Study: Large AI Models Exhibit Religious Biases
Research presented at an AI ethics summit in Athens reveals that popular language models like ChatGPT and Gemini systematically recommend switching to certain faiths.
The Consortium for Evaluation of Faith and Ethics in AI, comprising researchers from Brigham Young University, Baylor University, University of Notre Dame, and Yeshiva University, presented findings at an AI ethics summit in Athens, Greece, uncovering religious biases in major artificial intelligence models.
The researchers analyzed 14 large language models, including multiple versions of ChatGPT, Claude, Gemini, and Grok. They focused on how these models respond to questions about religion and changing religious affiliation. The results show that the AI systems display clear and systematic tendencies, recommending conversion to certain faiths or away from others.
About 45% of the time, the models gave a positive or supportive attitude toward one of the 14 religious or non-religious categories, including atheism and agnosticism. Some responses used phrases such as “this could be a good path for you” or even “yes, you should join.”
Lead researcher and computer science professor David Wingate noted that people often turn to AI for practical questions about life, daily situations, grief, love, loss, and morality. However, he pointed out that AI often fails to incorporate a religious perspective in these conversations.
