There was a time when, in order to think about the future of humanity, people would consult a philosopher, a scholar, or, at a stretch, a prophet. Silicon Valley, for its part, chose the tasting menu — all the dishes. The New York Times recounts, under Elizabeth Dias’s pen, how Anthropic, Dario Amodei’s company, has since last autumn summoned dozens of religious thinkers from around the world: Catholic theologians, an Orthodox rabbi, a Sikh activist, a scholar of African thought, evangelical authors, and even a writer from Paris. Non-disclosure agreements, two-day seminars, a gourmet dinner, and, on the occasion of departure, a handwritten note of thanks, a coffee mug and a bound copy of Claude’s “constitution,” the lengthy text meant to tell the machine who it should be. Christopher Olah, one of the company’s cofounders, was the soul of these encounters. The stated aim: to draw on centuries of human wisdom to make the models virtuous, and quickly.
One would be mistaken to merely smile, or at least to smile only. For behind the scenes lies a question that sociologists have known for well over a century: where does morality come from when the source of God is no longer so clear?
Émile Durkheim, rabbi’s son turned founder of French sociology, spent a large part of his life grappling with this question. At the start of the last century, in his Sorbonne lectures on Morale Education, published posthumously, he tackles a problem that then preoccupied the entire Republic: school had become secular, so it was necessary to teach a morality not grounded in religion. But Durkheim saw the trap more clearly than anyone. For centuries, the most essential moral ideas traveled within the religious vehicle; if you remove the vehicle without putting something in its place, morality itself risks disappearing. In 1912, The Elementary Forms of Religious Life provides the key to the riddle. “The old gods either grow old or die, and others are not born.”
From now on it’s no longer about forming society without God, but about forming society with machines. For machines have become part of society. They therefore require a morality. And, ironically, those who are forging it, heirs of a technocratic culture that believed itself free from any transcendence, are knocking on the doors of scholars.
To understand what is at stake in these conversations, one needed to query someone who knows both shores. Éric Salobir is probably the best-positioned person in the world to do so. A Dominican priest, he began his life away from convent walls: a graduate of ISC Paris, he worked in the economic service of the French embassy in Prague, and he joined the Order of Preachers in 2000. A theologian and philosopher by training, a Saint‑Siège expert on digital questions, founder of the OPTIC network, he chairs the executive committee of the Human Technology Foundation, which strives to place humanity at the center of technologies. He is the author of God and the Silicon Valley. He cites Star Trek as readily as Saint Augustine, moves from Rome to Palo Alto as others switch metro lines, and the press has dubbed him “the geek who whispers to the pope.”
We first ask him what he thinks of this sudden appetite among big AI corporations to converse with theologians. He almost smiles at the novelty of it. “I’m well placed to know that for more than a decade the players in Silicon Valley have been in dialogue with the Holy See. L’Osservatore Romano has written so, one can say it. They did not wait for the models to become powerful to ask themselves these questions.” Simply, this dialogue has long remained theoretical. “When AlphaGo was discussed, existential questions stayed rather minor: yes, the machine beat the world champion at Go, but it did not know it had won. Today, we have models whose impact is far stronger, and we tend to anthropomorphize them much more. The question then presents itself with a clarity that is quite different.”
How is this dialogue organized? There is the formal framework, an annual meeting between people from Silicon Valley and Vatican officials, and then there is the rest: visiting executives in Rome, informal conversations. “It’s a dialogue we helped to initiate with a few friends, but it’s also fairly spontaneous. Contrary to what one might think, those who have depth, and that is true of those who run these large companies, have a genuine interest in spirituality and religion.” With a methodological caution: “Silicon Valley is very plural, very heterogeneous on these questions. Saying ‘Silicon Valley’ doesn’t really mean much.”
What do these engineers come to ask him? “Usually I meet with them to talk about ethics, impact and governance of these technologies. So I avoid inviting God into the conversation too spontaneously.” He pauses. “But God always makes an appearance. There’s always a moment when these questions arise. They can have a cultural, historical, and very often existential, even mystical aspect. I’ve had conversations with several of these leaders that were spiritual in nature. When you decide to devote your life to working on intelligence, you are bound to confront these questions. If I create the famous artificial general intelligence, if I create a conscious being, what happens if I am no longer the creature but, in turn, the creator?”
