Artificial Intelligence Evaluates Religions

The Results Might Shock You

My Most Recent Guest

My most recent guest on Brave New World was paleontologist Peter Ward, who is a Professor at the University of Washington, and the author of several books, including the bestseller Rare Earth, The Medea Hypothesis, and A New History of Life. Peter was a guest on Brave New World last year, but there was still a lot to talk about. So I caught up with him again in Seattle to continue where we left off in episode 76 of the podcast series.

I’m awed by frontier scientists like Peter who study the origins and evolution of life on Earth, measuring events at timescales of millions and billions of years, and backing up their findings using hard evidence from the Earth’s fossil record. Peter mentioned that he was inspired by the great paleontologist and evolutionary biologist Stephen J Gould, who was a New Yorker, and apparently lived a stone’s throw away from me in Greenwich Village during several decades of his life.

I have read a number of Peter’s books and I’ve become a huge fan. Talking to Peter is always both fascinating and deeply educational, and my most recent conversation with him on Brave New World was no different. You can check out the episode at:

https://bravenewpodcast.com/episodes/2025/05/08/episode-95-peter-ward-on-the-evolution-of-life/

Religion and AI

Towards the end of my conversation with Peter, I asked him, as someone who studies life on Earth using evidence that covers billions of years of history, how he thinks about religion. He pointed to a big stone Buddha behind me, which he said gives him peace. “Buddhism,” he said, “reveres life.”

Religion is a sensitive topic. It’s hard to discuss freely. People dig into their positions and ignore the uncomfortable truths. Some people are easily offended.

But now, with the advent of conversational AI, we can have “safe” and rational conversations about religion (although I do wonder how long it might take the AI to realize that we’re crazy).

I say rational because, in principle, the machine has been programmed by its designers to be neutral and non-offensive, especially on sensitive topics such as religion. And it has access to a lot of data, so it’s a great place to start a conversation in a relatively neutral place.

I thought it would be interesting to perform a “risk analysis” of each religion. I prompted ChatGPT by giving it as minimal prompt as I could to express my broad intention:

Me: “I'd like to do a ’risk analysis‘ of different religions. What I mean is, how susceptible might various religions each be for ‘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. I asked the bot for a set “risk factors.” It came up with a list of nine risk factors, all on its own, which it described like this:

I then asked the bot to assign a High/Medium/Low score to each major religion, based on how its descriptions applied to each of them. It presented the following table for thirteen religions across the nine risk factors.

I thought the output was reasonable, but another user might just as easily push back on items of disagreement. The bot tends to be very accommodative to disagreement.

My next move was, for me anyway, the pièce de résistance.

Applying the Analytical Hierarchy Process

I asked the bot to apply the Analytic Hierarchy Process (AHP) to the data. The AHP is an algorithm developed by Tom Saaty in the 1970s that compares each choice (in this case each religion) pairwise on each factor using a nine point scale, where a 9:1 means that one choice “very strongly” dominates the other on that factor, whereas a 1:1 means they are equivalent. For example, Catholicism is High on Centralized Authority whereas Hinduism is Low, so their ratio would be 9:1 on that factor, whereas Hinduism versus Buddhism would be 1:1 since they have the same levels of Centralized Authority. Medium versus Low would be a 5:1 ratio.

I asked ChatGPT to apply the AHP to calculate the overall “risk score” for each religion assuming all factors to be of equal importance. (Historical footnote: I was actually privileged to work with Tom when he was developing the AHP. As part of our work, I also ended up writing the original code that implemented the methodology.)

Here are the overall risk scores that the bot computed for each, using the AHP. 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 independently.. Higher scores are riskier (the totals add up to one, ignoring rounding error):

Was ChatGPT’s analysis questionable? Without doubt. But the larger benefit to the analysis is that for the first time ever, by using AI, I was able to ask these types of questions within an ostensibly objective, analytical, framework and able to do so interactively.

There is almost certainly some degree of subjectivity in how the AI compares various practices across the religions to come up with its levels, and in a more robust analysis, we might correct such errors. For example, perhaps the bot was overly harsh in rating Catholicism High on “Infallible scripture” since many Catholics don’t practice to the letter what the Pope or the Church say, especially on issues of birth control or abortion. The degree to which the practice differs from scripture matters. And, I arbitrarily assumed that each of the risk factors should have equal weight.

Taking a step back, though, this type of analysis is an interesting grounding exercise. AI enables a useful kind of hypothetical reasoning, in which the user can choose and adjust the inputs for a scoring algorithm. As is the case with any evaluation framework, the outcome is influenced by the criteria we choose for comparison, and the available data on each of them. But this is the only influence the user may exert.

Personally, I felt like I was talking to a human with deep expertise who had a rich storehouse of data, and with whom I could discuss and debate a highly sensitive topic that is very difficult to do rationally with most people. 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.

There’s a great line in the movie The Two Popes, in which the Pope comments to a Cardinal, “We all suffer from spiritual pride.” The scene is impactful because of the irony of the head of the Catholic Church suggesting that no one religion has a monopoly on the truth, especially in the absence of hard evidence about the existence of God.

The prominent author Deepak Chopra posits that AI should help enhance spiritual awareness. Chopra blends Eastern spirituality with Western wellness and medical research by promoting alternative medicine and mind-body healing, and often encourages readers and listeners to think about their internal intelligence, emotions and spirituality. Chopra holds that AI should increase human intelligence by enabling us to contact the deeper wisdom within ourselves.

Who woulda thunk it? Arguments are still raging about whether AI is really “intelligent” (however that is defined) let alone spiritual. Yet maybe the AI will help moderate religious extremism by enlightening religious zealots. This enlightenment might also go a long ways towards achieving world peace.

Amen AI.