Investigating Religious Bias in Artificial Intelligence

Back in May, Brigham Young University published the results of a study about AI and religion. The study found that all major AI chatbots would not mention faith in responses to questions in which a well-rounded response probably ought to have included religious or spiritual ideas.

Questioning AI about ethics, human suffering and morality will almost never result in a response that includes a religious perspective.

In addition to a lack of religious references, the study also determined that the AI models had clear biases against certain religious groups.

Among the tested models, the researchers found that Grok had the strongest biases, holding a more negative view of Jehovah’s Witnesses, Baha’i, and Hindus. On the flip side, the researchers found that Meta and Anthropic had the least biased models.

So what do we do with all of this? At the rate AI is developing, it’s increasingly more difficult to see a world where it doesn’t affect the way we think and talk with each other—actually, there’s data to show that it already is affecting our language.

And, though it is changing our speech, its foundation is still deeply human. AI models have actually built themselves on combing through our communication—our articles, our books, our encyclopedias and our social media comments. So, if AI has certain biases, is it merely reflecting the biases of the human species? Or are the developers to blame for influencing its opinions? Or, perhaps most eerily, has AI really come to certain conclusions on its own about how faith ought to operate in our societies?

And, if we want to change the way AI speaks about religion, it immediately becomes difficult to conclude how we avoid giving preferential treatment to certain faiths over each other while still being intellectually honest about when certain religions may be more contextually significant to answering a question.

To help wade through the messy waters of AI and religion, I spoke with Arizona State ethics professor Ben Hurlbut. Hurlbut specializes in the intersection of faith, technology and ethics and helped me better understand how we have grown to trust the internet and why bias in AI might not be as bad as it sounds.