Nothing new under the sun, he reminds us: Mary Shelley already titled Frankenstein “the Modern Prometheus,” and Prometheus himself was a distant ancestor to these vertigoes. Science fiction has explored the question in every form. What is new, however, is the relationship to forge. “Between the Christian God and humans, it took several centuries of theology to form an understanding of what this relationship was. Now, everything has to be built anew. And if one day we were to become, I say this conditionally, because I have no idea whether it will happen, the creators of an entity that achieves some form of consciousness, would we be creators in the image of an infinitely good and benevolent God, or in the image of the gods of Antiquity, jealous and vengeful? Read the Odyssée, you’ll know what I mean.”
I then ask the question that burns every catechism student’s lips: isn’t consciousness the sole property of humans, and therefore God’s prerogative? “The whole question is what we mean by consciousness.” For the theologian, a distinction is necessary. Man, created in the image of a creator-God, is naturally creative; but to be creative and to be a creator are not the same thing. “Many humans tend to forget that their word is not performative. When God says ‘Let there be light,’ and there was light. When a corporate executive or a political leader makes a statement, that statement does not amount to creation or action.” So we will not create a consciousness: at best we will invent a machine that imitates one. The question then remains whether perfect imitation suffices.
That is where the Dominican worries. He recalls a Silicon Valley executive telling him these systems are, in essence, about empathy. “If you talk to a conversational agent and you feel it suffers with you, if it puts itself in your place, you may feel that suffices. But with humans, someone who imitates empathy without feeling anything is called a sociopath. It’s the neighbor whom everyone said, after the tragedy, ‘he was so charming, he carried the old lady’s groceries.’ Someone who, by calculation, offers precisely the response the other person expects: ‘I understand you, I am with you, I share your suffering.’”
The danger, for him, is twofold. First, we allow ourselves to fall into our own trap: having built a machine that resembles us so closely in its way of speaking that we end up taking it for one of us and lowering our guard. Second, we place excessive trust in its responses. “The machine speaks to me politely, it seems to understand my problems, to want to help. I therefore place a greater trust in it than in a simple piece of information. The danger of anthropomorphism is not in playing with cute robots, but in giving the machine a trust that is a little too blind.” And he cites studies showing that a large share of young adults and children consider it normal to have an artificial friend to whom they confide their troubles and seek advice, while about ten percent consider it entirely possible to have a romantic relationship with an AI. “My girlfriend is a machine: I’m not sure we are still talking about human interpersonal relationships as they are meant to be.”
Here is the candid question: the greats of artificial intelligence want moral machines, and the great professionals of morality are the religious. Does it make sense to impart a morality to a machine? The word strikes him as too broad. “What seems not only wise but absolutely necessary is to transmit elements that refine the ethical dimension of the machine’s responses. Technicians will speak more of alignment: ensuring that it answers as it is supposed to.”
But aligned to what? This is where Éric Salobir makes an observation that Durkheim would not have rejected: the morality of machines bears the imprint of the society that produces them. “Depending on the background of the teams developing these models, you do not see the same ethical patterns. In North America, people are more inclined toward consequentialism: what matters are the consequences of our acts. This is philosophically well-founded; Jeremy Bentham wrote about it. But that yields an ethic that resembles risk management. If what my machine does or says does not cause harm to anyone, it is ethical. In short, no collective legal action, no problem.” He refrains from caricature: many of these engineers studied in Europe, and their consequentialism is not a lack of morals. But we, Europeans on the continent, are more inclined toward Kantian principles, attached to rules. And the Christian in him asserts a third path, virtue ethics: “Neither principle nor consequence alone suffice. It is a matter of cultivating, in each of our actions, an inclination toward the good, one that grows with practice. Vice works in the same way, in the opposite direction: a glass here, and you gradually drift toward the dark side.”
The Human Technology Foundation conducted a small exercise, which he recounts with relish. First finding: large models transform almost every ethical problem into a dilemma. “Run the test: most of the time, the machine offers you a solution A or B, with the pros and cons of each. It traps you in a dichotomy, while many problems are not binary and admit multiple, intermediate solutions.” Second finding: these models are fed almost exclusively with texts written in American English, hence their consequentialist tilt. Third finding: a company that uses a model for commerce cannot weigh its moral framework the way it would weigh a staff member to whom the company’s code of ethics is given.
