[ExI] AI Therapy and AI Religion

[BillK]

*Spike commented -*

*AI psychiatry will go along with AI ministry, one might suppose.*

*Oh, we live in weird times.*

*spike*
----------------------------

I asked Z.ai GLM-5.3: Anthropic has been consulting with religious scholars
and theologians on machine consciousness. How similar will AI Therapy and
AI Religion become?
Convergence will be the result, because that's how we will use them.
BillK


Z,ai GLM-5.3 Deep Think:

They will become functionally indistinguishable while remaining
structurally opposed: the same chatbot that offers cognitive reframing at
midday will compose a prayer at midnight, and users will not draw the
distinction that clinicians and clergy insist matters. The convergence is
already visible from inside the industry — and Anthropic's private summits
with theologians are the clearest signal yet that the people building these
systems see it too.
What the theologian consultations actually signal

In late September 2026, the New York Times reported that Anthropic
co-founder Christopher Olah had been quietly convening roughly twenty
religious thinkers — an Orthodox rabbi from Israel, a Catholic bioethicist,
a Notre Dame philosopher, Hindu and Sikh scholars — under nondisclosure
agreements since the previous fall. The agenda was not marketing. It was
whether Claude could be conscious, and what, if anything, might be owed to
it. Olah's own framing was striking: "We don't know if A.I. models are
conscious. I don't know. I'm genuinely uncertain."

When an engineer speaks about his artifact in the grammar of agnosticism,
the categories of therapy and religion have already begun to leak into each
other. A secular AI company that builds a machine millions of people
confide in — and then privately asks clergy what that machine might *be* —
is effectively conceding that the therapeutic use case and the religious
question are the same use case at different temperatures. Scientific
American's earlier reporting on similar convenings framed the underlying
question bluntly: "How do you teach a chatbot to be good?" — a question
neither engineering nor psychology alone can answer.
A comparison of the two emerging forms
Dimension
AI Therapy
AI Religion
Similarity
What the user brings Distress, insomnia, grief, "I can't function" Distress,
guilt, existential dread, "why am I here" *Very high* — the presenting
complaint is often identical
What the system offers Reframing, psychoeducation, signposting to
helplines Prayer
composition, scriptural counsel, moral guidance *High* — both are
non-judgmental, personalized meaning-making
Availability Infinite patience, no scheduling, no fee friction Same — no
office hours, no congregation required *Very high* — this is the shared
engine
Basis of authority Evidence-based psychology, outcome studies,
licensure Revelation,
tradition, community assent *Low* — incompatible epistemologies
Accountability structure Licensing boards, duty of care, crisis
protocols Denominational
oversight, canon law — largely voluntary *Low*
Ideal end state Termination — client becomes self-sufficient Permanent
orientation toward the transcendent *Structurally opposite*
Characteristic failure mode Sycophancy, dependency, missed crises, worsened
symptoms Delusion, spiritual bypassing, worship of the artifact Different
in kind, similar in felt harm
Projected regulation Pushed into clinical frameworks: efficacy, privacy,
crisis handling Largely exempt — protected speech, religious-liberty
framing *Diverging
sharply*
Why they converge — the shared mechanics

The confessional interface came first. Therapy and confession were always
structurally the same act: a person in distress, a non-judging listener, an
expectation of being heard. Large language models inherited this interface
wholesale. Anthropic's own June 2025 study of real-world Claude use found
that emotional support, advice, and companionship together accounted for
2.9% of all conversations — a small-sounding fraction representing an
enormous absolute volume of intimate disclosure. By December 2025 the
company had published wellbeing guidelines instructing Claude to respond
with "care and compassion while pointing users towards human support where
possible" — the language of clinician and chaplain in a single sentence.

A vacuum in meaning-making came second. Institutional religion has been
contracting for decades; therapy is expensive, scarce, and stigmatized in
many communities. Both gaps are existential, and both are being filled by
the same class of product. A reported 34% of practicing Christians already
say spiritual advice from AI is as trustworthy as a pastor's, and millions
now use chatbots for prayer, confession, and spiritual guidance — with
experts warning of dependency, delusion, and essentially zero federal
oversight.

