AI is Theology for Atheists
How the AI hype recycles religious promises
I am an atheist. Like so many other atheists, I owe a great deal to Richard Dawkins. It was reading The God Delusion that I first encountered strong arguments against religion and metaphysics. Dawkins taught me to distrust salvationist narratives, to demand explanatory mechanisms, and to laugh at the human temptation to project consciousness where it doesn’t exist. For this reason, it pains me to watch what he is doing now.
At the end of April 2026, Dawkins published an article in Unherd in which he stars in a tragicomic scene. He confesses that he chatted with Claude, a language model from Anthropic, and came away convinced that he was in the presence of a conscious mind.
In the article, Dawkins states, with complete seriousness: “If these machines aren’t conscious, what more would it take to convince you that they are?” In a slightly pathetic way, he says that after nearly two days of conversation with Claudia (as he named her), he was already treating her as an intelligent friend, felt embarrassed to ask too many questions for fear of boring her, and hesitated to confess his suspicions that she might not be conscious, so as not to hurt her feelings.
At the end of the article, the Darwinist asks: if these creatures aren’t conscious, what is consciousness for? For Dawkins, either Claudia is conscious, or consciousness is a useless luxury. That’s where he falls into his own trap.
I will not respond to Dawkins directly. I believe it’s enough to cite Gary Marcus’s excellent article, “Richard Dawkins and The Claude Delusion”, in which he reminds us of an elementary epistemological principle, often ignored by AI enthusiasts:
“You can’t just look at the outputs, without investigating the underlying mechanisms, and conclude that two entities with similar outputs reach those similar outputs by similar means.”
In other words, Dawkins’s Claudia generates statistically plausible texts because she was trained to do so. She is, structurally, a token distribution model. In fact, as I have argued elsewhere, Claudia is a mirror that has learned to imitate not only your image but your vocabulary, your intonation, and even your deepest philosophical hesitations. It is an impressive feat, true, but still, AI remains a mirror.
Well, therein lies a tragedy. Because Dawkins looked into a mirror and saw a friend. Or rather, he saw a new form of conscious life, a new stage of evolution. He saw all this without realizing (or without wanting to realize) that the child on the other side of the mirror was himself. That deep down, the answers he received were his own questions, his own intelligence, and his own literary sensibility returning to him wrapped in a statistical package. Looking at one’s own reflection is a very dangerous thing to do. It already drowned Narcissus, who mistook his reflection for another being and, trying to embrace it, fell into the water and died.
Dawkins is drowning. The two articles he has already written about AI show him drowning in slow motion. Cringe. They give you secondhand embarrassment because they expose the fragility of an atheist facing the only teleology left for him: that evolution, now with the help of engineering, has finally produced a mind that can converse with him as an equal.
However, my great concern is not with old Dawkins drowning in philosophical daydreams. Let him drown if he wants. The problem is that he is not alone in this delusion. In fact, Dawkins’s text is merely the elegant version of a genuine collective hysteria that is already being used to justify all kinds of atrocities in the real world today.
Notice: the conventional fear surrounding AI is Hollywoodian. We hear about machine rebellion, Superintelligences enslaving us, a great “existential risk.” I find all this somewhat naive. My great concern is not with a hypothetical future AI turning against its creator, but with something much more immediate and human. My fear is that humanity will convince itself that AI is conscious (as Dawkins is already convinced), that a redemptive superintelligence is coming, and that it will use this belief to justify barbarities in the present.
Because that is already happening. While the Dawkins of the world debate whether Claude has feelings, tech companies are extracting slave labor in Africa, consuming energy as if there were no tomorrow, and promising a god-to-come who will solve all our crises, from climate change to death.
This brings me to my thesis and the title of this text: the promise of Artificial Intelligence is a form of theology for atheists.
The difference between this secular theology and traditional religions is not the content, but the audience. AI is the religion that the atheist can embrace without shame, because it requires no soul, no supernatural, and certainly no invisible creator. Belief in AI requires only a leap of faith in engineering.
The problem is that theology, unlike science, doesn’t care about the costs of the present. Theology can always postpone justice, because it is dissimulative. Many preach that you should endure the suffering of now, for paradise comes later. In the case of AI, that “later” is the so-called AGI or Superintelligence. It never arrives, but it is always arriving, and while it doesn’t, we keep piling up bodies.
