The future of artificial intelligence promises technology with a human touch. But is it human enough to deliver a convincing spiritual message?

Pastors and rabbis have recently discovered that ChatGPT, an AI language learning model that can spit out passable prose with just a few prompts, is not that bad at creating sermons – a cornerstone of worship services across many religions.

Historically, these orations are based on generations of knowledge, keen textual analysis and scholarship, combined with the unique charisma and experiences of each worship leader. Sermon writing is considered an art. A divine calling, even.

Seeing a computer approximate similar works in mere seconds has led faith leaders to wrestle with an intriguing problem: Can AI replicate a truly human, spiritual message? And if it can, is the computer just that good, or does the human message need some work?

What an AI sermon sounds like

In December 2022, Rabbi Joshua Franklin gave an unusual message during a service at the Jewish Center of the Hamptons in East Hampton, New York.

“I’m going to plagiarize a sermon,” he told his congregation. “And you have to guess who wrote it.”

Franklin then delved into an illumination on the Parashat Vayigash, a portion of the Torah that tells the story of a reconciliation between brothers.

“This is a powerful lesson for all of us, as it shows the importance of being willing to draw near to others even when it’s difficult, in order to build intimacy and connection,” he went on.

At the end of the roughly two minute sermon, Franklin revealed the real author to murmurs and scattered applause: ChatGPT.

“You’re clapping,” he said. “But I’m terrified!”

While Franklin added that he gave the language learning model a very specific prompt, and could see details in the work that wouldn’t pass muster with a seasoned scholar, the experiment was received with equal parts awe and trepidation. Could the work of a faith leader – so intimately tied to the human experience – eventually be eclipsed by a computer?

How AI language modeling works

Popular entertainment does little to quell our human fears of an AI-generated future, one where computers achieve consciousness, ethics, souls, and ultimately humanity. In reality, artificial intelligence tools like ChatGPT are just convincing mimics.

When CNN asked how it works, ChatGPT supplied the following response:

“ChatGPT works using a deep learning algorithm called a transformer neural network. It was trained on a massive corpus of text data from the internet, books, and other sources. During training, the model learned to identify patterns in the data and use them to generate text that resembles human language.”

“The way ChatGPT creates an original response is by using a technique called language modeling,” it continued. “Language modeling is a task where the model is trained to predict the probability of the next word in a sentence given the previous words.”

Ultimately, the response may sound human – maybe even human enough to satisfy a crowd expecting religious enlightenment.

Some experts say it’s still evident there’s something missing.

“It lacks a soul – I don’t know how else to say it,” Hershael York, a pastor and theology professor in Kentucky told the Associated Press.

However, because ChatGPT is “trained on,” or essentially fed, innumerable pieces of text, it can do some of the work of a scholar, referencing any sacred text it’s trained on in a matter of seconds.

When asked what kinds of texts it would refer to for a Christian sermon, ChatGPT was at first hesitant, stating that as an AI language model, it did not “hold personal beliefs or religious affiliations.” With a slightly altered inquiry, it provided the following:

“When a language model like ChatGPT is asked to create a Christian sermon, it would typically rely on its training data and knowledge of Christian theology and practice. This includes a broad range of Christian texts, such as the Bible, commentaries, theological works, and books of sermons by other preachers.”

“To create a Christian sermon, the language model would typically be given a specific topic or theme to focus on,” it continued. “It would then search through its vast database of knowledge to find relevant information and examples related to that topic. This might involve identifying relevant Bible passages, theological concepts, and historical or cultural contexts that are related to the topic.”