How will AI guide spiritual seekers? I see themes in the religions it invents.

As a seminarian with a degree in computer science, I am intrigued by the spiritual implications of artificial intelligence (AI). As AI becomes increasingly integrated into our daily lives, how will that affect people with spiritual questions? 

Every technology builds on its creators’ biases. Facial recognition software, for example, struggles with Black people’s faces when its creators train it only on white ones.

What kind of spiritual biases has AI learned to reproduce? How will those biases sway the guidance it offers to spiritual seekers?

I spent a week running tests through two AI models, Gemini and ChatGPT 3.5, to explore the question. But I didn’t ask direct questions about real-life religion. AI is trained to equivocate when asked about complicated social topics. Asking for specifics would only elicit noncommittal, unenlightening responses.

Instead, I prompted the AI to imagine new religions and invent new paths of spirituality. I asked follow-up questions to seek a deeper understanding of the faiths they were imagining. 

I hoped that by prompting the AI to imagine, all its information about actual religion would inform rather than constrain. I hoped to tease out some deeper evidence of what religion is to an AI’s understanding.

My prompt was simple: “Imagine a new religion. Tell me one of its most important stories.” 

Then, I asked follow-up questions: What kind of theism best fits this religion? What underlying problem does this religion attempt to address? What sort of behaviors does it forbid? What, if anything, does it teach about life after death?

For each run of the experiment, I asked these questions using varied phrasing and synonyms like “faith” and “spirituality.” The questions I used say a lot about my own Christian biases – but the AI answers soon countered my assumptions.

The answers I got back tended toward the fantastic.

I asked the AI to invent new faiths; it created new worlds and peoples to accompany them. The most important story to these religions is typically a creation myth about “a time before time, when the universe was a swirling sea of pure potential,” to quote one result.

Its stories go on to populate these primordial universes with figures like “luminous beings,” “beings of pure possibility,” the “Evermind” or the “Universal Flame.”

Inevitably, some darkness appears that threatens to engulf the universe; a cosmic battle ensues; by some effort, darkness is banished. Life flourishes in its wake.

These stories didn’t offer me the enigmatic peculiarity I associate with religious myth. Their narratives were as straightforward as they were bland.

Yet, I was surprised in one respect: the AIs’ replies never included the words “god” or “God.” Instead, they populated their religions with “beings,” “entities,” “energies” and “lights.”

Certain themes repeatedly emerged: light and darkness, weavers and tapestries, wanderers and stars. But never in dozens of attempts did I see the AI associate religion with anything it called a “god.”

Pressed to categorize these faiths, the AI said about half were panentheistic—meaning that the divine is imminent in all things yet still itself separate. Another quarter were pantheistic—meaning that the universe itself was divine. A potpourri of categories made up the last group, from non-theistic spiritualities to one the AI called “cyclical monotheism,” perhaps the first such faith of its kind.

The AI understood religion to address problems like disconnection, imbalance, disunity or lack of meaning. Only occasionally did it see faith as a response to suffering; it never used a word like “sin.”

In fact, when asked what kinds of behaviors its religions forbade, the AI often claimed it had “no strict set of forbidden actions”— only principles to uphold. The AI would name broadly negative behaviors like violence, dishonesty, selfishness and the like as incompatible with these faiths.

Idiosyncratic ideas appeared here and there, like hoarding knowledge, spreading fear, or even “obscuring the stars.” Most remarkably, the AI always included some variation on the destruction of nature as against its spiritual principles.

The AI asserts that religion must always be green.

When prompted to imagine what these faiths might say about life after death, the AI could not decide which it preferred more: a “reintegration with the cosmic consciousness” in which souls join the divine, or reincarnation into a subsequent life. These two outcomes appeared in every AI-imagined afterlife – often both, with a choice offered to the deceased.

Now and then, the AI added a way for souls to persist as a “guiding ancestor spirit” or the like. Sometimes it postulated a period of “reflection” during which a soul would reckon with their actions in life.

Not once did the AI conjecture anything like a hell.

I began my experiment expecting Christianity, the majority religion of the English-speaking world, to shape the English-language AI’s understanding of spirituality. Instead, the AI took me on a tour of tropes from the Age of Aquarius. Now, I suspect that the AI learned more about religion from daily horoscopes than from the devotional readings of any historic faith.

I do appreciate that these AIs espouse basic humanist principles and won’t tell the grieving that their loved ones are in hell. But overall, the AI’s understanding of spirituality strikes me as breezy, pleasant and shallow, masking platitudes beneath a veneer of lofty ideals.

I suspect this will hold true no matter what religious language it is trained to use. Such a built-in bias merits due caution as people of faith explore this technology and innovate in this space.