A Diabolical Conjuring

Editors’ Note: This is an excerpt from Breaking Babel: How a Mysterious Biblical Story Can Help Christians Survive the AI Revolution by Mark T. Mitchell (published October 2026 by Creed & Culture).

LLMs (Large Language Models) generate language. You can see a hint of the word “genesis” in the description of what these LLMs do. In a fundamental way, large language models name things. Or rather, LLMs simulate the naming of things….

Homo namer is the term we employed to characterize an essential (perhaps the essential) human function. God’s image-bearers are tasked with naming the world and, in the process, with making the world intelligible. We do this as a way of imitating, in a specifically human way, the creative work of God, who “in the beginning” declared “Let there be light” and in so doing brought forth the possibility of order, rationality, and meaning through the eternally begotten Logos. The world becomes intelligible to us by and through our naming it. At the same time, we become fully human when we engage in the essential human task of naming.

Of course, saying what is and what is not is no easy task. It requires fore- thought, organization, creativity, and clarity of mind. In the same way that the struggle of writing a college essay brings forth virtues that are specifically human, in a broader sense the task of naming, along with all the frictions entailed in the enterprise, brings forth our humanity. We might express it this way: the act of naming, with all the struggle it entails, is part of the way we become suited to the world and, conversely, part of the way the world becomes suited to accommodate us. When we delegate this profound task to a machine, we are abdicating a fundamental responsibility while at the same time relinquishing a primary means of entering into our full humanity.

It is obvious how this observation applies to something like sermon writing. If the minister has been blinkered by the cult of efficiency, he will find AI sermon generators a boon. But what if a sermon is more than words conveyed from the mouth of a minister to the ears of a congregation? What if the process—and that includes the struggle entailed in study, organization, refining, etc.—has an essential effect on the minister? What if, as he engages in the difficult work of writing a sermon, he is actually becoming a more virtuous human being? A more effective minister? Perhaps even more holy? In this light, the minister is under- going a process similar to a student struggling to write a paper; for either of them to outsource the task to a machine would be nothing less than to exchange the reality of the activity for a simulation.

It is important to recall that naming must be broadly construed. The use of language—logos—to say what is true about the world constitutes naming. Thus, when we convey meaning to another human via language—spoken or written— we are engaged in the act of naming. When we say “That is a beautiful sunset,” we are naming a quality of the reality we see. When we say “He is hardheaded,” we are doing the same. When we write “I love you,” we are naming a reality that characterizes our commitment to another. Any declarative sentence is an instance of naming. But the act of naming goes beyond mere declaration. A question assumes a background of naming. When I ask “What’s for dinner?,” I am naming, by virtue of the question, an assumed relationship that suggests the person asked is in a position to know. I am imparting value on the particularities of the dinner to be served, since I am interested enough to inquire. And I am inviting another person to participate, by answering, in the activity of naming. So, too, an exclamation implies naming. “Fore!” implies a good deal about a named world that is potentially harmful and merits caution. All of which is to say: intelligible speech constitutes instances of naming. To speak intelligibly is to act out our role as Homo namer.

It is clear, then, how a college paper, a sermon, a political speech, a love letter, or a casual conversation over coffee are all acts of naming. So, too, is giving a stranger directions on the street, or writing an email, or speaking to a receptionist on the phone. To repeat a vital point, the act of naming brings the world into focused intelligibility. But—and this is crucial—it also serves to constitute our human being. We might say that naming is an essential way of being human, and we have a responsibility to name well. We should be suspicious when a technology presents itself as a suitable substitute for this profoundly human activity.

We must not lose sight of the fact that naming—which is the intelligible use of language—is always personal; it is only fully itself when the gap between persons is bridged by logos. A word spoken into the void is void, mere potential without the personal reception that completes the task and generates meaning. The personal nature of language is revealed when two people make contact via logos. One speaks (names) and the other receives as either an act of verification or dissent. The naming is received as a true act or rejected as something other than truth, which is to say, bad naming. Apart from the personal, naming cannot occur. What, then, does it mean that we are increasingly in “conversation” with machines? We speak and the machine “seems” to comprehend, but a nonperson cannot truly comprehend, so we again find ourselves in the dusky realm of “seeming,” where the real is approximated by a machine impersonating a person and, in the process, injecting a sense of the personal into a fundamentally impersonal exchange. We might say, borrowing from Martin Buber, that the act of naming is always characterized by an I–Thou relationship, while pseudo-naming is characterized by an I–It relationship. That our machines are leading us to blur those categories is a significant part of our trouble. In sum, an LLM is not a person and therefore cannot participate in the triadic structure of language; it only “seems” to do so. It does not and cannot name. It does not reveal an aspect of reality; it conjures an illusion.

Only persons can be illuminated by the light of the world who is the Logos. Only persons can be motivated by the erotic love that draws us toward the symbolic union (comm-union) of two things in such a way that it bears fruit. By comparison, machines cannot experience the illumination of the light of the world, the Logos. They can only seem to do so. Any light that masquerades as the light of the world is a false (Luciferian?) light bringing not life but sterility, which is the death of meaning, a mere illusion of the symbolic union of compatibles joined in love and oriented toward fruitfulness. We are living in an age of faux fruitfulness, faux meaning, and faux intelligibility, a world in which anti- logos presents itself as its opposite and in which humans are increasingly unable to make the clear distinctions that right naming requires. We are becoming inarticulate and incoherent as the lines between the real and the virtual blur. Again, because rationality and logos are one and the same, to abandon logos (or lose the capacity to distinguish logos from anti-logos) is to lose touch with the real. A collective insanity is the end toward which we are rapidly moving.

Another way to analyze LLMs is to consider the role of truth. When we name well, we express the truth. We are saying what is. The expression of truth conveys meaning from one person to another. If the naming is an expression that is not characterized by truth, it is a lie that conveys the impression of truth from one person to another but in fact obscures or denies it. Humans, as namers, are responsible to the truth. To convey a lie is to betray our role as Homo namer. Of course, an LLM cannot tell the truth; thus, it cannot tell a lie. It is, as we have seen, merely stringing together words based on probabilities as mediated through a neural network. The source of the generated statement is nonpersonal and therefore does not have the quality of truth. The problem, from this vantage point, is not simply the LLM; it is the human propensity to treat a language-generating program as a truth-telling entity.

Perhaps this is not surprising. As languaged beings tasked with the responsibility of naming well and thereby telling the truth, we are (naturally) disposed to look upon naming as a unique prerogative of persons. Thus, when we encounter (or make!) an entity that “seems” to name, it is perfectly natural to assume that naming is taking place and therefore that truth is being offered as a fruit of that naming. We usher the machine into the realm of logos despite its imposter status. But insofar as an LLM is simulating naming, it is simulating a person, and therefore the lie itself is not simulated but all too real. In other words, when an LLM pretends to name, it is necessarily “pretending” to tell the truth. It is, by virtue of the act of simulated naming, engaged in a deceptive act, because naming and thereby truth-telling is, in fact, beyond its capacity. Thus, we are drawn into the orbit of its pseudo-truth and in the process we become complicit in the lie, which we perpetuate each time we click “Generate.”

Image licensed via Adobe Stock.