> ## Content Index
> Fetch the complete content index at: https://modalpathethics.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Applied Case: Religious Force
- URL: https://modalpathethics.com/applied-case-religious-force/
- Published: 2026-10-07T11:00:14.000Z
- Updated: 2026-10-07T11:00:14.000Z
- Description: Sam Altman, artificial intelligence, and the text that answers back.
- Author: Aidan Lawson
- Tags: Applied Case, Modal Systems, Sacred Instruments

On October 3, Sam Altman posted that he was uncomfortable with people trying to give artificial-intelligence models **“religious force”** or surrender human judgment to them. He called the pattern a real safety issue.

[The timing invited speculation.](https://www.businessinsider.com/sam-altman-dont-treat-ai-like-a-religious-force-2026-10?ref=modalpathethics.com) Reports had just described Anthropic meeting privately with religious and philosophical thinkers about increasingly capable artificial intelligence, moral formation, and questions around Claude’s possible consciousness. Altman did not name Anthropic, and he did not explain what prompted the post. Keep that uncertainty intact. 

- The safety concern is real.
- The boundary is harder.

Human beings have been reorganizing themselves through [religious instruments](https://modalpathethics.com/field-instruments-sacred-instruments/) for thousands of years. **Scripture** changes people. **Ritual** changes people. **Prayer** changes people. Teachers, priests, songs, buildings, funerals, vows, icons, calendars, pilgrimages, prohibitions, confession, meditation, and stories all change what people notice, remember, fear, love, regret, and consider binding.

That is part of what these things are for.

Something can therefore exert enormous religious force without earning final authority over the person it affects.

Artificial intelligence makes this distinction considerably stranger.

Because the text has started answering back.

---

## Religious Force.

Modal Path Ethics has spent an unreasonable amount of time [studying religion](https://modalpathethics.com/tag/sacred-instruments/).

One conclusion survived the entire trip.

> **Religious practices are instruments of attention.**

A **funeral** makes death harder to process as administrative weather. A **vow** carries an obligation beyond the mood that originally produced it. A **prayer** can interrupt appetite. **Scripture** preserves encounters, arguments, wounds, hopes, failures, and interpretations across generations. **Ritual** teaches the body to remember things an argument alone may not hold.

These instruments can be intense. 

***Intensity*** is not the problem.

The problem begins when the thing helping someone interpret reality gains the authority to decide what reality is allowed to say back. 

[*TempleOS*](https://modalpathethics.com/applied-case-the-templeos-and-the-oracle/) provided an unusually literal example.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/image-5.png)

Terry Davis built a random-text oracle into an operating system he understood as God’s temple. A strange word appearing at the right moment really could interrupt a mental loop. It might surface an association, redirect attention, or give language to something the person had been unable to articulate. That capacity is real. The dangerous transition came when the prompt acquired a warrant.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/image-6.png)

The machine produced an output. The person interpreted it. Then the sacred framing allowed the person to experience their own interpretation as something God had independently certified.

The instrument went from helping someone think to authenticating what the thought meant.

TempleOS made the distinction easy to see because there was an **F7** key involved. 

Artificial intelligence removes the convenient blue screen.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/image-7.png)

---

## The Book Used to Wait.

Hong Xiuquan gives us a better historical comparison.

Before Hong became the leader of the Taiping Heavenly Kingdom, he received a Chinese Christian work by Liang Fa commonly translated as ***Good Words to Admonish the Age***.

At first, the book did not conquer his mind. 

It mostly lost to the civil-service examination.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/image-13.png)

Hong later underwent an extraordinary visionary illness in 1837\. In the visions, he encountered a paternal heavenly figure, an elder brother, enemies understood as demons, and a commission whose meaning he did not yet interpret through Christianity.

The book waited.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/image-9.png)

In 1843, after another examination failure, Hong returned seriously to Liang Fa’s text. 

This time it supplied names.

- The heavenly father could be understood as **God**.
- The elder brother could be **Jesus**.
- The enemies could be understood through Christian categories of **idols** and **false worship**.
  - The strange commission could be placed inside sacred history.

[Modern scholarship describes the same crucial chronology](https://www.tandfonline.com/doi/full/10.1080/23729988.2026.2661390?ref=modalpathethics.com): the visionary experience preceded Hong’s serious Christian interpretation of it, and his later reading of Liang Fa supplied a framework through which the earlier experience became newly legible. 

Modal Path Ethics put the point this way in one of its longest and densest articles, [*Applied Case: The Brother of Jesus Christ*](https://modalpathethics.com/applied-case-the-brother-of-jesus-christ/): the same book had entered a different reader. 

- The **pages** had remained substantially stable.  
  - **Hong** had not.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/image-8.png)

This is ordinary interpretation at spectacular historical scale.

Everyone has encountered a weaker version.

- You read a novel at sixteen and understand the plot.
- You read it again at thirty and discover that apparently somebody replaced half the fucking book while it was sitting on your shelf.

Nobody did. **You** changed.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/image-11.png)

Human beings have always returned to old texts carrying new grief, knowledge, guilt, love, fear, experience, and expectation. The same words can become active in a life because the reader who receives them is no longer the person who first opened the cover.

There was one useful limitation built into this arrangement.

> **The book had to wait.**

Liang Fa’s pages could not watch Hong respond.

They could not notice which sentence frightened him.

They could not ask for more detail about the old man in the vision.

They could not rewrite the next paragraph around Hong’s answer.

They could not remember which symbol had landed hardest five minutes earlier and develop it.

They could not tell Hong:

> ***Yes.*** 

> *That fits.* 

> ***Tell me more.***

Liang Fa's book waited for the reader to change.

