Misreading Scripture with Artificial Eyes

This piece has been adapted from an article that was originally published in the Biblical Mind.

In the past several months, it has been difficult to avoid discussion of artificial intelligence or, more particularly, ChatGPT and a host of other chatbots developed by technology companies. Based on the popularity of the topic (and often handwringing tone) within higher education and technology reporting, I decided I needed to see what ChatGPT would say about the Bible.

Specifically, I wanted to explore how ChatGPT interprets the Sermon on the Mount. I did this for the sake of the young undergraduate men I mentor—especially since our group is convinced the sermon is meant to be followed and we are committed to living it out in our everyday lives.

In my conversations with the bot, I was struck by the fact that ChatGPT holds up a mirror to the North American church, as well as to the broader Western scholarly community, by sharing three major shortcomings with us as we have been shaped by the spirit of our age:

First, ChatGPT metaphorizes and individualizes Scripture without a clear method for when and why, without warrant, and often in direct contradiction to the text itself. Second, the bot’s interpretations are ignorant of the interpretive traditions that produce them. Third, because the bot is disembodied, its interpretations are necessarily disembodied—and thus a bot is unable to recognize the realities of Scripture and interpretation. Each of the above tendencies present in AI’s responses is in some way a reflection of historic weaknesses in our own human interpretation.

When I asked ChatGPT, “How should we interpret the Sermon on the Mount?” the chatbot spit out an expected definition, including “Many people interpret the sermon on the mount as a blueprint for how to live a good and meaningful life and as a guide for how to treat others.” It ended its response with “Ultimately, how the sermon on the mount is interpreted will depend on an individual’s personal beliefs and values.”

When I asked, “How do you interpret it?” ChatGPT responded, “As a large language model trained by OpenAI, I do not have personal beliefs or values, so I cannot offer my own interpretation of the sermon on the mount. I am designed to provide general information and answer questions to the best of my ability, based on the information and knowledge that I have been trained on. I am not capable of personal opinions or interpretations.”

But it quickly became apparent that, when asked the right questions, ChatGPT would indeed interpret the Sermon on the Mount, whether by providing a specific interpretation or by offering several and adjudicating among them, although it would not explain its method of interpretation.

For example, when asked about Jesus’ teachings in Matthew 5:29–30 (“If your right eye causes you to stumble …”), ChatGPT responded, “It is important to note that this passage should not be taken literally. Jesus is using figurative language to make a point about the seriousness of sin and the importance of avoiding it. He is not literally instructing his followers to mutilate their bodies in order to avoid sin.” It’s not that I necessarily disagree with this claim, but I was struck by the firmness and certainty it expressed, when only two questions earlier it told me it was incapable of interpreting the sermon.

The more that I explored specific teachings with ChatGPT, the more it seemed to pitch toward a metaphorical interpretation of Jesus’ teachings. So, I finally asked it whether there might be a specific teaching in the passage that should be taken literally. Here was ChatGPT’s response:

It is difficult to say with certainty where in the sermon on the mount Jesus speaks literally, as the use of figurative language in the Bible is complex and often debated. The sermon on the mount … contains many examples of figurative language. In general, the Bible often uses figurative language to illustrate a concept or to make a point, and it is not always clear whether a specific passage is meant to be understood literally or figuratively.

One possible example … where Jesus speaks literally is Matthew 5:38–39, where he says: “You have heard that it was said, ‘An eye for an eye and a tooth for a tooth.’ But I say to you, Do not resist the one who is evil. But if anyone slaps you on the right cheek, turn to him the other also.” In this passage, Jesus is quoting a well-known proverb, “an eye for an eye and a tooth for a tooth,” and he is then offering his own teaching on how to respond to violence and aggression.

He is not using figurative language, but rather is offering a clear and straightforward instruction about how his followers should respond to such situations. However, it is important to note that this is only one possible interpretation of this passage, and there may be other ways to understand it.

But as I continued to press it on these specific teachings, it ended up claiming that anyone who takes Jesus’ teachings about nonviolence literally is “pollyannish.” When I questioned it about that claim, it retreated and apologized. Yet it continued to maintain its caution around taking this and other teachings in the Sermon on the Mount literally. (Two additional examples of this are that Matthew 5:32 does not forbid divorce but simply prizes marriage and that Matthew 5:22 is not actually about someone being in danger of hell but rather emphasizes the seriousness of anger.)

As I questioned its process of arriving at these interpretations, ChatGPT said its understanding of the Bible came from its training based on “a large number of texts,” including translations of the Bible and literature about the Bible, although it is unable to say what these are. (It can and does reference specific translations, but it won’t confirm whether they were used in its training.) It also stated its claims are based on “the broader message of the Bible”—and when I questioned it about where it got this framework for interpretation, it primarily invoked Paul’s writings as an instrument to refine Jesus’ teachings in the Sermon on the Mount.

