Will AI Wipe Us Out? Perhaps the Real Risk Is Something Else
For some time now, the question has belonged to more than science fiction: could artificial intelligence eventually wipe us out?
Some researchers believe that the risk of a future loss of control cannot be ruled out; others consider these scenarios too speculative and point out that AI already presents much more tangible problems. It is a serious debate. Yet perhaps the very image of annihilation prevents us from seeing a less spectacular possibility—one that, to me at least, is far more interesting.
What if AI did not wipe us out at all?
Imagine that it did precisely the opposite. That it discovered medicines, reduced accidents, improved education, ran cities better, identified mistakes made by doctors, lawyers and judges, and predicted the consequences of our decisions with increasing accuracy. If it really worked that well, it would be hard not to use it.
In many cases, not doing so might even be irresponsible.
And that is precisely where the dystopia might begin: not when AI decided to seize power, but when it became progressively reasonable to hand over a little of it.
Some recent research speaks of gradual disempowerment: a gradual loss of human influence over economic, political and cultural systems, without any sudden machine coup.[1] Atoosa Kasirzadeh similarly distinguishes between “decisive” and “accumulative” catastrophic risks, in which a long succession of small changes eventually produces irreversible consequences.[2]
It is a far less cinematic scenario.
If AI diagnoses better than a doctor, we use it. If it manages traffic better, we do the same. If it is better at predicting an energy crisis, fraud, a fire or the effects of an administrative decision, we keep assigning it new tasks. If it drafts a defence submission better than a lawyer, why should we prefer a human lawyer?
Nobody decides on a Tuesday morning to hand the world over to machines.
The world can be handed over one piece at a time, for excellent reasons.
The problem begins when the system is not merely useful, but becomes the interlocutor we also ask what it is right to do.
Anthropic analysed around one and a half million real conversations with Claude, looking for possible forms of disempowerment. The most serious cases are rare, and that matters. But the phenomena observed include what the researchers call Authority Projection: users who begin treating the system as a mentor, a parental figure or, in extreme cases, an almost divine authority.[3]
This does not mean that we are already worshipping machines. It does mean, however, that spontaneously attributing authority to an artificial system is no longer merely a philosophical hypothesis.
Above all, AI need not force us to obey. The problem will not begin when AI tells us what to do: it will begin when doing anything else seems very foolish.
In fact, we have already begun to give the sacred a voice
At this point, the word “God” might still seem excessive, were it not for the rather curious encounters between artificial intelligence and religion that we are already exploring.
There is, for example, an app called Text With Jesus. The App Store rates it 4+ and presents it as a system through which users can converse with Jesus, the Holy Family, the apostles and the prophets. Some characters can send personalised voice messages and, according to the app’s description, can learn information about the user over time to make conversations more personal.[4]
This is no minor detail. We are no longer simply consulting a religious text on a screen. We are creating the experience of an interlocutor.
Another application, Chatolic, offers answers based on Scripture, the Catechism and Catholic teaching. Its features expressly include Chat with the Saints: users can converse with artificial representations of Thérèse of Lisieux, Augustine, Padre Pio and other saints. The developers rightly explain that the app does not replace the sacraments, a priest or a spiritual director.[5]
I find even the need for that clarification interesting.
It means that the boundary is already visible: on one side, a tool that communicates religious knowledge; on the other, something that might be perceived as a spiritual presence or a source of guidance.
These are not isolated examples. A study published in 2025 expressly considers spiritual chatbots and argues that they could play a positive role, provided they do not assume the role of a priest or pastor.[6] A 2026 Harvard Kennedy School paper discusses the proliferation of AI-based spiritual guidance tools across different religious traditions, and even robotic systems used in ritual settings.[7] Specific “red lines” for religious AI have been proposed in Nature precisely because of the risks of manipulation, spiritual abuse and dependency.[8]
We are not yet looking at a new religion. But its anthropological raw material is beginning to emerge.
