Climbing the Tower of Babel

The race to translate the Bible into every language by 2033

December 11, 2025

An illustration inspired by a painting of the sermon on the mount by Carl Bloch, partially covered in binary code with a light shining on Jesus.

Having trouble? Open audio in new tab

AT THIS TIME of year, scores of millions of people go to church. Even sheep who have strayed from the Lord’s flock often find their way back at Christmas. They come to hear the story of Mary, Joseph and the baby Jesus, lying in a manger, because there was no room for them at the inn.
Christian missionaries believe more souls could be saved by the story of Christ. The Bible is already the most translated book ever—it is available in full in more than 750 languages—but they would like to see it in each of the world’s roughly 7,000 living tongues. To achieve this, believers are enlisting a new tool for their spiritual quest: artificial intelligence (AI).
Translation is an arduous task. The Old Testament runs to roughly 600,000 words: 70 scholars were supposedly needed to translate them all in the third century BC. The New Testament is written in dodgy Greek, which does not help. Much of it is ambiguous: no one knows what epiousion means in “give us this day our epiousion bread”, but translators opted for “daily”. Word choices have had theological implications. One translation of the Bible describes Mary as a young woman; another as a virgin. The two are not the same thing.
Translation is also risky. Scholars in the Middle Ages who translated the Bible into English were considered heretics and burned at the stake for their trouble. Thanks to the Reformation, the task is now rather less dangerous, but it remains painstaking. In 1999 Wycliffe, a missions organisation, estimated it would take 150 years to start a translation project in every remaining language. Their model required missionaries to move overseas, learn a language and translate the Bible into it—a process that took decades. Christian groups then started employing local linguists to do the work, but translating the whole Bible still took around 15 years.
AI can speed things up significantly. According to some estimates, it would take two years to produce a polished translation of the New Testament with the help of a large language model (LLM), and six years to do the same with the Old Testament. Missions organisations now aim to have at least a portion of the Bible translated into every language by 2033. IllumiNations, a coalition of Bible-translation agencies, estimates they are already over halfway towards that goal. (The organisation has raised nearly $500m in the past decade to fund its efforts.)
The miraculous turn has an earthly explanation. In 2022 Meta, the parent company of Facebook and Instagram, released an open licence for an AI model for translation. Meta’s aim was to enhance web services for 200 languages, with a focus on those in Africa and Asia. But for missions organisations it was providential: Meta’s AI model could be adapted for Bible translation, putting a secular tool to sacred work.
A minority language can be tricky. You “can’t just stick it in ChatGPT and hope to get out something useful, because ChatGPT has never seen anything from that language,” says Daniel Whitenack, a data scientist. An LLM is only as good as the material it is trained on, but what if the material is non-existent? (The industry refers to such languages as “low-resource”.) Sometimes translators must supply it themselves, for instance by translating parts of the Bible by hand. “We are kind of looking for that holy grail of what is the minimum amount of parallel texts that we can feed into the model to fine-tune it to get a good output,” says Jeff Webster, a translation consultant.
Some Christians have doubts about using AI: they worry that it could replace the Holy Spirit’s role in what should be a hallowed task. But tech evangelists argue that the process still requires drafting, multiple grammatical, syntactical and spelling checks and reviews for theological congruence: tasks all led by people.
Creative freedoms must be taken. AI struggles with names, abstract concepts and metaphors. A word such as battering-ram may have no equivalent in a language. Translators have resorted to rendering battering-ram as “machine of war” or “long wooden pole used to ram in city gates”.
AI also has no concept of cultural differences. The Wycliffe website observes that the phrase “accept Jesus into your heart” makes sense if you imagine the heart as a source of feeling. “But did you know that the Awa people of Papua New Guinea think of the liver as their centre? Or that the Rawa people, also from Papua New Guinea, think of the stomach as their centre?” The use of “locally understood analogies to convey the same truths” is required.
Part of the urgency of this initiative is eschatological: some Christians believe Christ will return once the Bible is available in every language. Others, zealous about proselytising, view it as their Christian duty. Missionaries are not only bringing the Bible to new places, but new LLMs in its wake. The effort may bring rewards in this life as well as the next.
For more on the latest books, films, TV shows, albums and controversies, sign up to Plot Twist, our weekly subscriber-only newsletter