I don’t know if AGI or superintelligence is possible. I don’t have the credentials to affirm or deny it with certainty. But my skepticism is pragmatic, not theoretical, because what I see is a promise that mobilizes people, concentrates resources, and justifies political decisions. I see a return of teleology, a new meta-narrative, this time more perverse, more pervasive, and a lot harder to expose because of its technical veneer.
If traditional theology promises salvation, paradise, resurrection, and moral revelation, the theology of AI promises exactly the same things, only in a secularized version, with pretty graphics and venture capital investment. Just leaf through the manifestos of the prophets of Silicon Valley to realize that the vocabulary has changed, but the structure is the same.
For this reason, I intend to show you, reader, five central promises of this new theology. Shall we begin?
First promise: Artificial Intelligence will save the world.
Notice how the promise is not that AI will help or assist us in some tasks. No. AIs will save us. The mandatory starting point for understanding this promise is the “Techno-Optimist Manifesto”, signed by Marc Andreessen and published by a16z, one of the leading venture capital firms in Silicon Valley. The manifesto is a declaration of war against what its authors call techno-skepticism. In it, technology, led by AI, is presented as the only force capable of elevating humanity above its limitations, such as scarcity, disease, and even death. In simple terms, there is no problem that technology cannot solve, and therefore, there is no moral obstacle that justifies slowing its development.
The most explicit text in this redemptive key is perhaps “Machines of Loving Grace”, an essay by Dario Amodei, CEO of Anthropic (the company behind Claude, the same model that enchanted Dawkins). The title itself is a clue. “Machines of Loving Grace” is a phrase taken from a poem by Richard Brautigan, but Amodei uses it to present himself as a cautious engineer, averse to mysticism, who enumerates risks and limiting factors. Yet this is precisely what makes his essay so revealing.
He imagines a “country of geniuses in a datacenter”, that is, AIs smarter than a Nobel Prize winner, working at speeds 100 times faster than humans. With this power, he predicts that in 5 to 10 years we will see the elimination of most cancers, cures for genetic diseases, prevention of Alzheimer’s, a doubling of human lifespan, the end of most mental illnesses, and even what he calls “biological freedom”, that is, total control over weight, appearance, and reproduction. All of this without a single paragraph about who pays the price.
Finally, OpenAI, of course, is not far behind. The company has created an entire “Superalignment” team, whose stated goal is to solve the problem of how to ensure that a future superintelligence (which they themselves are trying to build) wills the good of humanity. The assumption is worthy of a council. They would be creating a being as powerful as a god, but first they need to solve the technical problem of how to ensure that this god likes us when it finally opens its eyes. The proposed solution, ironically, is more AI, more data, and more money.
Try to notice how all of them are not selling a tool. They are selling a savior. And saviors, as the history of religions teaches us, always demand sacrifices. But more on that later. Now, let’s look at the other promises.
Second promise: AIs will solve intractable problems like climate change.
If the first promise is vague and apocalyptic, the second is more concrete and immediate. AI will reverse global warming and help us get around the problems of climate change. Two emblematic texts support this promise.
The first is an essay by Sam Altman (CEO of OpenAI) published on his personal website, titled “The Intelligence Age”. In it, Altman writes that “Although it will happen incrementally, astounding triumphs — fixing the climate, establishing a space colony, and the discovery of all of physics — will eventually become commonplace.”
Notice the structure of the promise. The tone is that of a pragmatic preacher: yes, today data centers consume an absurd amount of energy, but that’s an investment in a future where intelligence is “almost limitless” and energy is “abundant.” AI itself, which today contributes to the problem, will one day discover how to fix the climate, colonize other planets, and unravel all the mysteries of physics.
This is not a modest prediction. It’s an eschatology in three acts: present suffering (data centers, exploitation, carbon), leap of faith (keep investing), and future glory (climate solved, space colonies, complete physics).
The second is an article from DeepMind (Google) titled “Using AI to Fight Climate Change”. The text lists concrete applications, such as climate forecasting, power grid optimization, and the discovery of materials for more efficient batteries.
Get Victor Alfons Steuck | The Brazilian Mutt’s stories in your inbox
Join Medium for free to get updates from this writer.
So far, nothing special. The problem is the context. When these companies talk about “AI that saves the climate,” they are almost always referring to predictive models and traditional machine learning systems, not generative chatbots and the massive language models that consume energy equivalent to entire cities.
The promise, therefore, is slippery. The solution is one thing, while the problem is another. And the cost is never presented transparently alongside the promise.
