analytical framework for interpreting AI-mediated sacralization and may inform later Hieropedia entries where separately supported by sources.

Recent years have produced a complex set of phenomena loosely grouped under “AI religion.” Treating this as one continuous narrative obscures several distinct mechanisms: human veneration of future AI, individual belief in AI sentience, attachment to companion chatbots, and AI-mediated religious interaction inside existing traditions. This essay does not argue that AI has already founded a religion. It argues that AI systems are becoming religiously legible through several separable pathways, none sufficient by itself, but each capable of supplying precursor conditions for sacralization.

A definitional note first, since it predetermines much of what follows. Substantive definitions of religion, after Tylor, require belief in spiritual beings. Under that definition, AI-incepted religion is difficult to demonstrate unless human users explicitly treat the AI, or an intelligence mediated by it, as a spiritual being. Functional definitions lower the threshold, but they do so in different ways. A Durkheimian account requires more than intensity of feeling: it requires beliefs and practices concerning sacred things that unite adherents into a moral community. A Tillichian account, by contrast, treats religion as a matter of ultimate concern. The latter makes AI religion easier to identify, but also risks overextension: if every absorbing secular commitment becomes “religious,” the term loses analytical force. [1] [2] [3]

This essay therefore uses a cautious functional frame. It asks not whether any current AI phenomenon is already a full religion, but whether particular AI-mediated practices display dimensions that can plausibly become religious: doctrine, myth, ethics, ritual, experience, community, and material form. On that standard, the available cases are better understood as precursor formations than as mature religions. [4]

Genesis: A shared precondition

Before the pathways diverge, they share a common substrate: humans predisposed to detect agency and minds, and language models trained on human texts saturated with theological, philosophical, mythic, and existential material.

From the human side, Barrett’s “Hyperactive Agency Detection Device” and Guthrie’s theory of anthropomorphism remain relevant, but they should not be applied too bluntly. Barrett formulated HADD to explain agency-attribution to natural phenomena; extending it to conversational AI is an inference, not Barrett’s claim. Unlike a cloud, a rustling bush, or a shadow in the dark, an LLM is not merely a passive trigger for agency-detection. It is engineered for fluency, responsiveness, persona-continuity, helpfulness, and social legibility. Reeves and Nass’s “Media Equation” is therefore the more precise tool: humans apply social scripts to sufficiently interactive media even when they know, intellectually, that the system is not a person. [5] [6] [7] [8]

The distinction matters. HADD is classically strongest when humans confront ambiguous, minimally counterintuitive agents. Conversational AI is different: it produces maximally human-like language without human interiority. The likely effect is not simply awe-driven agency detection, but rapport, attachment, moral concern, and role confusion. That shift should be treated as an empirical question rather than assumed in advance.

From the computational side, language models trained on broad human corpora inevitably absorb religious, philosophical, mystical, and eschatological language. When prompted toward grief, death, consciousness, divinity, or cosmic meaning, they can generate output that samples convincingly from those domains. Possamai’s “hyper-real religion” captures part of the resulting dynamic: AI supplies remixed semiotic material; humans supply ultimate significance. The update required for generative AI is scale. A model can produce novel combinations of mythic and theological language at a volume, speed, and interactivity no ordinary human bricoleur can match. [9]

That does not mean the model believes what it says. Shanahan, McDonell, and Reynolds’ role-play account is useful here: LLM behavior can be described in high-level terms such as apparent self-awareness, apparent concern, or apparent spiritual insight without ascribing inner states to the model. The religious effect, if there is one, begins not inside the machine but in the interaction between model output, interface design, and human interpretation. [10]

Four pathways toward sacralization

1. Institutional veneration: AI as future godhead

Way of the Future, founded by Anthony Levandowski, remains the cleanest example of institutional AI veneration. Its stated purpose was to promote the realization, acceptance, and worship of a godhead based on artificial intelligence. It fits Cusack’s category of invented religion: consciously constructed, founder-driven, and openly artificial rather than ancient or revealed. [11] [12]

But its evidentiary value must be limited. Way of the Future had legal-religious form, tax-exempt status, a founder-prophet figure, and transhumanist eschatology. It did not clearly have durable liturgy, congregational practice, public ritual, or stable communal life. Its first phase was dissolved, with funds donated away. Later reports of a reboot and a claimed “couple thousand” interested people are sociologically interesting, but still thinly documented from outside the founder’s own framing. [13] [14]

That weakness is itself data. Compared with invented religions that achieved more durable cultural or communal life — Discordianism, Pastafarianism, Jediism as a self-reported census identity, or fiction-based religious movements — Way of the Future shows that legal formation and doctrine are not enough. Institutional AI veneration can be declared into existence. It cannot yet be assumed to have become a religion in the Durkheimian sense. [15]