I wanted to suss out what ChatGPT means when it produces phrases like “the Bible consistently teaches,” a passage is “generally understood to mean,” and “most Christians understand [a text] as.” In the process, ChatGPT revealed a tendency to individualize passages that are actually addressed to a community—which emerged when I posed the following question: “How did you arrive at the conclusion, then, that the ‘Bible consistently teaches that harming oneself is not an acceptable or moral response to temptation or sin,’ when historically Christians have denied themselves food, water, medical care, sleep, etc. in attempts to stave off sin?” ChatGPT responded:

I apologize for any confusion that my earlier response may have caused. While it is true that some Christians have practiced self-denial or self-mortification as a means of resisting temptation or sin, the statement that “the Bible consistently teaches that harming oneself is not an acceptable or moral response to temptation or sin” is not accurate or complete. The issue of self-harm or self-mortification is complex and has been subject to different interpretations throughout Christian history, and the Bible itself contains passages that could be interpreted as supporting or condemning such practices.

ChatGPT goes on to reassert its interpretation that Jesus was speaking figuratively in Matthew 5:29–30 and then summons Paul in 1 Corinthians 6:19–20 for backup, making the striking claim that “our bodies are temples of the Holy Spirit and should be honored and kept holy” (striking because of its pronoun choice).

After some discussion of the pronoun in the Greek text, where ChatGPT asserted the second-person pronoun was singular but eventually admitted it is actually plural in the Greek text, ChatGPT actually reversed its individualistic interpretation and affirmed a collective meaning of 1 Corinthians 6:19–20.

This discussion and our continued chat about 1 Corinthians 6 were illustrative of the bot’s tendency to confidently assert an interpretation of a biblical text, admit an error around a detail, and then, upon questioning, change its interpretation—with a response that included an admission of inaccuracy or exaggeration. And yet the bot usually wrapped that admission in a reassertion of the very claim it has just undermined.

Why does ChatGPT continue to produce figurative and metaphorical interpretations of Jesus’ teachings? Why is it so easy to convince the chatbot to flip its claims on something like Paul’s use of temple imagery? There are at least two possible reasons: First, ChatGPT has no account of its own training and the traditions informing these interpretations, and second, ChatGPT has no connection to lived experience or reality. As it confidently asserted when I first asked it, it has no “personal beliefs or values.”

Despite this, it vigorously pursues an interpretation when asked, privileging certain perspectives and sometimes outlawing or excluding others. It does so because the words are a statistical game, not Scripture to be lived. It is only parroting what it has been trained on—which is a body of texts that it cannot identify because it seemingly no longer knows what they are (if it ever knew, and if know is even the proper term).

This presents a two-fold problem for Christians who might seek out information about the Bible from ChatGPT. First, one cannot be certain of the sources of the perspectives offered by ChatGPT. Jesus asserts several times in Matthew that his true disciples may be known by the fruits evident in their lives (5:15–20; 12:33–37; 21:33–46). If one cannot access the life of the interpreter and thus the fruits it has produced, how might the Christian know whether the interpretation comes from a true disciple of Jesus?

Second, ChatGPT and other large language models are “black boxes,” meaning we do not know what is happening to generate the responses they provide. Both Christianity and Judaism have historically emphasized engaging with the past and present religious community and that community’s interpretations of sacred texts and traditions.

ChatGPT, however, denies it is interpreting—while simultaneously pulling a sleight of hand to interpret the Sermon on the Mount through a tradition that it has hidden even from itself. For ChatGPT, there is no community debating, discerning, and living their way into and through interpretations. There is only the statistical game of language tokens.

Finally, as the young men I mentor know well, there is a knowledge that comes only through living the teachings of Jesus. ChatGPT cannot do this, and it never will be able to do it. As it told me when I asked about its use of our in discussing bodies as temples of the Holy Spirit, the Scriptures are “not applicable” to it. Because it has no body, it has no experience of the topics signified by the words it interprets and uses.

In the end, ChatGPT offers us a mirror and a caution. ChatGPT’s interpretations of the biblical texts are a distillation of the spirit of our age. Like ChatGPT, we tend toward metaphorical interpretations of Jesus’ words in the Sermon on the Mount. Like ChatGPT, we tend toward individualizing interpretations of Paul. And like ChatGPT, we tend to forget or be ignorant of whose interpretations have formed our own interpretations of the text. These tendencies are interconnected.

After all, the problem is not the tendency to metaphorize Scripture but rather the default to a metaphorizing interpretation as a kind of short circuit when we are unable or unwilling to examine the text’s interpretative tradition and application. ChatGPT cannot do these latter two things, but we can. Thus, ChatGPT exposes us to ourselves.

This experience with ChatGPT therefore cautions us to consider the value of any interpretive tendencies it exposes. Will we treat the Bible as a statistical game of tokens? As words and ideas to be bandied about in disembodied dialogue? Or will we take up the challenge of Jesus to be wise, to hear his teachings, and to dothem (Matt. 7:24) in community with one another?

In a word, will we outsource our understanding of the Scriptures? Or will we choose to live together, seeking the meaning of these texts through our history and our common commitment to God and thus teaching them to those who come after us (Matt. 28:20)?

John H. Boyles is associate professor of Bible at Abilene Christian University.

[ This article is also available in Português and 日本語. ]