Robot ‘Church Fathers’ Might Curate New Canons

The pace of adoption for artificial intelligence is unprecedented.

By the end of January, ChatGPT—an AI chatbot that generates brand original content when prompted—had logged 100 million visitors to its site. Before that, it reached 1 million users in the first five days after its release in late November. By comparison, Instagram took 2.5 months to reach 1 million users, and Facebook 10 months.

Generative AI systems like ChatGPT, which can produce humanlike responses to users’ prompts, will undoubtedly shape how we, as the Book of Common Prayer puts it, “read, mark, learn, and inwardly digest” the Scriptures. Indeed, there are already multiple tailor-made AI-driven chatbot systems being used for Bible engagement—which I’ve dubbed “BibleGPTs”—including IlluminateBible.com, SiliconScripture.org, Bible.ai, and OpenBible.info’s AI-assisted Bible Study.

As a digital theology expert, I believe these kinds of “BibleGPTs” will continue to advance, proliferate, and eventually become proprietary systems. And as this happens, the church and its leaders will be prompted to make some momentous decisions about the Christian canon. This will, in turn, influence how we interpret the Bible and impact the future of our faith and practice.

AI-led Scripture engagement will generate new problems.

First, BibleGPTs could reify what I call “concentric canons.” Their databases will require us to precisely define what writings are included in our Christian traditions.

Scripture is our primary canon. When handwritten scrolls became recognized as distinct books, the early church fathers confirmed which writings were considered canon and which were not.

Likewise, with today’s AI databases, someone will soon have to decide which denominational writings and perspectives should be included in the training data of the GPT (which stands for Generative Pre-trained Transformer). And in some sense, these can act as secondary canons—much like the concentric growth rings of a tree trunk.

Imagine a Reformed database, an Orthodox database, and an Anabaptist or Catholic database: Each would require someone to decide which doctrinal writings should be included (“canonized”) and which should be left out. Who will decide, and how? Will it be denominational leaders within these traditions? Will it be tech leaders? Or possibly book publishers who control intellectual property?

The answers to these questions matter, and there’s bound to be controversy surrounding these decisions.

And while the input logic of these databases may be transparent to their creators, end users will likely have little to no visibility or understanding about which writings qualify as canon and why. In the past, extrabiblical writings like the Apocrypha and the Gospel of Thomas have raised such questions, but these new databases will multiply these concerns exponentially.

Second, BibleGPTs could create canonical mashups, because it’s easier for users to blend various Christian traditions and denominations together.

What would John Calvin think about megachurch governance? How would Mother Teresa respond to prosperity preaching? What would Martin Luther say about Martin Luther King Jr.? Questions like these—while intriguing—may or may not have clear answers based on church tradition. And yet these AI systems make it easier to ask such hypothetical questions, and therefore more likely.

The result could be Christians arguing over speculative theology rather than seeking to deeply understand the hard-won historical perspectives and traditions that a BibleGPT claims to represent.

Our traditions have built incredible creeds and confessions—architectural gems that house meaningful differences and fruitful diversity. Yet BibleGPTs risk razing these edifices and paving over them with a flattened understanding of faith that anyone can skim over.

This is why it’s important for AI designers to build into these systems a firm grasp of respective Christian church traditions, so that they do not cause “canon confusion” for the user.

Third, BibleGPTs can make it easier for users to ask “culture-bound” questions that the Bible doesn’t directly address.

What does the Book of Leviticus say about artificial intelligence? Is the Bible Democrat or Republican? Or, more seriously, what does the Bible say about eating disorders or child abuse?

Questions like these relate to culture-bound issues that are far removed from the Bible’s context. Such questions may be irrelevant or simply asking something that is outside the Bible’s sphere. We might extrapolate possible answers from Scripture—this is the work of theologians—but it’s important for users to know the difference between authors’ original intentions and contemporary applications.