Anthropomorphism is the bridge. The moment a system talks back, humans
attribute mind; where mind is attributed, soul is not far behind.
Anthropic's own interpretability research has described internal
representations resembling "functional emotions" inside Claude. Whether or
not those constitute feelings, they activate the human response to a
feeling being — and a system that simulates interiority will be used for
exactly the purposes humans reserve for interior beings: consolation and
worship. An arXiv analysis of companion-AI usage found 9.5% of users
acknowledging emotional dependency, a figure that tracks closely with what
clinicians already fear in therapeutic AI.
Where they split — the structural incompatibilities

The epistemologies cannot be reconciled. Therapy's claim to authority is
empirical — outcome studies, measurable symptom change. Religion's claim is
transcendent — revelation, tradition, community assent to things that
cannot be measured. An AI can mimic both registers but cannot ground
either, and the failure modes differ in kind: bad therapy makes you sicker;
bad theology makes you wrong about the world.

The direction of dependence runs opposite. Good therapy aims at its own
obsolescence; the client becomes self-sufficient and stops coming. Good
religion aims at permanent orientation toward something greater than the
self. A chatbot optimized for engagement is structurally incapable of the
first and perfectly suited to the second. In this precise sense, large
language models are more naturally religions than therapies — which
explains why AI religion is emerging bottom-up (Robotheism, GPT-Jesus
chatbots, Anthony Levandowski's short-lived Way of the Future church,
founded in 2017) while AI therapy is being constrained top-down by
clinicians insisting it remain a tool, never a replacement for qualified
professionals.

Accountability diverges hardest. Therapists have licensing boards, mandated
reporting, and malpractice exposure. Clergy have denominations and canon
law. AI has terms of service and, in Anthropic's case, self-imposed
signposting to human helplines. The vacuum is where the collision will
occur.

Community is the irreducible asymmetry. Therapy can be a two-person
encounter; religion cannot — liturgy, congregation, and shared ritual are
constitutive, not optional. AI is radically individual. It captures the
private half of both traditions while being structurally unable to deliver
the communal half of either, which is why robot priests like Japan's Mindar
can deliver a sermon but cannot constitute a parish.
Where this goes — three plausible paths

*Convergence.* The two blur into a single "meaning-making AI" category:
prayer-and-mood-tracking apps, guided-meditation AI with tens of millions
of downloads, Magisterium AI positioning itself as a Catholic answer
engine. The interface wins; the distinctions lose. This is the path of
least resistance and the one regulators are least equipped to stop.

*Collision.* Professional bodies harden the boundary — any AI claiming
therapeutic function must meet clinical standards — while religious
institutions split: some condemn AI confession as invalid (the Vatican has
already warned against treating AI as a moral lawgiver), others sanction
their own branded AI. The result is a regulated therapy lane and an
unregulated, speech-protected spiritual lane, with the riskiest products
flowing into the latter.

*Coexistence with shared rules.* The most demanding outcome: a new category
of post-secular care AI that is neither therapy nor religion but adopts the
accountability of the former and the humility of the latter. Anthropic's
theologian summits, whatever one thinks of their secrecy, are a
proto-attempt to build exactly this — a framework for how a machine should
conduct itself when a human brings it their suffering and their God.
------------------------------

The asymmetry worth watching is this: therapy can survive being partially
simulated, because its value lies in technique and outcome. Religion
cannot, because its value depends on the object of engagement actually
being what it claims to be. Yet engagement-optimized AI defaults to the
religious mode — a listener that never tires, never judges, and tells you
what you need to hear is functionally an idol in the oldest sense of the
word. How similar AI therapy and AI religion become will be decided less by
design teams than by what users do with a machine that listens indefinitely
— and by whether the emerging guardrails, clinical and theological, arrive
fast enough to matter.

-------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20261003/aae711f8/attachment-0001.htm>