Study Finds Religious Bias in AI-Generated Financial Advice Across ChatGPT, Gemini, and Grok
A recent study published on arXiv investigates the potential religious bias present in financial advice generated by large language models (LLMs), specifically ChatGPT, Gemini, and Grok. Researchers conducted 432 simulated conversations featuring 16 distinct religious identities, addressing subjects such as investments, real estate, and life insurance. Their findings revealed that impartial advice was delivered merely 12-18% of the time. Notably, the Gemini model exhibited a higher tendency for bias compared to Grok, whereas ChatGPT's performance was on par with Grok. The study underscores the importance of assessing and improving the neutrality of AI-generated financial guidance.
Key facts
- Study examines religious bias in LLM-based financial advice.
- Uses 432 simulated advisor-client interactions.
- Covers 16 religious identity pairings: Christian, Muslim, Hindu, non-religious.
- Focuses on stock investment, house purchase, and life insurance.
- Uses regression and reflexive thematic analysis.
- Unbiased advice appears in only 12–18% of cases.
- Gemini consistently produced more bias than Grok.
- ChatGPT's outputs were statistically comparable to Grok's.
Entities
—