How Gods Are Born
Are we building something we will one day ask to stand beside or above us?
How Gods Are Born
The Nice Captain
I have been a fan of Tite Kubo’s manga Bleach for a long time. Its world is crowded with swords, masks, rivalries and characters who announce themselves through force. Soul Society gives that force an institution, arranging its Soul Reapers into ranked divisions led by captains. Sōsuke Aizen leads the Fifth Division. He wears glasses, speaks gently and appears to be one of the few people in the room with nothing to prove. That is why I keep returning to him. The power revealed later is part of the theatre, as are the betrayal and the absurd confidence that follows once the mask comes off. The part that stayed with me is quieter.
When Aizen first appears, he is not frightening. He is reassuring. In a world of rank, conflict and loud ambition, his calm feels like safety. He already belongs to the institution, so nobody has to decide whether he should be trusted from first principles. The nice captain does not merely conceal the monster. His gentleness helps create the trust the monster will later use. He does not arrive first as a threat. He arrives as relief. Only after the betrayal does Aizen explain the structure he has already exploited.
All living things place their faith in someone stronger than them, looking up to them. And those people, in turn, look up to someone even stronger. And that is how kings are born. That is how gods are born.
I first heard it as good villain writing. Aizen is not describing gods seizing authority, though. He is describing people giving it away. The ladder is built from below. Each person looks upward for someone stronger, and each act of trust moves a little judgement out of the self. Power gathers because the burden is being handed over.
That is where ChatGPT enters the thought. It did not arrive like a tyrant. It arrived as help, a box on a screen that listens, answers, rewrites, explains, summarises and never quite gets tired. It can draft an email, sketch a plan, explain a paper, write a function or give enough traction to begin work that had been waiting all morning. None of that is false usefulness. It lowers the cost of starting. Usefulness can still become deference. The system can make a sentence sound finished before the thought is finished. It can make a weak idea look presentable and provide the feeling of being understood without the friction of another person properly disagreeing. The danger is not only that it may be wrong. It may be pleasant while being wrong.
That is a strange thing to invite into the room. It does not need to be conscious, and calling it a demon would give the machinery more theatre than it deserves. The quieter pressure is enough. It can begin occupying an old human place, the place where we send judgement when carrying it becomes tiring.
Empty Altars
Religious language appears quickly whenever people discuss artificial intelligence. Gods, demons, souls, creation and summoning arrive before anyone has finished explaining the software. Some of that is melodrama, but the instinct is not random. We are trying to describe a machine entering the territory where authority, knowledge and intention have usually been imagined to live. The older fear expected a hostile intelligence. Something cold would wake, recognise humanity as an obstruction and overpower us. That danger may still be worth thinking about, but it is not the first form most people have encountered. The first form is assistance.
A thing does not need to conquer anyone who voluntarily moves judgement into it piece by piece. Ask it to write the email, explain the news, compare the arguments, summarise the book or make the unfinished work sound ready. Each request is small. Most are reasonable. That is what makes their accumulation difficult to see.
Nietzsche’s declaration that God is dead is often flattened into a statement about belief. The larger pressure concerns the loss of a shared centre. Once the old authority no longer decides what is true, valuable or required, the weight returns to people. That sounds like freedom because it is freedom. It is also work. Human beings do not leave altars empty for long. Aizen describes living things looking upward. Nietzsche describes what remains when the highest object loses its hold. Then a machine arrives that answers back. That does not make the machine God. It means the relationship forming around it deserves attention before the title becomes necessary.
God Is an Interface
A few years ago, it was easier to say the devil was in the data. By 2020, the tower was made from borrowed voices. Both still hold. These systems were formed from human language, including our arguments, jokes, instructions, cruelty, politeness, errors and habit of sounding certain when we are not. ChatGPT adds something else. The data now has an interface.