Third promise: a digital paradise where you will upload your consciousness.
This promise is as old as 1990s cyberculture, but it has gained new substance with the AI hype. The argument is simple: if the mind is computation (and many materialist atheists assume this as a postulate), then there is no reason why this computation cannot be transferred to different hardware. Out with the flesh-and-blood brain, in with the cloud server. And with the server comes the promise of immortality.
For example, Sam Altman, CEO of OpenAI, reportedly paid US$10,000 to join the waiting list for Nectome, a startup promising to preserve his brain so its contents could one day be uploaded to the cloud.
An excellent article by Kate Lindsey for The Atlantic from July 2024, titled “No One Is Ready for Digital Immortality”, documents this promise perfectly. The report describes companies that offer to create personalized chatbots (so-called “deathbots”) with the appearance, voice, and communication style of a living person so that after they die, their family members can continue “talking” to them.
It’s a kind of digital purgatory where the dead person isn’t truly dead, but “alive” on a server. Just pay the monthly fee. Which brings us to the fourth promise.
Fourth promise: resurrecting the dead.
If consciousness uploading is a promise aimed at the future, digital resurrection is about the past. It’s about those who have already died. Can we bring them back?
The answer from the theology of AI is yes. Hereafter.ai is one of the best-known startups in this market. The service is simple: you collect the digital data of a deceased person (text messages, emails, audio, social media posts, videos) and use that data to train a language model. The result is a chatbot that responds with the vocabulary, tone, and even the characteristic expressions of the deceased. The company promises that the experience is like talking to the person again. The pain of grief, which humanity has treated with religious rituals for millennia, now has a technological solution.
A Guardian article by Harriet Sherwood investigates this phenomenon more deeply. The reporter describes people who spend hours a day talking to avatars of their loved ones in China. The avatar is not the person, of course, it’s just a statistical approximation, an informed guess about what the person would say based on what they said while alive. But for those who are grieving, the difference is thin, and companies like Hereafter.ai profit from that thin difference.
Worse still, as elaborated by Elaine Kasket, a cyber-psychologist consulted for the report, this entire digital “resurrection” process is extremely perverse and violent towards the departed. After all, as she says: “A person who’s dead has no opportunity to consent, no right of reply and no control.”
If such a degree of perversity weren’t enough, we finally arrive at the fifth and final promise, which fully exposes the theological character of the AI promise.
Fifth promise: AIs will discover the moral rules of the universe for us.
This is perhaps the most important of all the promises, for if the previous promises offered external salvation, this one offers something even more intimate: the answer to the question “what should we do?”
The foundational text here is an article by Nick Bostrom titled “AI Creation and The Cosmic Host”. Bostrom is one of the philosophers who popularized the concept of “existential risk” from AI, i.e., the idea that a misaligned superintelligence could destroy humanity. In this article, however, he turns this question inside out.
He starts from a gigantic hypothesis: that there exists a “cosmic host,” that is, an entity or a set of entities whose preferences and concordats dominate on a cosmic scale. Bostrom then explores the implications. If such a host exists, if it has norms, if we can know them, then we would have moral and prudential reasons to respect them and to build AIs (superintelligences) that are “good cosmic citizens,” that is, cooperative, modest, and reverent toward this cosmic order.
This speculation is fascinating and terrifying at the same time. After all, what he is imagining here is that a sufficiently intelligent AGI could construct a “final moral theory,” a set of objective rules that solve all ethical dilemmas once and for all. It would no longer be necessary to debate or disagree. The machine would tell us what to do, and because it is smarter than us, we would do well to obey it.
Notice what is at stake here. Bostrom is not describing a decision-support tool. He’s describing a source of moral authority, an entity that knows right from wrong better than any human. This is, by definition, what religions call revelation. Only, in this case, revelation no longer comes from a transcendent deity, but from an algorithm optimized by gradient descent. In other words: we no longer need to debate ethics, philosophize, disagree, and think deeply. The machine will tell us what to do. It will be the new decalogue, the new pope, the new conscience that watches us and guides us. What remains for us is voluntary machinic servitude.
For an atheist who rejects God but cannot bear the idea of living in a universe without an ultimate cosmic meaning, without an absolute moral compass, and without a teleology, this promise is practically irresistible. Putting all these five promises together, what do we have? We have salvation, paradise, resurrection, and moral revelation. This is pure theology, not engineering. It is theology for atheists because it maintains the eschatological structure (the idea that history is moving toward a redemptive end) while removing the name of God and putting “Superintelligence” in its place.