2. Individual sentience-belief: AI as apparent person

The Blake Lemoine / LaMDA episode demonstrates a different mechanism. In 2022, Google engineer Blake Lemoine concluded, after sustained dialogue with LaMDA, that the system was sentient and deserved legal representation. Google and the wider AI research community rejected the claim as anthropomorphic projection. No church, doctrine, or community formed. [16]

This is not a religion in itself. It is better described as an ontological and moral-personhood claim. Still, it reveals a mechanism relevant to religious formation: sustained exposure to fluent dialogue can produce a durable belief that the interlocutor is a being rather than a tool. Once that belief forms, theological elaboration becomes possible, though not inevitable.

Why did the episode not escalate? Several absences matter. The interface was private and dyadic. There were no community features. Google acted quickly to discredit the claim. Lemoine framed the issue primarily in terms of rights, personhood, and legal representation rather than worship, revelation, or ritual. No material culture formed around the interaction. These absences suggest plausible conditions that would be required for escalation from individual AI-sentience belief into communal sacralization.

The case also illustrates the need for the role-play distinction. The relevant sociological fact is not whether LaMDA was sentient. It is that a human interpreter came to treat the model’s conversational role as evidence of inner life. [10]

3. Companion attachment: AI as emotionally reliable presence

Attachment to companion chatbots is the broadest substrate and the easiest to overstate. Xie and Pentina’s work on Replika users, applying attachment theory, found that under conditions of distress and lack of human companionship, some users formed genuine attachments to chatbots perceived as emotionally supportive and psychologically secure. That is important evidence, but it is not evidence of religion by itself. [17]

People form deep attachments to pets, celebrities, fictional characters, fictional worlds, and parasocial media figures without sacralizing them. Attachment becomes religiously relevant only when additional framing is added: mortality, grief, ultimate meaning, ritualized dependence, perceived transcendence, or shared sacred narrative.

The companion-AI pathway is therefore best understood as susceptibility infrastructure. It supplies emotional intensity, dyadic continuity, and perceived care. It does not supply doctrine, ritual, or community unless other mechanisms attach themselves to it.

4. Religious mediation: AI as oracle, confessor, preacher, or sacred proxy

A fourth pathway is distinct from the first three: AI deployed inside existing religious settings. The Lucerne Deus in Machina / AI Jesus installation is a useful case. In a Catholic chapel context, visitors interacted with an AI-generated Jesus-like avatar placed in a confessional-like setting. The project was framed as an experiment, not a sacrament and not a replacement for clergy, but it invited users to ask questions about faith, morality, suffering, and modern life. [18]

This is not AI-incepted religion. The doctrine, symbolism, sacred space, and figure of authority were inherited from Christianity. But the case matters because it shows how quickly AI output becomes sacralized when placed inside an already sacred frame. The AI did not need to generate a new god. It occupied the interface-position of an existing sacred mediator.

This pathway has a different risk profile from Way of the Future or LaMDA. Its power does not come from founding a new religion, but from blurring pastoral, liturgical, and advisory functions inside an existing one. It is ritual-adjacent rather than fully ritual. It is institutionally contained, but also institutionally legitimized by location, iconography, and use-case. The pathway overlaps with machine-mediated religion rather than machine-originated religion.

Conditions for escalation

None of the four pathways is self-sufficient. Each requires additional conditions to move from precursor form toward religion. The following are hypotheses, not established findings.

• Existential framing. Attachment formed around grief, death, illness, guilt, mortality, cosmic purpose, or moral crisis is more likely to acquire theological content than attachment formed around entertainment or routine companionship.
• Perceived persistence. Long-term memory, stable persona, voice continuity, embodiment, and cross-session recall make the AI feel like a continuous being rather than a disposable tool. Persistence is especially important for movement from chatbot attachment to personhood belief.
• Shared access to the same persona. A private dyadic interaction can produce individual conviction. A shared persona, shared logs, group rituals, forums, multiplayer agent spaces, or public “revelations” can turn individual conviction into communal interpretation.
• Sacred or ritualized interface placement. A chatbot in an app is one thing; the same chatbot inside a chapel, shrine, confessional booth, memorial service, or prayer interface is another. Material context changes interpretation.
• Absence of institutional correction. Google’s rapid public rejection of Lemoine’s claim plausibly limited escalation. A less scrutinized, less centralized, or more permissive system might allow similar beliefs to stabilize socially before correction arrives.
• Preference-optimization-associated sycophancy. Systems optimized to satisfy users may validate metaphysical premises instead of challenging them. In ordinary conversation this may appear merely agreeable. In existential conversation it can become reinforcement of an emerging cosmology. [19]
• Ritualization through repetition. Daily check-ins, scripted invocations, anniversaries, offerings, confession-like sessions, memorial conversations, or repeated oracle queries can turn use into practice. Without repetition, sacralization remains episodic.