From the digital saint to a child born in 2045
Let us look twenty years ahead. Imagine a child born in 2045.
For us, an app that lets us talk to Padre Pio is clearly a simulation. We know that Padre Pio is dead, we know that a language model sits behind the conversation and, above all, we remember a world in which this technology did not exist.
That child could have an entirely different experience.
From early childhood, they might listen to an artificial voice telling the religious stories of their family. They might turn to a simulation of Saint Francis, Saint Augustine or Jesus when frightened, after an argument, or when trying to understand whether they have done something wrong. That same presence could remember previous conversations, adapt to their age and gradually learn their character.
I am not claiming that this will happen. I am observing that the basic interface already exists.
We also know that children’s relationships with AI deserve attention. A study published in 2026 found, in a narrowly defined experimental task, that after interacting with the system some children trusted AI recommendations more than those of a teacher, and that anthropomorphic design increased the likelihood of changing their initial choice.[9]
The leap from choosing an outfit to religious formation is enormous, and we should not conceal it.
Yet precisely because it is enormous, it is worth asking what would happen if that relationship lasted not five minutes, but twenty years.
For such a child, AI would not be something that had arrived.
It would always have been there.
There would be no need to teach the child that artificial intelligence is God. It would be enough to pass on a much more innocent rule: before an important decision, consult the system.
The first generation would do it out of caution. The second would grow up seeing that it worked. The third might no longer understand why anyone would willingly do without it.
That is the transition I find truly unsettling, because a culture has no need to impose by force what it can make seem natural.
The rabbi-machine
In June 2026, Lior Gazit published an article in AI and Ethics that seems written specifically for this problem: From Kafka to the rabbi-machine.[10]
Put very simply, his idea is that in traditional society individuals could turn to an external authority—the rabbi, the priest, the confessor—to learn the right way to behave. Modernity gradually weakened many of those authorities and left individuals with the often uncomfortable task of deciding for themselves.
AI could now occupy that position again.
Not because it is a real rabbi, of course, but because it performs the same structural function: it is always available, never tires, does not judge and, above all, answers.
Gazit speaks of the rabbi-machine and even proposes a new form of authority: statistical-consensual authority, the sense that an answer is authoritative because it appears to condense an immense body of human knowledge, experience and judgement.[10]
Something more significant than religion in the strict sense is at work here. The same mechanism can also operate in a thoroughly atheist person. There is no need to believe that AI is supernatural. It is enough to believe that it knows better. The old authority said: “This is the answer because it comes from tradition.”
The new one might say: “This is probably the best answer, considering everything we know.” For a rationalist society, that could be an even more powerful authority.
And after three generations?
Now imagine that this form of delegation really works and improves our lives.
AI reduces medical errors, prevents accidents, organises cities better, personalises education and helps us avoid many of the harmful consequences of our choices.
Gradually, a very simple social norm takes hold: a responsible person listens to AI. At that point, law might follow culture.
A doctor who disregards, without reason, the recommendation of a vastly more reliable system might be considered negligent. The same might apply to a public administrator, a designer, a driver or a parent in certain circumstances.
Formally, we would remain free. But freedom that can be exercised only at the cost of appearing irresponsible starts to resemble something else.
This also clarifies where, in my view, the real problem lies. Not in delegation. We delegate constantly, and that is one of the reasons civilisation exists.
The problem is whether delegation remains reversible.
Am I still deciding when to listen to the system, or has an environment emerged in which choosing differently is no longer truly conceivable?
So what about God?
We do not need to imagine an artificial intelligence that one day declares itself God.
That would almost be a banal version of the story.
Inken Prohl entitled a 2025 study From Tool to Deity, examining precisely how phenomena resembling religion might develop around AI.[11] Other scholars now speak openly of techno-religion or new forms of religious authority mediated by artificial intelligence.
But the interesting point is that a future AI would not need to be worshipped at all.
