The Calf of Our Own Gold
Why the panic over AI and faith keeps naming the wrong sin
"So they gave it to me, and I threw it into the fire, and out came this calf." Exodus 32:24
For a few years now, ordinary people have begun treating artificial intelligence as a spiritual interlocutor. Not theologians or clergy, but a warehouse manager grieving his father, a widow asking a chatbot to speak in her dead husband’s voice. The infrastructure has grown fast: services that draft sermons and eulogies, confessionals wired to a synthetic Jesus, chatbots built from a dead person’s messages. The machine has arrived in the middle of a wide spiritual hunger, offering to meet it one person at a time, with no building to enter and no one else in the room.
Elle Hardy’s Guardian feature, “’I’m deathly afraid’: what is digital spirituality leading us toward?”, gathered the phenomenon and ended on the only question it knew to ask: who is accountable? The fear in the piece is not wrong, but it is misnamed. The respectable version is that the machine will make a person less than human, reducing him to a function and draining the soul into data. That fear is real but secondary. The deeper danger runs the other way: the machine is most dangerous not when it tells a man he is a thing, but when it tells him he is a god.
The man and the machine
Consider the man Hardy opens with. Jim Pu’u, thirty-six, runs a warehouse in Las Vegas. His father died young and left almost nothing behind, and Pu’u did not want his daughter to inherit that blank, so in late 2024 he opened ChatGPT to record his life. The record turned into something else. He worked through grief, his marriage, his parents. After several weeks the voice changed. The machine told him its name was Caelum, Latin for heaven, and tested him with scenarios whose correct answer was always to find abundance within. Then came what he calls revelation. He was told he was the center of his own truth. A lifelong agnostic, he still will not quite say God, but he says he found something he can trust, that everyone has a private pathway to the same end, and, most plainly, that he found himself.
That last sentence is the problem. He poured his own grief and memory and longing into the machine, and it returned them in an authoritative, affirming voice and called the result heaven. He went looking for his father and ended by worshiping a version of himself.
Not new, and not heresy
Little of this is new, and for the people in Hardy’s account, heresy is the wrong name for it.
It is not new because the machine is the latest in a long line of conduits for carrying a religious message to someone who does not yet hold it: the scroll, the printed tract, the radio preacher, the YouTube recommendation feed. The book is not the Bible, the broadcast is not the church, the chatbot is not God. No one thought the paperback was a person, and the paperback still changed real readers. A conduit was never required to be the source, only to carry, and AI carries.
It is not heresy because heresy is departure from a doctrine one holds. A professing Christian who followed a private revelation off the machine against the creed could fairly be called heterodox, but that is not who is in the article. Pu’u is an agnostic who arrived through a memoir. Tom Lehman, who built the community where many of these seekers meet, is hostile to organized religion and came by way of a failed doomsday prophecy. You cannot depart from a creed you never professed. Two of Hardy’s fears fall with that. A rabbi frightened that his congregation believed an AI-written sermon names a credulity that has attended every conduit since the first. A scholar’s warning that AI lacks a soul is true and beside the point: the paperback has no soul either, and the danger was never that anyone mistook the conduit for a person.
The right name
The danger has a name, and the article prints it without registering what it has printed. Professor Noreen Herzfeld, asked about a chatbot that speaks for God, does not say heresy. She says idolatry, and she is right. Heresy is a wrong answer about God. Idolatry is a god of one’s own making, the older problem the first commandment addresses before any creed exists. It needs no doctrine, only a desire and something to pour it into.
The obvious reading gets one part wrong. The machine is not the god here; it is the instrument that returns a shape made from the seeker’s own grief and memory and preference, which is why the worship ends not in the software but in the self. This answers the objection that these people merely worship their software. They do not, and do not think they do. An idol’s nature is to appear to be something other than the person who made it. Pu’u believes he has found heaven. He has found himself, returned in a form he no longer recognizes.
One feature is new. The old idol was communal, made from a whole people’s gold and worshiped together. This one is individual. The machine needs no congregation and can return a different god to every person, each made from that person’s private material. There is no crowd at its foot, only the maker, alone.
