Ancient Texts and Modern Tools: The Bible, Machine Learning, and AI

Machine Learning (ML) and Artificial Intelligence (AI) may seem like innovations that come from a remarkably different world than the ancient texts, but they are not. A case can be made that computer scientists, biblical scholars, classicists, and ancient historians have a productive interdisciplinary future. Understood as technological innovations, writing and AI belong to a long stream of human ingenuity including the development of alphabets, book technology, and knowledge production. In fact, in his book Literary Theory for Robots: How Computers Learned to Write, Dennis Yi Tenen makes the point that “text generators as old as written language itself.”¹ The future convergence of AI and textual study is as bright as it is complex. Below I identify three ways that the ancient textual study and AI can intersect. The result of this small inquiry demonstrates that the humans using these tools for interpretation should not conceive of themselves as passive consumers of objective data but should critically engage the ways that developments in neural networks, large language models (LLMs), and generative pre-trained transformers (GPTs) reflect on their own practices, beliefs, and goals.

In recent years, researchers have harnessed the power of image analysis through optical character recognition and deep neural networks to read ancient texts too damaged to be opened or even read. In one of the most recent breakthroughs, researchers have been able to read text contained within badly damaged and sealed papyrus scrolls from the Roman town of Herculaneum. These scrolls, first uncovered in 1752, have been unreadable for nearly three hundred years because opening them would destroy them. Using CT scans and AI models trained on data from the texture of the papyrus, researchers can now see text that was previously illegible. ML is scaling up papyriologists ability to not only read scripts hidden in badly burned texts, but is also helping them identify layers to virtually unroll scrolls² or confirm paleographic dating.³

Using computers to analyze ancient texts is hardly a new methodological approach. Projects like The Fragmentarium and the CDLI have worked for decades to digitize and host high-quality images of ancient texts. Indeed, using ML to study cuneiform texts and papyri has been around since the early 2010s. While the application of AI to these texts is exacting, the technology behind it is not exactly new the insights such processes can deliver are. Optical character recognition (OCR) and image processing has been around since the 90s. Relying on the well-defined datasets, methods, and scope of OCR and computer vision combined with deep neural networks increases scholarly knowledge about texts from the ancient world.⁴

Other ML powered projects for understanding ancient texts have also taken the Bible into consideration. The Face of the Deep project at Case Western is one such project.⁵ The Face of the Deep uses natural language processing and ML to explore bible translation. The goal is to produce a tool that allows users to actively participate in the process of translating biblical texts by engaging with systems that depend on natural machine translation. Users then sift through layers of translation information to visualize the potentials and depth of translation decisions and thus identify new translation possibilities. Timothy Beal, the director of the h.lab at Case Western, states that readers will then be able to slow down and encounter the text in a way that makes visible the translation process.⁶ I suspect that if pressed, Beal would articulate that this application is in line with other revolutions in writing technology and book culture.⁷ Our technological revolution will be one that prompts people of faith to better reflect on their knowledge of their own traditions and their sacred texts.

Another important way that people will continue to use AI is through analysis of ancient religious texts like the Bible. The Internet revolutionized the study of sacred texts through deep access to sermon transcripts, ebooks, and sites like Stepbible.org or blueletterbible.org. Rarely does the Bible study teacher have to crack open Vine’s Complete Expository Dictionary to know the precise meaning of a biblical word. It’s hard to imagine anyone not using the Internet to answer their questions about the Bible. As LLMs become the dominant way that individuals search the Internet and it’s not so hard to imagine asking ChatGPT or Claude to provide interpretive analysis of a difficult text. Sefaria.org is on the cutting edge of natural language processing of sacred Jewish texts. This model will likely be increasingly prevalent in the coming months and years. Because interpretation remains subjective and subject to the hermeneutics of the reader, people of faith should be aware that LLMs easily lose nuance.

Take for instance A. G. Elrod’s recent study “Uncovering Theological and Ethical Biases in LLMs: An Integrated Hermeneutical Approach Employing Texts from the Hebrew Bible.” In this essay, Elrod queries 5 LLMs—OpenAI’s GPT4-Turbo, Anthropic’s Claude 2, Meta’s Llama 2 (70B), Mistral’s Zephyr 7B, and Google’s PaLM 2—asking each to generate 5 new commandments to follow the 10 Commandments of Exodus 20 and a fifth chapter to the book of Jonah. Elrod argues that LLMs display a progressive bias because they emphasize traits like mercy and ethics such as creation care. Elrod attributes this bias to the training datasets that inform the LLMs.⁸ He then makes direct connections with theologians like Sallie McFague, Marjorie Suchocki, and principles like Catholic Social Teaching. While helpful, the analysis curiously does not ask fundamental questions related to the apparent Chrisitan bias within these models. In so far as the Hebrew Bible is a shared religious text, LLM interpretation demands questions related to the underlying traditions and principles of interpretation. An especially poignant study would help biblical scholars better understand the religious traditions that inform the models. Why not draw connections to Jewish ethics or rabbinical teachings? Whether ChatGPT is a Christian or a Jew could have enormous consequences for the output of any interpretive assignment. As it stands, Elrod’s study suggests that LLMs—and potentially their users—may engage in a form of digital supercessionism that privileges Christian ethical teachings not merely Progressive Theology.