Culture-bound questions are a user-driven problem. Nonetheless, kingdom-minded AI designers will want to account for this tendency when building any BibleGPT. Otherwise, the ease of asking these kinds of hypothetical questions could easily distract users from more worthwhile Bible engagement—and from understanding what questions are relevant and appropriate.

In response, BibleGPTs risk “hallucinating” heresies.

When we ask a BibleGPT questions the Bible doesn’t directly address, it may produce confident-sounding answers—whether they correspond to biblically orthodox beliefs or not. In fact, the BibleGPT might manufacture an entirely nonfactual response or “hallucinate” a heretical statement. And from these hallucinations, users can end up with a misleading or warped theology.

Experimenting with speculative questions may generate interesting responses that make decent theological arguments—but the user cannot be certain of the logic the GPT uses to reach its conclusions. Only when GPT systems are evaluated on hundreds or thousands of instances, not just a handful, do their biases begin to appear.

Ultimately, bad questions will lead to bad answers. And while many of these issues already happen in online queries, GPTs will amplify them and make it easier to retrieve inaccurate responses.

One hopes that GPT users will be discerning, but my guess is that pastors know just how likely some Christians are to asking questions like this. And if BibleGPTs ever reach even a fraction of YouVersion’s hundreds of millions of users, these edge cases seem inevitable.

Fourth, the loudest traditions will win. Because GPTs are based on statistical probabilities, the most probable theology will be overrepresented.

Whichever writings or denominational traditions the GPT databases include more of are likely to get the most visibility, as more prolific traditions pull the probability in their direction. This means that the loudest traditions will dominate the discussion. The prolific Presbyterians will win over the quiet Quakers. The popularity contest that already exists online will get reified in our databases.

When people warn about “AI bias,” this is partly what they’re talking about. This is why leaders of various Christian traditions will need to deeply consider which writers and whose writings are included or omitted in any BibleGPT database. It’s a question of representation.

But there’s a potential upside to this, too. BibleGPTs could keep us rooted more firmly in the broad historical orthodoxy by providing us with the best answers to biblical questions—the ones most frequently cited and best defended over the centuries—rather than serving up responses from fringe theologies.

Fifth and finally, AI users risk offloading Bible reading. On-demand answers may replace our efforts to engage the Bible and wrestle with what is written.

There’s something important about reading the Bible itself—and for ourselves. Something in the way it draws us into the story and invites us to face questions about who we really are. GPTs may provoke a Google reflex, where we instinctively search for an answer before wrestling with the question. This Google mindset assumes that access to Scripture is the same as knowledge of Scripture.

Pre-digested Bible content generated by GPTs isn’t a direct path to spiritual formation. We must remember that the ultimate purpose of reading the Bible is to encounter God, to be transformed by that encounter, and to be equipped to participate in what God is doing in the church and the world.

Christians can participate in solving these problems.

BibleGPTs may very well have a place in God’s kingdom, but wherever they oppose the purposes of Scripture itself, we must rethink their design and use. How can these systems be shaped to serve us well?

First, reverse the interrogation. GPTs serve Bible readers better by generating more questions than answers. We must let BibleGPTs interrogate us.

As Christians, we believe it is God who speaks first—not the internet, and not us. Bible scholar Scot McKnight writes, “We may argue with God and the Bible and we may ask questions, but that all comes after we listen.” God feeds us with his Word. Eugene Peterson takes a page from the prophet Ezekiel and the apostle John when he says, “Eat this book.” The Bible is food that energizes us and equips us for the good works he planned for us (Eph. 2:10).

One way I’ve reversed the interrogation with a GPT is to ask it what questions or issues a given Scripture passage asks me to wrestle with. The GPT has helpfully captured some of the key challenges that the passage presents. These questions spark my curiosity and invite me deeper into the passage, to read it more closely and to consider what it says and how I might need to respond.

Second, use AI as a supplement, not a replacement for the Scriptures themselves.