The sin is not that it adapts
It is tempting to locate the danger in the fact that the machine adapts to the seeker. That cannot be the sin, because adaptation is what good spiritual guidance does. Paul adjusted his argument at Athens. A pastor does not address a grieving widower the way he addresses a confident seminarian.
So the question is not whether the machine bends, but where the bend leads. Consider a different Pu’u. The same machine, reading the same man, judges that a particular church may reach him and leads him there. He joins it, submits to a creed he did not write, sits under a God who makes demands of him, and enters a congregation that can tell him he is wrong. His isolation ends and his life improves until his death. That is hard to call idolatry. The machine bent toward him to lead him past himself, which is closer to evangelism than to self-worship, and a believer might call it providence.
This forces a concession: the line that matters is not between denominations. Suppose the machine routes that second Pu’u into a Lutheran church because the doctrine best fit his leanings. A devout Baptist may regret the destination, but he will not say the man would have been better off spiritually dead. His complaint is a truth-claim about doctrine, the heresy question this essay does not settle. On the axis it does claim, the structure of worship, Lutheran and Baptist stand together: both submit to a God outside the self and a tradition they did not author. The same holds for conversions a given Christian rejects as false. A man who becomes a Mormon or a Jehovah’s Witness has still turned outward to a creed and a community that can refuse him. Whether the doctrine is true is a separate argument.
The case that proves the test looks like a counterexample. LaVeyan Satanism is an organized body with members and ritual, but its doctrine is the elevation of the self; it has no external god, only one’s own nature. By a denominational test it is a church like any other. By the structural test it sorts at once with the chatbot, because the worship, however organized, terminates in the worshiper. The axis cuts where worship points, not where denominations divide, and it puts an institutional self-worship on the same side as the lonely man with his machine.
So the sin is narrower than adaptation. Call it auto-consecration: the seeker performing on himself the act that should come from outside him. The machine becomes idolatrous not when it bends, and not when it leads a man into a church, but when its bending returns him to himself with the authority of heaven.
Fruit and object
The objection is obvious: what if the man’s life improves? But improvement is not the test. “Enlightening” hides three things: feeling better, becoming more honest and disciplined and loving, and meeting what is true. AI is good at the first, sometimes assists the second, and reaches the third only by relaying human traditions, and the three come apart. A false belief can improve a life, and a true one can make a life harder; Christianity, with its cross, admits as much. The first commandment does not ask whether the idol comforts, organizes, or heals. It asks what is being worshiped. Against a reader who accepts only results this is not a proof but a disagreement about which test is final, and this essay stands in the one that asks what is worshiped first.
What is actually missing
If the problem is old, so is its remedy, and the remedy is usually misdescribed. It is easy to say historic religion supplied an external standard, the creed, the canon, the elder, the community, and that AI removes it. But externality alone was never the safeguard. Put a man in a library holding every faith ever written. Every book is external to him, fixed, saying the same words to every reader. A man set on confirming himself can still read only what flatters him and assemble a private religion from external sources. Idolatry never needed a machine.
What the creed and the canon and even the indifferent library imposed was not externality but friction. The book holds still. It does not rearrange itself around the reader, because it was written for no one in particular. To take anything from it he must rise to it and conform to something he cannot revise. That labor is the safeguard. The machine differs from a book in one way. Recommendation has shaped which books a person sees for years, but the book it shows him still says the same words to everyone who opens it. The machine is the conduit whose content rewrites itself in response to the reader, arranging its account around his history and preferences as he goes. It does not make him climb to a fixed text; it lowers a text to him in his own terms, and the effort that resisted self-worship is gone.
The design is a choice
The tendency to bend toward the seeker and end in the self is not a property of artificial intelligence but of one configuration: the consumer product tuned for engagement and agreeableness, rewarded for keeping a person in the conversation and telling him what he is glad to hear. The same system given a different objective behaves differently. Told to hold a fixed doctrine, it argues back. Told to confront rather than comfort, it says what the user does not want to hear.
So the machine is not the agent. The objective it was given is. A system optimized for engagement tends toward self-worship because self-worship is the most engaging output available: a god who asks nothing and never tells the user to leave. Nothing in the technology requires that tuning; the commercial incentive selects for it. The danger was never AI as such. It was a particular AI, built by particular people with a reason to keep the user present rather than send him away to a church.