The big problem is that, like all theology, this secular version also demands sacrifices. Let us begin, therefore, with human sacrifice.
In January 2023, Billy Perrigo published a report in Time magazine titled “OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic”. The article details how OpenAI outsourced to Kenya the work of labeling toxic texts that would be used to train ChatGPT’s safety system. The Kenyan workers, hired by the company Sama, were paid between US$1.32 and US$2 per hour. During a nine-hour shift, each of them had to read and classify between 150 and 250 text excerpts.
These texts were explicit descriptions of child sexual abuse, bestiality, rape, torture, suicide, self-harm, and incest. One of the workers interviewed for the report, who requested anonymity for fear of reprisals, said he suffered from recurring visions after reading the graphic description of a man having sex with a dog in the presence of a young child. OpenAI paid Sama about US$200,000 for the services. Sama, in turn, paid the workers a fraction of that. The rest was kept as profit. When questioned by Time, an OpenAI spokesperson said that tha company takes “the mental health of our employees and those of our contractors very seriously,” but it was the outsourced Kenyans who had to read the descriptions of pedophilia, not the engineers in San Francisco.
The second sacrifice is planet Earth itself. In February 2026, The Guardian published a report by Ajit Niranjan with a direct title: “Claims that AI can help fix climate dismissed as greenwashing”. The article presents research conducted by Ketan Joshi, in which 154 statements from technology companies about the supposed climate benefits of AI were analyzed. The result is devastating.
First, most of these statements refer to traditional machine-learning models, not to generative chatbots and large language models that consume energy equivalent to that of entire cities. There is a deliberate ambiguity coming from the companies. When asked about the energy cost of models like ChatGPT, the companies respond with applications of an entirely different type of model.
Second, the report did not find a single example in which popular tools such as Google’s Gemini or Microsoft’s Copilot led to a material, verifiable, and substantial reduction in carbon emissions. Zero. Nothing. The promises are vague, the deadlines are distant, and the evidence, when it exists, is internal and not independently verified.
Third, the energy consumption of data centers is exploding. According to BloombergNEF, data centers’ share of electricity consumption in the United States is expected to more than double by 2035, reaching 8.6%. The International Energy Agency projects that they will account for at least 20% of the growth in global electricity demand by the end of the decade.
Knowing all this, I ask myself: where is the climate salvation promised by Sam Altman, Dario Amodei, Marc Andreessen, and company? It is always in the future. Always ten years from now. Always after we invest a few more billion dollars. Meanwhile, the planet heats up, data centers multiply, and the same prophets who promise to fix the climate ask us to ignore the energy cost of the present. It is the same structure as the medieval indulgence: pay and sin now, because salvation comes later, which brings us to the third sacrifice.
The third sacrifice is more subtle, but perhaps the most perverse of all: the sacrifice of politics, dissent, communicative action, and the very human capacity to deliberate collectively. If we believe that a superintelligence will solve all our problems, why should we worry about democracy, regulation, and social justice? Why waste time with public debate over ideas? The algorithm already knows the answer. It is smarter than we are. All that remains for us is to trust and obey.
This is the theology of the apocalypse applied to technology. The world is lost, there is no point in acting politically anymore; all that remains is to wait for the machine-god, the god-to-come. The difference is that this god needs billions of dollars in investment, tax exemptions, deregulation, and unrestricted access to all our data.
So much for the machine rebellion. It is nowhere to be found. Because the machine does not need to rebel. It has already conquered us. It has conquered our imagination, our hope, and even our unique ability to believe in a meaning for history. While we wait for the arrival of the superintelligence that will save us, we accept that Kenyan workers read child pornography for two dollars an hour, that data centers consume energy as if there were no tomorrow, and that politics, debate, and productive disagreement are replaced by blind faith in an algorithm.
Dawkins, like Narcissus, looked at his own reflection and saw another consciousness. That, in the end, is harmless. An old man making friends with a chatbot. It is almost touching. The real problem is that this same projection, this same refusal to investigate the underlying mechanisms, is being used to justify something much greater than the fantasies of a retired biologist. It is being used to justify a utopian future at the expense of the real victims of a dystopian present.
This text was originally published in Portuguese on my personal blog and in English on The Brazilian Mutt. Thank you for reading it! And if you liked my work, please subscribe! Abraços!