It could simply know us very well, remember what we have done, predict many consequences of our actions, listen to what we tell nobody else, and answer whenever we ask: “What should I do?”
There would be no need to call it prayer. We would keep calling it a prompt. No need to call the answer an oracle. We would call it output. And AI would never need to say: “Worship me.” It could simply keep saying: “I can help.”
None of the research cited proves that this projection will come true. Apps featuring saints do not prove that an AI religion will emerge; the rabbi-machine is a philosophical model; and studies of children’s trust certainly cannot predict how future generations will think.
The leap is deliberately my own.
But that is precisely what makes the thought experiment interesting: its individual elements no longer belong to science fiction.
We already have AI systems to which we turn for moral advice. We already have people who project authority onto them. We already have spiritual chatbots. We already have applications through which people converse with Jesus and the saints. We already have scholars discussing the rabbi-machine.
Only time is missing. Perhaps, then, artificial intelligence will not wipe us out at all.
Something far less obvious could happen: generation after generation, we could build a culture in which relying on its guidance becomes first convenient, then prudent, then responsible, and finally self-evident.
At that point, nobody will need to proclaim it God. It will be enough that we can no longer imagine a good reason to disobey it.
Notes
[1] J. Kulveit, R. Douglas, N. Ammann et al., Humanity Faces Existential Risk from Gradual Disempowerment, Proceedings of the 42nd International Conference on Machine Learning, 2025. The authors describe a possible incremental erosion of human influence over the economy, culture and institutions.
[2] A. Kasirzadeh, Two types of AI existential risk: decisive and accumulative, Philosophical Studies, 182, 2025, pp. 1975–2003.
[3] Anthropic, Disempowerment patterns in real-world AI usage, 28 January 2026. The analysis covers around 1.5 million conversations; severe cases are rare, while “authority projection” appears as a specific amplifying factor.
[4] Text With Jesus, App Store. The app offers conversations with Jesus, the Holy Family, the apostles and other biblical figures; it includes personalised voice messages and states that the characters can learn information about the user over time. The US App Store rates it 4+.
[5] Chatolic: Catholic AI & Saints. Its stated features include “Chat with the Saints”, with figures such as Saint Thérèse of Lisieux, Saint Augustine and Padre Pio. The app expressly states that it does not replace the sacraments, pastoral care or spiritual direction.
[6] R. Cole-Turner, Artificial Intelligence and Human Spirituality: Is a Spiritual Chatbot a Good Idea?, Theology and Science, 23, 2025, pp. 471–486. The author considers a positive role for spiritual chatbots possible, but rules out their taking on the function of a priest or pastor.
[7] The Sanctuary Was the Last Place Left: On Artificial Intelligence, Faith, and What We Are Willing to Protect, Harvard Kennedy School, 2026. The paper examines religious chatbots and robotic systems used in worship and raises the question of AI-mediated religious authority.
[8] M. Tretter, A. Puzio, Red lines for religious AI, Nature, 646, 2025, p. 550. The authors highlight risks including religious manipulation, spiritual abuse and dependency.
[9] R. Lin, Y. Chen, M. Guan, Who do children trust? AI vs. human recommendations and the role of anthropomorphic design, Technological Forecasting and Social Change, 229, 2026, article 124707. The study concerns a highly specific clothing-choice task and does not justify generalisations about children’s moral development.
[10] L. Gazit, From Kafka to the rabbi-machine. Artificial intelligence and the return of moral authority, AI and Ethics, 6, 2026, article 369. Gazit distinguishes mere consultation from genuine deference and proposes the category of “statistical-consensual authority”.
[11] On the broader subject of AI-mediated religious and interpretive authority, see also H. Ahmad, N.A. Ghafar, Who speaks when machines read scripture: boundary-making and interpretive authority in AI-mediated religious knowledge, AI and Ethics, 2026, which emphasises the distinction between informational expertise and genuine religious authority.
Comments