Experience Stern | Faculty & Research

The recent Papal inauguration was business as usual: 133 old men electing a new leader. The new Pope now faces the challenge of balancing Catholic Church doctrine with the realities of the modern world. Many of its followers do not adhere to its teachings on birth control, abortion, homosexuality, and gender inequality. This disconnect is not unique to Catholicism. The world’s major religions are noticeably out of touch in the modern era of scientific progress, social equity, and globalization. But on top of their traditional challenges, religions must now contend with a new phenomenon that is very likely to change them, probably for better: Artificial Intelligence (AI). 

With its emphasis on data and rationality, AI is poised to mediate the centuries-old conflict between religion and science. By helping people better understand their own religion and those with which it is in conflict, AI could become a moderating influence on religions which are plagued by extremist behavior.

The magic of modern AI is that its large language models (LLMs) have learned a lot about everything, including the world’s major religions, by ingesting from the Internet whatever humanity has said about them through scriptures, sermons, scandals, and recorded history. For the very first time in the history of mankind, individuals can converse with AI about religion in a way that has been virtually impossible until now because believers tend to dig into their positions. And most religions find it difficult to accommodate differences, to confront their own uncomfortable truths, and perhaps most importantly, to control violent extremism that adherents justify in the name of religion. 

In contrast, AI provides a relatively unbiased interface to religion thanks to the heavy Reinforcement Learning Human Feedback (RLHF) that has gone into ensuring that chatbots stay neutral and non-offensive to all religious groups. This is an essential starting point for an impartial discussion on the subject. And its views on religion are grounded in volumes of data that transcend any single human’s capability to digest. Think of it as a new kind of oracle.

A related reason that people will engage with AI about religions is that the AI chatbot is flexible and open to entertaining challenges to its assumptions. For example, if someone suspects that its data are biased, the bot is accommodative to being challenged and corrected while it keeps the discussion going. 

Similarly, it is easy to discuss and compare religions with AI using any lens. For example, I ran an experiment on ChatGPT to evaluate religions through a “risk” lens. I started the conversation with a minimal prompt:

Me: I'd like to do a risk analysis of different religions. What I mean is, how susceptible might various religions each be to ‘going off the rails.’ Do you see where I'm coming from when I say risk-based analysis?

AI: Yes, I see exactly where you’re coming from. You’re proposing a risk-based framework to evaluate how susceptible different religions are to extremism, or “going off the rails”— much like how organizations do risk assessments for operational vulnerabilities. This kind of analysis would involve identifying factors that correlate with higher risk—not necessarily judging the truth of the religion, but assessing its structural or doctrinal vulnerability to misuse, radicalization, or abuse.

I thought this was a great start, so I asked the bot to come up with the set of “risk factors.”

It came up with a list of ten risk factors, all on its own:  Authority Structure, Infallible Scripture or Leader, Apocalyptic Worldview, Us-vs-Them Mentality, Suppression of Dissent, Obsession with Purity, Closed Community or Isolationism, Emphasis on Sacrifice, Precedents of Extremism, and Gender Hierarchy. It also explained why each of these are relevant risk factors.

Again, I thought this answer was reasonable, so I asked it to assign High/Medium/Low scores to each religion on each risk factor. It created the following table for thirteen religions across the ten risk factors.

I then asked the AI to apply a principled decision-making algorithm called the Analytical Hierarchy Process (AHP) to the data in order to generate an overall risk score for each religion. The method compares each pair of choices on each factor to calculate a score for each choice. Below are the overall risk scores that the bot computed for each of the religions used in my experiment, using the AHP (higher scores are riskier). It bears repeating that these rankings are computed entirely from the bot’s initial assessment of the levels of each of the risk factors for each religion. I expressed no opinions.
 

We can see that both major sects of Islam, Catholicism, and Eastern Orthodox rank high on risk, whereas Protestantism, Judaism, Buddhism, and Sikhism rank low, and Hinduism, Jainism, Taoism, Confucianism, and Shinto rank lower still on overall risk. 

Is ChatGPT’s analysis questionable? Without doubt. There is no single right answer. But the larger benefit to the analysis is that for the first time ever, I was able to ask these types of questions within an ostensibly objective, analytical, framework and able to do so interactively. That’s new.

Given the coincidental timing of my article “AI Evaluates Religions” and the election of the new Pope who might better align the church with modern practices, I received some fascinating critiques. 

Personally, in my conversation with the AI, I felt like I was talking to a human with deep expertise and a rich storehouse of data, and with whom I could discuss and safely debate a highly sensitive topic. The machine seemed to act as an impartial judge, following an objective process and doing so without a heavy predisposition bias, namely, the tendency to interpret new information in a way that aligns with pre-existing preferences, beliefs, or emotional leanings—regardless of objective evidence to the contrary. 

It is all but certain that future generations will spend a lot more time on their screens than at shrines. Because of its neutrality and inclusiveness that cuts across religions, AI could well become a moderating influence on religious extremism in the future digital era. That would go a long way towards contributing to world peace.
___

Vasant Dhar is a Professor at the NYU Stern School of Business and the Center for Data Science, host of the podcast Brave New World, and author of the forthcoming book Thinking With Machines.