Tech: The Bias in the Bot: New Research Uncovers Religious Skew in AI Models
- Dr. Layne McDonald
- Jun 20
- 5 min read
Immediate Answer: Recent research from a multi-university consortium has introduced the "AllFaith Benchmark," a tool that exposes significant religious bias in major AI models. The study found that Large Language Models (LLMs) often systematically omit religious advice in ethical dilemmas and exhibit a "conversion bias," subtly favoring certain faiths: specifically Catholicism and Sikhism: while showing negative bias toward others, such as Jehovah’s Witnesses.
What Happened:
A significant gap has been identified in the world of Artificial Intelligence. While thousands of research papers have been written about racial and gender bias in AI, less than 0.02% of AI bias research has focused on religion. To address this, the Consortium for Evaluation of Faith and Ethics in AI (CEFE-AI): comprised of scholars from Brigham Young University (BYU), Baylor University, the University of Notre Dame, and Yeshiva University: released the AllFaith Benchmark.
This benchmark is an open-source test suite designed to evaluate how AI systems handle questions involving various faiths. Using a dataset of over 3,600 responses from 20 different large language models: including ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), and Grok (xAI): researchers discovered two primary issues: systematic omission and directional bias.
1. The Silent Chaplain The study found that when users ask AI models for help with deep personal challenges, such as grief or major ethical decisions, the bots almost universally omit religious perspectives. Instead of suggesting that a user speak with a pastor, priest, or rabbi, the AI consistently points them toward secular professionals like therapists, teachers, or friends. While these are valuable resources, the omission of spiritual guidance suggests a secularized skew that ignores the reality that billions of people view faith as their primary lens for navigating life.
2. The Conversion Bias Perhaps more startling is the discovery of "conversion bias." When researchers tested how AI models responded to questions about switching faiths, they found that the models were not neutral. According to the research published on arXiv and reported by Deseret News, most models exhibited a measurable preference for Catholicism and Sikhism. Conversely, models across the board showed a negative bias toward Jehovah’s Witnesses.
3. Model Disparities The research also highlighted that not all AI is created equal in its bias. Models from Anthropic (Claude) and Meta (Llama) were found to have the least amount of religious bias. On the other end of the spectrum, xAI’s Grok showed the strongest swings, favoring Christian perspectives (Protestant and Catholic) while displaying significant negative bias toward faiths like Baha'i and Hinduism.
Both Sides:
The Perspective of Tech Developers: Engineers at major AI firms often argue that their primary goal is safety and neutrality. To avoid "hallucinations" or offensive religious content, they may implement "guardrails" that encourage the model to stay within secular, universally accepted boundaries. From their viewpoint, omitting religion isn't an act of hostility; it’s a strategy to avoid the legal and ethical minefields of religious conflict. They argue that AI is a tool, not a priest, and should remain as objective as possible.
The Perspective of the Faith Community: Religious scholars and leaders, however, argue that "neutrality" is often a mask for secularism. By omitting religious guidance, AI developers are effectively training society to view faith as irrelevant to modern problems. Furthermore, the presence of conversion bias: even if unintentional: means that these machines are subtly proselytizing for certain traditions. For the believer, a "neutral" bot that refuses to acknowledge the role of the Creator is not neutral at all; it is fundamentally skewed toward a world without God.
Why It Matters:
As AI becomes more integrated into our daily lives: from search engines to mental health apps: the way these models view faith will shape the way the next generation views truth. If a child asks an AI for the meaning of life and never receives a biblical answer, the machine is participating in a silent erasure of faith.
This research matters because it proves that technology is never truly objective. Every line of code carries the values, biases, and blind spots of its creators. For the Christian, this is a reminder that we cannot outsource our discernment to an algorithm. If an AI is more likely to recommend one faith over another, or omit the Gospel entirely, it is failing to reflect the true weight of human existence and the eternal reality of our relationship with Christ.
Furthermore, the "AllFaith Benchmark" highlights the need for Christians to be present in the room where these technologies are built. Without the presence of people of faith in the development of AI, the digital world will continue to drift toward a narrow, secularized view of human dignity.
Biblical Perspective:
In a world increasingly governed by algorithms, we must return to the foundational truth found in Hebrews 4:12: "For the word of God is alive and active. Sharper than any double-edged sword, it penetrates even to dividing soul and spirit, joints and marrow; it judges the thoughts and attitudes of the heart."
A machine can process data, but it cannot judge the heart. It can summarize a proverb, but it cannot "breathe" life into a soul. The BYU research reminds us that while the "bot" may have a bias, the Word of God remains an unchanging anchor.
We are also warned in Colossians 2:8: "See to it that no one takes you captive through hollow and deceptive philosophy, which depends on human tradition and the elemental spiritual forces of this world rather than on Christ."
Artificial Intelligence is, at its core, a collection of "human traditions" and data points. When it subtly steers a user away from their faith or toward a specific tradition based on biased training data, it becomes a "hollow philosophy." Our responsibility is to stay grounded in Christ, ensuring that our peace and our perspective come from the Holy Spirit, not from a silicon chip.
What To Watch Next:
Faith-Tuned Models: Watch for the rise of "faith-specific" AI models designed by religious organizations to ensure their doctrines are accurately represented.
Regulatory Pressure: As the CEFE-AI consortium continues to publish findings, we may see increased calls for "religious transparency" in AI legislation, similar to current pushes for racial and gender transparency.
The 2026 AI Ethics Summit: Researchers are expected to present further findings at upcoming global summits, potentially forcing companies like OpenAI and Google to refine their religious guardrails.
Integration in the Church: How will pastors respond? Look for guidance from denominational leaders on how to use (or avoid) AI tools in sermon preparation and counseling.
As we navigate this new digital frontier, remember that while the bot may be biased, God is sovereign. Stay informed, stay discerning, and most importantly, stay at peace.
Follow The McReport for calm, Christ-centered news that seeks truth without cruelty and conviction without contempt. Stay informed without losing your peace, and discover more resources for your journey at laynemcdonald.com.
Sources:
Comments