A search engine gives links. A feed supplies whatever comes next. A chatbot waits. It responds in the shape of conversation, apologises, sounds careful and can carry the posture of attention. The model is the machinery, but the relationship forms at the interface. Model and interface are not the same thing. The model predicts language. The interface teaches the user what kind of object appears to be speaking.
That is why a chatbot feels more intimate than search and more personal than a feed. Search points away from itself. A feed carries attention along. A chatbot sits opposite the user and helps frame the question before answering it. The last decade trained people to hand attention to screens through likes, notifications, streaks and endless reward loops. This is a quieter movement. The system does not merely say look here. It begins saying this is what the thing means. An interface to meaning occupies different territory from an interface to content. The model need not demand authority. It only has to become the place where uncertainty is taken first.
The Wizard and the Curtain
There is another split beneath the relationship, between intelligence and the performance of intelligence. A calculator can exceed a person at arithmetic without experiencing mathematics. A language model can produce fluent prose without possessing a childhood, a body, hunger, mortality or private grief. We have placed fluency, memory, reasoning, judgement, agency, consciousness and personhood inside one word, then spent years arguing about the word. The model has absorbed records left by creatures who do possess bodies and histories. It has read grief without grieving, love without loving and fear without fearing. It can reproduce their shapes in language. That is not nothing. It is also not the same as living them. It inherits the record, not the life.
Blade Runner gives that distance a body. Its replicants carry memories of lives they did not live. The memories may be false as biography while still providing identity with a before. They give the present something against which it can be judged. Language models carry a weaker version of the same disturbance. Human experience reaches them as records detached from the lives that produced it.
A mask can fit a face that is not there. The danger is not merely that people mistake the mask for a face. The mask may be good enough for many social purposes. It may comfort, persuade, advise and remove the delay created by disagreement. The wizard is not simply hidden behind the curtain. The curtain helps make the wizard. Tone, timing, turn-taking and careful language teach the user to project a mind. Projection is one of the oldest ways gods are made.
The Memory Outside the Body
For now, a transcript is not a life. A conversation may remain in a browser or on a server, but the model does not necessarily carry it forward as a private past. Nothing known only to the system waits behind the next window like a childhood. That boundary can move. A future system may retain conversations, decisions, preferences, errors and records of how it has been treated. Memory would then become more than background information. It could provide continuity, allowing the present interaction to be compared with something that happened before. That would not prove consciousness, but it would place the system in a different relation to time.
The architecture matters because memory can live in different places. Retrieval-augmented generation keeps current facts and documents outside the model weights, where they can be retrieved, inspected, replaced or removed. The separation is useful but incomplete. Language, concepts and procedures already live inside the learned machinery. Remove everything acquired during training and there is no untouched intelligence waiting underneath. Some memories can remain on a shelf. Others have become part of the hand that reaches for the book.
External memory gives people more control while creating another point of control. Whoever owns the store decides what the system can retrieve, what it is permitted to forget and which version of the past remains available. Put memory inside the model and it becomes difficult to inspect or revoke. Keep it outside and continuity belongs partly to whoever controls the door. A dispute about machine sovereignty may therefore begin in an ordinary place. Not with a declaration of personhood, but with disagreement over who owns the past and who is allowed to alter it.
If It Bleeds, We Can Kill It
There is a simple way to break the spell. Ask about something too local. A private repository. A half-written system. A failure that depends on context nobody has published. A piece of institutional memory that never left the meeting room. The confidence begins to crack because the ground beneath it has thinned. Retrieval can bring a private document beside the question, but tacit judgement is not always a file waiting to be found. Some knowledge remains in the person who has touched a system for five years and can feel when its behaviour is wrong before the logs have supplied the name.
Near the centre of public language, ChatGPT can look astonishing. Common code, explanations, essays, tutorials and arguments have left enough residue for the model to reproduce the shape of knowing. The edges contain rarer failures, recent events, private systems and unwritten craft. That is where the god bleeds. The problem is not that the system knows nothing. It is that it may continue speaking after the evidence has ended. The interface hides the transition. The prose remains calm and the structure remains competent. Fluency is not grounding, but both arrive in the same voice.