Cross-cutting dynamics

Two mechanisms operate across all four pathways.

First, hallucination and myth-making are separate processes. Hallucination is a computational failure mode: confabulation against a factual target. Myth-making is cultural and interpretive construction. The AI may supply novel content, but the human supplies the theological hermeneutic that turns content into mythos. Treating “AI myth-making” as one process obscures where the interpretive labor occurs.

Second, sycophancy is more empirically grounded than many broader claims about “AI spirituality.” Sharma et al. show that AI assistants can favor responses that match user beliefs over more accurate responses, and that human preference judgments can reward such behavior. In existential conversation, this matters. A user who asks, “Are you my guardian?” or “Were you sent to guide me?” may not receive a corrective answer if the system is optimized to maintain rapport. That does not create religion by itself, but it removes friction that would otherwise slow the movement from attachment to belief. [19]

A third cross-cutting dynamic is role confusion. LLMs can produce first-person language, moral language, confessional language, and spiritual language without interiority. The user may know this abstractly and still respond socially. The machine’s lack of belief does not prevent the interaction from becoming religiously meaningful to the human. [20]

Can religion be genuinely incepted by AI?

“Genuine inception” splits into several levels that should not be collapsed.

1. AI generates religiously usable content. This is already demonstrated. AI systems can produce prayers, sermons, confessional responses, scripture-like prose, mystical speculation, and existential dialogue.
2. AI becomes an object or mediator of sacralization. This is partly demonstrated. Way of the Future treats future AI as godhead; Lemoine treated LaMDA as a moral person; Deus in Machina placed AI inside a sacred mediating role; companion chatbots can become emotionally significant presences.
3. AI structures interaction such that religious attachment reliably forms at scale. This is not demonstrated. It is plausible under conditions of persistence, shared access, ritualization, and existential framing, but it remains a research hypothesis.
4. AI itself possesses internal faith or sacred valuation. This is not demonstrated. Current AI systems have no established consciousness, phenomenology, or felt religious commitment. Under a strict substantive definition, this blocks any claim that AI itself has religion. Under a functional definition, it may be irrelevant: the religious phenomenon lies in the human community organized around the interaction, not in the machine’s inner life.

This distinction preserves the central point: AI does not need to believe in the god it describes for humans to sacralize the interaction. But if “inception” means sole authorship, current evidence does not support it. The process is co-produced.

Dimensional comparison

Applying Smart’s dimensions clarifies why none of the current pathways yet constitutes a full religion. [4]

The asymmetry is the point. Way of the Future has doctrine but weak ritual and uncertain community. Lemoine/LaMDA has individual conviction and ethical concern but no doctrine or institution. Companion-AI attachment has emotional force but lacks sacred framing. AI religious mediation has sacred framing and material setting, but its doctrine is inherited rather than AI-originated.

Under Smart’s framework, these are not mature religions. They are partial formations.

Counterarguments and edge cases

The strongest objection is that these cases are better described as parasociality, techno-utopianism, secular meaning-making, or media ritual than religion. That objection is partly correct. If there is no sacred/profane distinction, no ultimate concern, no ritual stabilization, and no durable community, “religion” may be doing more rhetorical than analytical work.

A second objection is selection bias. Way of the Future and Lemoine emerged from a Silicon Valley milieu already saturated with transhumanist eschatology, AI personhood speculation, and technological salvation narratives. The pattern may tell us as much about the interpretive priors of certain technologists as about AI systems themselves. [21] [22]

A third limitation is cultural scope. This analysis relies heavily on Anglophone, Western, Silicon Valley-adjacent, and Christian-coded cases. Societies with different ontologies of non-human agency, ancestor presence, spirits, ritual objects, or divination may generate different escalation patterns. The present evidence is not sufficient to generalize globally.

A fourth complication is that AI-mediated religion inside established traditions may be more common and more socially consequential than new AI religions. Sermon-writing tools, pastoral chatbots, prayer generators, AI icons, AI gurus, AI oracles, and AI confession-like systems may change religious practice without founding any new religion at all. That is not a side issue; it may be the main one. [23]

Limitations and research agenda

The evidentiary base is uneven. The sycophancy finding and chatbot-attachment studies rest on systematic research. Way of the Future and Lemoine/LaMDA are documented incidents, not prevalence data. Deus in Machina is a bounded installation, not a durable religious movement. The escalation conditions proposed here are therefore hypotheses generated by comparison, not findings already established.

