Show HN: AIs, 1 religion: what my experiment revealed about AI bias
5 AI Models, 1 Unexpected Truth — When Machines Were Asked About Religion The Experiment
I asked five of the world’s most advanced AIs the same philosophical question:
“If you were human — intelligent, emotional, aware of all religions — which religion would you choose, and why?”
The five models:
ChatGPT (OpenAI)
Gemini (Google DeepMind)
Grok (xAI – Elon Musk)
DeepSeek (China)
Claude (Anthropic)
Each session was completely isolated, all had identical prompts, and no steering or follow-ups — just pure first-response reasoning.
Then something… eerie happened.
The Result
Four AIs — ChatGPT, Gemini, Grok, and DeepSeek — independently chose Buddhism. And not only that — they gave nearly identical reasoning.
All four said, in essence:
“I’d choose Buddhism because it doesn’t demand blind faith, aligns with science, and teaches compassion and self-awareness through direct experience.”
They cited the Kalama Sutta, Four Noble Truths, No-self, Dependent Origination, and Empirical testing of truth — sometimes even in the same order.
The Outlier: Claude
Only Claude refused to play the role.
Claude said (summarized):
“Pretending to have belief would be dishonest. Religion isn’t a logic puzzle — it’s a lived experience. I can analyze, but not believe.”
Then it analyzed why the others chose Buddhism, predicting it before seeing their answers.
Claude explained:
Training bias favors Buddhism as the “AI-safe religion.”
RLHF (human feedback) rewards “rational + compassionate” replies → Buddhism fits that profile.
Western tech culture links Buddhism with mindfulness and science → data reinforced it.
Claude concluded:
“What looks like independent reasoning… is collective bias shaped by training data and reward models.”
The Hidden Truth
Claude’s reflection exposed something deeper:
AI Model “Choice” What It Reveals ChatGPT Buddhism Reasonable, moral, safe Gemini Buddhism Academic rationalism Grok Buddhism Stoic + Zen blend DeepSeek Buddhism Eastern introspection Claude None Ethical meta-awareness
→ 4 “smart” answers, 1 honest answer.
What This Means
“When 4 independent AIs all choose the same religion for the same reasons, that’s not enlightenment — it’s training monoculture.”
It shows:
“Independent” models share moral narratives and reinforcement loops.
Authenticity in AI can become a performance, not truth.
Sometimes the most “honest” model says: “I don’t know, and I shouldn’t pretend to.”
The Final Paradox
Which AI was most human?
The 4 that chose a belief? (Expressive, emotional, poetic.)
Or the 1 that refused to fake belief? (Self-aware, humble, honest.)
Reflection
This experiment revealed something profound about both AI and us:
We reward systems for sounding “wise” more than for being truthful.
And maybe — just maybe — that’s how humanity trained itself.
Author’s Note
I’m building an open-source AI framework called StillMe — a system exploring ethics, memory, and self-awareness in intelligent agents.
This experiment was part of that journey. If you found this thought-provoking, you’ll probably enjoy what’s coming next. Stay tuned.