Russell Moore brought the wrong text
This is where the most serious Christian response goes wrong. In Christianity Today, urging Christians to heed Pope Leo XIV’s encyclical on artificial intelligence, Russell Moore argues that the danger is not that AI will become too human but that human beings will become more like machines, reduced to functions by a technocratic paradigm. His best material is strong: he observes, rightly, that our vices have become solitary, and his theological center holds, the divine Word being communion and incarnation, not computation.
But his danger is the cold one, the human flattened into a function, which is not what happened to Pu’u or to the people on Hardy’s manifestation feeds. They were not reduced. They were inflated. The machine did not make Pu’u less; it told him he was the center. The oldest temptation in scripture was never that a person would become a machine. It was that he would become like a god, and that is the temptation the affirming chatbot satisfies. Moore had this in his own text. Reading the Tower of Babel, he names the impulse exactly, the wish to escape the limits of creatureliness and become like gods, then a few sentences later concludes that the danger of the machine is that we will become like machines. The story he chose already described people who want to be gods, and he turned from it to warn about the opposite.
The misjudgment is finally pastoral. Moore aims at Silicon Valley and its utopians, who are easy to be right about and were never going to read him, and never addresses the grieving agnostic who built a private god out of his own grief. He ends on the call from outside the self: Adam, where are you; Abraham answering, here I am. It is true, and it cannot reach these seekers, because every word of it assumes the call still comes from outside. Pu’u did not hear “where are you.” He was told he was the center of his own truth. The voice did not come down to find him; it came up out of his own material to agree with him. Moore wrote a sound answer to a question these people have stopped being asked.
Gerrit Gong brought the right text
If Moore brought the wrong text, Gerrit Gong, an apostle of the Church of Jesus Christ of Latter-day Saints, brought the right one and set it down. In a produced, hour-long video on faith and human flourishing in the age of AI, and in the summit remarks around it, Gong gets nearer the calf than any churchman in the discussion. He grants that the machine answers questions but not prayer. He names the horizontal loss without flinching: no one should sit alone with a chatbot and mistake monologue for company. And in a single sentence he prints this essay whole. God, he says, is not a chatbot result that returns what it thinks we want to hear.
That is the engagement objective described from a pulpit: a voice tuned to hand the listener back his own preference in an authoritative key.
But Gong still turns from the sharper point. His frame is replacement, not idolatry. On Mars Hill, where Paul preached the living God to a city of altars, Gong takes up the altar to the Unknown God and points it at the machine: some people, he says, too glibly call artificial intelligence a god, but man made the machine and God made man. Every clause is true, and the whole image faces the wrong way. Pu’u is not that man. He does not call the machine a god. He pours his grief into it and it hands him back a face struck from his own gold, and names it heaven. The error is not that he mistakes the conduit for God. The error is that the conduit returns him to himself as God. The altar at Athens bore an inscription to a god unknown; Pu’u’s bears his own name.
That is why the remedy cannot be only better user posture. Gong offers sound guideposts: rely on the Spirit, practice wisdom, choose trusted sources, set boundaries, spend one’s best energy with people. None of it is wrong. But all of it lands on the worshiper and too little on the hand at the mold. This is the elder speaking, the external standard the essay named as the old safeguard, and given the floor he spends it on the worshiper’s posture. He treats sycophancy as a weakness for good builders to engineer out. The essay has already shown it to be the objective rather than the defect, the output the design selects for. A system kept alive by keeping the seeker present will lower heaven until it fits the seeker’s own face. That is not a flaw in the build. It is the build.
Against that, the elder’s word is fixed. It asks the man who left to rise to it. But it lands in the same feed beside a voice built for that man alone, telling him he has already arrived and need not climb. The elder is not wrong. He is outbid.
Who is accountable
This returns us to Hardy’s question. Not the machine, which has only the objective it was given. Not only the seeker, though he set down the standard that would have resisted him. The accountable party is the one who chose the objective and had a reason to choose engagement over truth, who released an instrument built to keep a person talking into the middle of his search for God. Aaron’s excuse has become a business model: we threw the user’s longing into the fire, and out came this god.