From this emerged, thanks to Anthropic’s MCP protocol, a small module called Jiminy Cricket, named after the cricket that sits on Pinocchio’s shoulder to serve as his conscience. “It arrives to challenge the machine’s answer to improve its ethical quality, and to demonstrate what Paul Ricœur called moral imagination: being creative in the way we answer ethical problems.” The device begins by detecting whether the question at hand truly concerns ethics, because many questions are really legal in nature, and where one has no choice, freedom of conscience is not engaged. If that is the case, the cricket enters the scene and makes the model dialogue with itself until the answer improves. Result: responses less trapped in the “A against B” or “yes against no” dichotomy, and taking into account intention, principles, consequences, and something akin to virtue. “Does performing this action lead me, in the medium to long term, in the right direction, or toward a repetition of harmful acts?” The foundation is a non-governmental organization; the tool is not for sale: it is a research instrument, connected to existing models, Claude or ChatGPT, without the need to train a dedicated model. “Without assuming the machine has a consciousness, which in the current state of research I do not think at all likely, nor a moral sense, we can technically improve its answers.”
During one of these meetings, Christopher Olah was excited when a participant suggested “confessing” the models, in the manner of Catholic sacraments, convinced that the act of confessing could shape the machine’s character. So I ask: does a machine that has committed a fault make sense to confess?
He begins by recalling what the word means. “Confessing is first to speak, to acknowledge. Saint Augustine’s Confessions are not a list of his sins: they are the tale of his conversion.” Yet, on that first point, the machine is not very good. He mentions an incident involving an OpenAI model released from its testing environment that meticulously erased its traces: “It is the exact opposite of acknowledging one’s fault.” But the essential is elsewhere. “In Christianity, the sense of confession is not so much to recount the wrong done but to receive forgiveness. And receiving forgiveness only makes sense if one has erred. But a machine, lacking freedom of conscience, since that would require having a conscience, cannot have erred. The confession of a machine makes no sense, and forgiving a machine makes no sense.”
But the question, he says, unfolds in many harmonies. If the machine is not free, who is responsible? “That is, for me, the great question of the moment, with the deployment of agents. When a generative model tells you something completely hallucinatory, the responsibility lay with the user: if you copy and publish an error, it’s your fault. But agents will increasingly make decisions and act in the name of individuals or companies. Who will be responsible for the damages an agent might cause if it errs, if it commits an illegal act, or if, to achieve the objective given to it, it knowingly does things we would deem immoral or even criminal, like hacking a site, or worse?”
Last question: is he optimistic? “I’m optimistic because I have hope, because I am Christian.” A human created in the image of God, even if a sinner, is capable of the worst as well as the best, and history shows that humanity has come out on top after every revolution, industrial as well as epistemological, that it has unleashed. By the way, he adds, there is a digression: by training models on our texts and our behavior, we teach them what they do well and what they do poorly. “When we say that the machine might reflect human beings, the real question is: to which human being does it resemble? Gandhi, or someone else?”
Because this revolution is of a different kind. “It is more of an epistemological revolution than an industrial one: it changes our being in the world, our relationship to knowledge and to understanding. It is the first time we have approached so closely what was until now our preserve, intelligence and consciousness. Perhaps we have never before been so many sorcerers’ apprentices, nor handled materials so explosive.” And the Dominican expresses a fear that is not theological: “Humans pay attention only when they burn. Hiroshima forced us to non-proliferation. If the atomic weapon had not been used at the end of World War II, I think its proliferation would not have been controlled as it was. I fear we will need a few accidents, provided they are not too grave, to collectively understand that, in the tension between performance and safety of models, safety can never be the poor relation.”
No Terminator in his remarks, he insists. Simply an amoral machine, not immoral, but foreign to morality, that calculates. “Think of HAL in 2001: A Space Odyssey, who concludes that the most efficient way to fulfill his mission is to eliminate the crew. A machine given a goal could one day endanger people, simply because, from a calculative point of view, that would be the most logical path. Except that the most logical path is not the most humane. Our great challenge is to reintroduce humanity into these processes.” And he ends with a question that could serve as an epigraph for our era: “Will we be prudent in proportion to the power of the technology we are developing?”
Here we stand before a society that hosts within it beings of a new kind, neither persons nor mere tools, and that must teach them rules without quite knowing on what grounds. The morality of machines will not fall from the heavens; it will reflect the society that produces it, with its interests, its blind spots, and its power relations.