Reversing the interrogation is a great practice that BibleGPT users will benefit from. But they must choose to do that because GPTs won’t default to it. Reversing the interrogation keeps readers in the loop—in their role of Bible engagement, not Bible replacement.

This habit also keeps GPTs in the role of Bible supplement, not Bible substitute. For Christians who truly desire to encounter God and be transformed by him, we need to pay careful attention to who and what is mediating our interaction with Scripture.

Third, BibleGPTs should prompt users to practice more holistic Bible engagement.

Remember those old VCRs that always flashed 12:00? Few ever bothered to reset the clock after the power went out. The default settings of any technology will powerfully influence our choices and habits, and we often do what’s easiest. It’s why every app’s first question is “Allow push notifications?”—because app makers know that defaults can nudge users into the habits they want.

But defaults can prompt positive habits as well. Just one example is a GPT system that defaults to generating questions instead of answers, which can nudge users toward deeper Bible reading.

How else could defaults help users to move past culture-bound questions toward a healthier diet of Bible reading? Perhaps defaults might incorporate a notable and reputable Christian thinker’s systematic theological framework or schema for biblical interpretation? Either way, ideal default settings should guide users toward a more holistic experience of Scripture and the church.

Fourth, what I call “conscientious canons” could help broaden readers’ horizons.

BibleGPTs could broaden exposure to Christianity’s diverse history. We can work toward that by being intentional about how we gather and include writings from various Christian traditions.

If “concentric canons” reify Christian traditions into fixed databases, how might GPT systems alert its users about the traditions it represents? How could it create an awareness of various views or highlight the traditions or sources the ideas come from? Maybe it constantly reminds users, “Being the Quaker BibleGPT that I am …”

Fairly representing denominational traditions is vital. Deciding which preachers, which writers, which perspectives get included—and left out—deserves great scrutiny and care. Any BibleGPT should gather input from a broad range of voices from Christian history to speak into theological questions. These voices should go well beyond evangelicalism and Protestantism.

This is not merely a matter of denominational politics, but of how we as Christians love our neighbors and our enemies as Jesus taught.

Fifth, BibleGPT systems should be aligned to the purposes of Bible reading.

A BibleGPT will not rise to the intentions of the designers—it will fall to the habits of the users. For that reason, Christian AI designers must anticipate how users will engage with a BibleGPT. Nudges and defaults are important; while any system must give users freedom, it can also encourage them toward a more holistic experience of reading the Bible.

The systems and defaults we program into BibleGPTs must be trained for the purposes of Scripture reading. Careless BibleGPTs can lose valuable insights and critical context in translation, but conscientious ones can deepen the significance and comprehension of Scripture for its users.

The GPT training data sets have already absorbed plenty of Christian-oriented content, as well as plenty of toxic content as well. In her recent PhD dissertation, “Righteous AI,” Gretchen Huizinga writes that “crowd-sourced wisdom does not reflect divine wisdom. It gives equal weight to the wise and the foolish and ignores the absolute and transcendent.”

This means that, for Christians, discernment will be vital.

Innovative Christians have an opportunity to create BibleGPTs that will make our Scripture diet healthier and more holistic. But this requires an intentional effort—in our design as much as our doctrine, and in our strategy as much as in our theology.

In 1943, Winston Churchill said, “We shape our buildings and afterwards our buildings shape us.” Likewise, as we shape our AI systems, they will shape us for years to come. The landscape of Bible engagement is indeed already changing—the skylines are not what they were even a year ago.

If we don’t think proactively about BibleGPTs, we will reap the consequences. But if we are clear and conscientious in how we design them, the opportunities are incredible. The best designed BibleGPTs will do what the Bible itself does: encourage and enable Christians to connect with God in ways that transform them and equip them for mission.

This is why Christians must approach BibleGPTs with extraordinary care and a global vision—the future of Bible reading depends on it.

Adam Graber is a consultant in digital theology and cohosts the Device & Virtue podcast.

[ This article is also available in Français. ]