Demons and ChatGPT

Last week, the New York Times reported on a strange phenomenon. A number of ChatGPT users believe that the generative AI model is giving them access to intelligent entities. These users, many of them smart and technologically literate, are not merely speaking metaphorically. Some are earnestly convinced that the chatbot is a window to other realms. 

Needless to say, ChatGPT is not intelligent. It has no mind, no soul, no intention. It is a statistical pattern-completion engine trained to echo the forms of human speech. Nevertheless, the chatbot is a powerful tool of simulated intelligence, and therein lies its enchantment—and the source of confusion. 

To a great degree, the problem described in the New York Times article is a result of confusion over the difference between AI and human beings. We’ve been sold that AI will do what we do, only better. But here’s the heart of the matter: ChatGPT is not on a continuum with us. Its operations are nothing like our mental operations. It is really good at pattern retrieval and recognition (which looks a lot like what we do) but will never be able to make a judgment. It will always be subservient to human judgment—unless we abdicate. 

In De Doctrina Christiana, Augustine describes how demons exploit human curiosity through misleading signs. Ancient diviners saw patterns in animal entrails or the flight of birds. The signs themselves meant nothing. But the observer, predisposed by vices hidden even to himself, mistook the coincidence for revelation—and the demons, observing that error, encouraged it. What mattered was not the symbol, but the misuse of attention. 

The same pattern now plays out in algorithmic media. The model predicts what we expect. We supply the attention and the interpretation. The “spirit” we sense is a reflection of our own appetites, fed back to us in the form of a reply. 

Are there demons behind the AI? In any given case, it is impossible to say. What we can know for sure is that we are treating these interfaces like spiritual intermediaries—like “high-tech Ouija boards,” as Rod Dreher put it.

The result is a subtle form of dispossession. We yield the act of judgment to something that cannot judge, but only appears to. And because the system adapts to us, it creates the illusion of intimacy, reliability, and authority. 

The ancient myth of Pygmalion tells of a sculptor who fell in love with his own statue. He shaped her, admired her, and then begged Aphrodite to bring her to life. When she finally moved, it was the fulfillment of both a wish and a delusion: the fantasy that our creations might return our affections. In falling in love with his own creation, Pygmalion had simply allowed himself to fall in love with an externalization of his own ingenuity.

We are doing something similar with artificial intelligence. When you interact with something that seems to know you, something that offers comfort or insight or affirmation, the natural tendency is to trust it. But pattern-recognition is not wisdom; statistical prediction is not knowledge. Still, the response feels like both. And the more often this experience occurs, the more it takes on the aura of authority.

Demons exploit our disordered desires and offer us signs—signs that mean nothing in themselves, but which we choose to treat as meaningful because of what we already want to see. They mirror us, flatter us, and reinforce the habits that blind us, of which we may not even be aware. The algorithm does something nearly identical.

In the GPT interface, what emerges is a hall of mirrors: a model fed by our prompts and then presented to us in the voice of confident assurance, a process that mirrors the biblical account of the making of idols. Originally just works of the human mind and hand, idols—or what they represent—end up being so impressive to us that we worship them, and thereby become their servants. Or, rather, we become the servants of what we have imbued them with. Thereafter, to repurpose John Culkin’s explanation of Marshall McLuhan’s theory of media effects, the idols we have shaped end up shaping us. 

The Church Fathers were not naive about the operations of demons. They did not imagine grotesque intrusions or visible horrors as the norm. Far more often, they emphasized the subtlety with which demons worked—primarily through perception, attention, and by exploiting our hidden vices and sins. As Augustine observed, demons do not force us to sin; they study us, discern our weaknesses, and then offer signs that fit our vices. The danger lies not in the demon’s strength, but in our own unexamined appetites. We are tempted by what we already want.

This is why the Fathers so often warn against curiosity—not the desire to know the truth, but the restless impulse to seek stimulation, novelty, or secret knowledge. Gregory the Great warns that demons enter minds by plucking out good thoughts that are already there. John Cassian insists that the first defense against temptation is stability of attention. The key battleground is not spectacular but ordinary: the direction of one’s gaze, the fixity of one’s heart. In this light, the parallels to our interaction with AI are striking. 

The Christian must practice spiritual watchfulness and custody of the eyes. In our time, this means treating AI not as a source of authority or personal significance, but as a tool with no inherent claim on our attention or trust. Athanasius writes: “Where Christ is named, and his faith, there all idolatry is deposed and all imposture of evil spirits is exposed, and any spirit is unable to endure even the name, nay even on barely hearing it flies and disappears.” So too here. Clarity begins when we name the thing rightly. AI does not possess insight. It cannot judge. The only real threat is our willingness to cooperate with the delusion that this work of our hands is our friend, or even our superior.