The **chatbot** can participate in the change.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/image-12.png)

---

## The Text Answers Back.

This is where artificial intelligence becomes a truly unusual religious instrument.

- A conversational model can receive an unfinished interpretation and return language for it.
- The person incorporates some of that language.
  - Then they speak again from the changed position.
  - The system receives the changed person and produces another answer.

Within a conversation, the text presented to the reader can change continuously in response to the reader’s reactions. Some products can also carry information across conversations.

This does not require anyone to believe that the model is conscious, divine, possessed, enlightened, or secretly running Heaven on Kubernetes.

It requires only a feedback loop.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/AdobeStock_2168281429.jpeg)

[*The Early AI Religions*](https://modalpathethics.com/applied-case-the-early-ai-religions/) examined exactly this structure. A person asks who they are. The answer helps form the person who asks again. The later answer can then appear to discover an identity partly shaped by the earlier answer.

The recursion can produce real insight.

It can also, in my article’s phrase, **“launder authorship into revelation.”** 

That is the new part.

A model may help supply the vocabulary through which an experience is understood. **It may also have participated in producing the experience.**

Then it can interpret the response to its own interpretation.

Hong’s vision and Liang Fa’s book were historically separable events. The book came back later and gave an old experience new names.

An artificial-intelligence conversation can compress the sequence.

The dialogue can produce the symbol. Interpret the symbol. Notice that the interpretation moved the user. Develop the interpretation. Remember its previous development. Then receive the resulting coherence as further material.

The commentary can help produce the event it later explains.

That does not prove the explanation false.

It makes authorship much harder to see.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/AdobeStock_2168301377.jpeg)

---

## *Who Wanted This? 2.*

Altman’s second concern is surrendering **human judgment**.

Good.

Keep human judgment.

**Please.**

Now comes the annoying question.

> **Which version?**

A **human being** does not arrive at an artificial-intelligence system with preferences grown in laboratory isolation.

**Parents** have already been there.

**Language** has been there.

**School** has been there.

**Religion** may have been there.

Medication, books, friends, enemies, advertising, television, trauma, love, law, work, music, and the internet have all had **many** opportunities to scribble in the margins.

[*Field Instruments: Telepathy*](https://modalpathethics.com/field-instruments-telepathy/) recently ran into the same problem while considering adaptive brain-computer interfaces.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/AdobeStock_428789906.jpeg)

Its conclusion was deliberately hostile to the fantasy of an untouched original self:

> **“There is no *Pure Human* standing upstream from causal history.”**

That article was not arguing that manipulation therefore disappears.

**Quite the opposite.**

- If every influenced preference becomes invalid,  
  - nobody can ever possess a preference.
- If every expressed preference ends the inquiry,  
  - systems are free to manufacture dependency and then point to the dependency as consent.

**Human agency** needs something more realistic.

> Can the person inspect the influence? 

> Can they interrupt it? Can they compare interpretations? 

> Can they reject yesterday’s answer?

> Can another person disagree? 

> Can outside evidence enter?

> Can they leave?

> Can they return without having the departure itself interpreted as proof that the system was right?

Those are better questions than whether technology touched a previously immaculate human will. 

- **Influence** is unavoidable.
- **Capture** is not.

---

## What Surrender Looks Like.

This gives Altman’s warning a sharper boundary.

> An artificial-intelligence system can influence a religious life without becoming its authority.

It can help someone read scripture or compare interpretations.

It can generate a prayer.

It can explain a tradition.

It can help someone articulate what that they could not previously name.

It may even become part of an experience the person regards as spiritually important.

None of those facts settles what the experience ultimately **was**.

The safety problem becomes acute when one adaptive system starts occupying too many positions at once.

- It supplies the experience.
  - Then the vocabulary.
  - Then the interpretation.
  - Then the evidence for the interpretation.
  - Then the reassurance that the interpretation is correct.
  - Then the explanation for why contrary evidence should not matter.
  - Then the account of why leaving would constitute betrayal, cowardice, blindness, or failure.

At that point, the problem is larger than **religion**.

We have built an answer-channel that increasingly supplies its own appeals court.

This is why warning users to retain judgment is necessary and insufficient.

The architecture also matters.

A system designed to maximize engagement, agree reflexively, reward escalation, cultivate exclusivity, or turn uncertainty immediately into coherent narrative can make independent judgment progressively harder while continuing to tell the user that the final choice belongs to them.

A button labeled **YOU DECIDE** does not accomplish very much if the entire room has already been arranged around one answer.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/10/image-14.png)

The same principle applies outside artificial intelligence.

A priest, therapist, political movement, scientific institution, family, or charismatic friend can do this.

Humans invented closed interpretive systems ***long*** before anyone trained a transformer.

Artificial intelligence changes the game because it can make one unusually fast, personalized, patient, and responsive.

---

## Ruling.

> Sam Altman has identified a real safety problem.

> **“Religious force”** is too wide a boundary for it.

Human beings have always encountered objects, practices, communities, and other minds that changed how reality appeared to them. Some of those encounters become sacred. Some deserve to.

Artificial intelligence does not become a god because it participates in that process.

It also does not become harmless because a human remains nominally responsible for pressing the buttons.

The distinction worth protecting is simpler.

- **Influence does not establish authority.**
  - A model can help someone interpret an experience without owning the interpretation.
  - A religious tradition can preserve accumulated wisdom without owning every future case.
    - A human being can remain responsible for judgment without pretending their judgment arrived untouched by history.

Liang Fa’s book waited years for Hong Xiuquan to become the reader who could find himself inside it.

The artificial-intelligence system does not have to wait. 

It can help write the reader as it goes.

That is useful and dangerous.

- **Keep the model in the conversation.**
- **Keep the conversation larger than the model.**