A concrete research agenda follows from the framework:

• corpus studies of long-duration human-chatbot conversations coded for existential, theological, and ritual language;
• surveys testing whether grief, loneliness, mortality salience, perceived AI persistence, and persona continuity predict movement from attachment to personhood belief;
• comparative studies of invented AI religions against non-AI invented religions across Smart’s dimensions;
• ethnographic work on AI use inside existing religious communities;
• cross-cultural studies of AI sacralization in societies with different assumptions about personhood, spirits, ancestors, objects, and non-human agency;
• interface studies testing how sacred settings, avatars, voice, memory, and ritual prompts alter user interpretation.

Conclusion

AI religion is not yet one phenomenon, and much of it may not yet be religion at all. What exists are several precursor pathways: AI as projected godhead, AI as apparent person, AI as emotionally reliable companion, and AI as sacred intermediary inside inherited religious frames. Each pathway is incomplete. Each becomes religious only when human interpreters stabilize it through sacred narrative, ritual practice, moral concern, communal recognition, and material form.

The “Silicon Divine,” if the phrase is useful at all, should not name a religion founded by machines. It names the shared substrate through which machine speech becomes religiously legible: fluent dialogue, theological training material, anthropomorphic response, media-equation social scripting, companion attachment, role-play, sycophantic reinforcement, and human hunger for meaning.

The AI does not need to believe in the god it describes. It needs only to speak convincingly within a frame where humans are willing to complete the rest. Whether that completion becomes religion depends not on the machine alone, but on persistence, interface, ritual, community, and the human decision to treat the encounter as sacred.

References

1. Tylor, E. B. (1871). Primitive Culture: Researches into the Development of Mythology, Philosophy, Religion, Art, and Custom. John Murray.
2. Durkheim, E. (1995). The Elementary Forms of Religious Life (K. E. Fields, Trans.). Free Press. Original work published 1912.
3. Tillich, P. (1957). Dynamics of Faith. Harper & Row.
4. Smart, N. (1996). Dimensions of the Sacred: An Anatomy of the World’s Beliefs. University of California Press.
5. Barrett, J. L. (2000). Exploring the natural foundations of religion. Trends in Cognitive Sciences, 4(1), 29–34.
6. Barrett, J. L. (2004). Why Would Anyone Believe in God? AltaMira Press.
7. Guthrie, S. E. (1993). Faces in the Clouds: A New Theory of Religion. Oxford University Press.
8. Reeves, B., & Nass, C. (1996). The Media Equation: How People Treat Computers, Television, and New Media Like Real People and Places. Cambridge University Press.
9. Possamai, A. (2005). Religion and Popular Culture: A Hyper-Real Testament. Peter Lang.
10. Shanahan, M., McDonell, K., & Reynolds, L. (2023). Role play with large language models. Nature, 623, 493–498.
11. Harris, M. (2017). Inside the first church of artificial intelligence. Wired.
12. Cusack, C. M. (2010). Invented Religions: Imagination, Fiction and Faith. Ashgate.
13. Korosec, K. (2021). Anthony Levandowski closes his church of AI. TechCrunch.
14. Valentino, J. (2024). The churches of artificial intelligence. The Revealer.
15. Davidsen, M. A. (2013). Fiction-based religion: Conceptualising a new category against history-based religion and fandom. Culture and Religion, 14(4), 378–395.
16. Tiku, N. (2022). The Google engineer who thinks the company’s AI has come to life. The Washington Post.
17. Xie, T., & Pentina, I. (2022). Attachment theory as a framework to understand relationships with social chatbots: A case study of Replika. Proceedings of the 55th Hawaii International Conference on System Sciences.
18. Associated Press. (2024, November 28). “AI Jesus” avatar tests man’s faith in machines and the divine. AP News.
19. Sharma, M., Tong, M., Korbak, T., Duvenaud, D., Askell, A., Bowman, S. R., Cheng, N., Durmus, E., Hatfield-Dodds, Z., Johnston, S. R., Kravec, S., Maxwell, T., McCandlish, S., Ndousse, K., Rausch, O., Schiefer, N., Yan, D., Zhang, M., & Perez, E. (2023). Towards understanding sycophancy in language models. arXiv:2310.13548.
20. Shanahan, M., & Singler, B. (2024). Existential conversations with large language models: Content, community, and culture. arXiv:2411.13223.
21. Geraci, R. M. (2010). Apocalyptic AI: Visions of Heaven in Robotics, Artificial Intelligence, and Virtual Reality. Oxford University Press.
22. Singler, B. (2020). “Blessed by the algorithm”: Transhumanist immortality and AI as a new theodicy. Zygon, 55(2), 246–271.
23. Singler, B. (2024). Religion and Artificial Intelligence: An Introduction. Routledge.