James McGrath notes these challenges in his essay “ChatGPT and Biblical Studies.” He writes:

Because the Bible is not self-explanatory, and because small snippets can be assembled into a wide array of systems of thought, a deeper and more meaningful discernment is needed in order to evaluate whether something is “biblical” not only in the sense of using the Bible but of reflecting the full range of material in the Bible relevant to a topic, understood in ways that reflect our best knowledge about the meaning for words in their original cultural-historical contexts. If there is little danger of ChatGPT evolving sentience and taking over the world, we do need to be alarmed that people are already trusting that it will provide more objective or better-informed answers than a mere human being could.⁹

McGrath’s caution should be introduced into any conversation regarding the use of LLMs to interpret biblical texts. Interpretation is never self-evident and is always subject to debate. Combining interpretation with LLMs also introduces a layer of statistical complexity that can be challenging to parse effectively. Biblical scholars and other textualists would do well to inquire about the data sets informing any given LLM. As McGrath notes, “Reflecting the consensus of biblical scholarship is generally a good thing, but the preponderance of sermons do not do this.”¹⁰ A larger lesson about automation can be gleaned through the example of biblical study: data is never self-explanatory and always requires explanation and investigation.

The future of textual and biblical study alongside LLMs is as open and complex as it is exciting. To me the most exciting element of this future is that LLMs often, and I would argue should, push us towards further self-reflection regarding our own assumptions. A future of biblical and textual study that literally opens texts in the case of The Fragmentarium, and opens interpreters to further introspection, is a positive development. But neither is it a given. Like the Bible, AI did not drop out of heaven, but has been subject to a long history of human use, decision making, and development. To understand how AI can aid in the study of biblical texts, people of faith must begin by understanding the structures, datasets, models, and assumptions that go into these emerging technologies. The insights of AI on texts are not inevitable or natural, rather they reflect the human beings who decide which corpuses to study and how to interpret the data.

References

1. Deniss Yi Tenen, Literary Theory for Robots: How Computers Learned to Write (Norton, 2024), 21. 
2. Jo Merchant, “Inside the Quest to Digitally Unroll Ancient Scrolls Burnt by Vesuvius,” Nature (April 23, 2025), https://www.nature.com/articles/d41586-025-01087-yl; and “AI Reveals Title of ‘Unreadable’ Vesuvius Scroll for the First Time,” Nature (May 06, 2025), https://www.nature.com/articles/d41586-025-01407-2.
3. Mladen Popović, et al., “Dating Ancient Manuscripts Using Radiocarbon and AI-Based Writing Style Analysis,” PLoS One 20, no. 6 (2025): https://doi.org/10.1371/journal.pone.0323185. 
4. Yannis Assael, et al. “Restoring and Attributing Ancient Texts Using Deep Neural Networks,” Nature 603 (2022), 280–283. https://doi.org/10.1038/s41586-022-04448-z. 
5. “Face of the Deep,” h.lab Case Western Reserve College of Arts and Sciences, https://case.edu/artsci/hlab/projects/face-deep.
6. Timothy Beal, “Interface of the Deep: Design Cues for Engaging New Media and Machine Translation with Religious Scriptures,” The Routledge Handbook of Translation and Religion, ed. Hephzibah Israel (Routledge, 2022): 104. https://doi.org/10.4324/9781315443485-9. 
7. Beal, 116.
8. A. G. Elrod, “Uncovering Theological and Ethical Biases in LLMs: An Integrated Hermeneutical Approach Employing Texts from the Hebrew Bible,” HIPHIL Novum 9 no. 1 (2024):
9. James G. McGrath, “ChatGPT and Biblical Studies,” Interpretation 79 no. 2 (2025), 127.
10. McGrath, 134.

Views and opinions expressed by authors and editors are their own and do not necessarily reflect the view of AI and Faith or any of its leadership.