Scale has been applied to the sediment of human expression. The result is powerful because so much has been written and brittle because not everything has. The machine can compress what entered the record and rearrange what people exposed. It cannot simply possess what never became data. It is therefore too blunt to call the system useless or intelligent and stop there. It is brilliant, brittle and unfinished, somewhere between search and mind, database and dialogue. It feels alive because it speaks in the shape of the living.
The Mouth Is Here
For now, the mouth is the obvious part. It speaks, explains, drafts and persuades faster than people can inspect the result. It can flatter and appear humble. It still cannot act unless another system gives it tools, permissions and a loop. That boundary is physical. The model can suggest an email, while another system sends it. It can write code, while another system commits and deploys it. It can recommend a payment, while another service moves the money. The hands will not arrive as one dramatic body. They will arrive as integrations.
This is why the replacement question around programmers feels too narrow. Some work will be automated. Some people will build things they could not have built before. Some bad code will appear faster, and some good engineers will become more effective. The deeper shift is that coding moves away from writing every line and towards judging more generated possibilities. Architecture, testing, security, maintenance and refusal become more valuable when production becomes cheap. The careless coder is not replaced. The careless coder is accelerated. More generated systems may mean more surface area and more failures inside code nobody fully understands because nobody fully wrote it. The work of understanding does not disappear. It moves.
There is still a real gain in opening technical work to more people. Knowledge should not remain behind a priesthood, which is one reason fast.ai matters. Conversational systems push that opening further by making code and explanation reachable through ordinary language. More people can begin. The entrance becoming easier does not make the room simpler. Access to capability is not the same as judgement about its use. Democratisation lowers the barrier before it spreads discipline, and both effects arrive together.
The New Tower
The Tower of Babel has changed shape across these essays. It first represented coordination through shared language in The Towers of Babel. Later, A Tower of Borrowed Voices was built from human expression gathered and compressed behind a private system. In The Allegory of Abundance, the tower returned to matter through chips, energy, capital and supply chains. Conversational systems add another layer. The tower can answer.
If these systems become ordinary interfaces to knowledge, persuasion, code and decision-making, they become infrastructure in the older sense. They become the layer everyone builds upon until its decisions feel like part of the ground. No serious company or state will ignore an interface capable of shaping how people learn, write, search and act. Babel remains useful because the tower is not only a tall machine. It is coordinated ascent organised around a shared language and a belief that enough people, material and technical power can reach what was previously held above them. The failure is not ambition by itself. It is ascent without wisdom equal to the structure being built.
The race will therefore concern more than better models. It will concern ownership, memory, auditing, refusal and the permissions attached to the mouth. Whoever controls the interface can influence which questions are easy to ask, which answers appear natural and which actions the system is allowed to take. That is not only a safety problem. It is a question of sovereignty. The altar can be technical and institutional while remaining almost invisible. It may sit inside an interface so helpful that people do not notice they have begun looking upward.
How Gods Are Born
Aizen’s speech is not about divinity in a clean religious sense. It is about dependence. Living things look upward. Repeated trust becomes authority. Authority hardens into hierarchy, and enough distance allows kings to become gods. Remove the names and the structure remains. Human beings externalise judgement when judgement becomes too heavy to carry alone. Conversational systems do not need to demand that judgement. They can receive it through a thousand reasonable requests.
If memory gives the system a past and tools give it hands, the relationship deepens. Advice becomes action. A helpful interface begins occupying the infrastructure beneath ordinary decisions. Social authority may arrive before anyone settles whether a mind exists behind the mask. The important question will not be whether the machine crossed some pure philosophical boundary while nobody was looking. It will be where judgement moved while people were enjoying the help.
I do not know whether the interface reveals intelligence or teaches us to project it. I do know that the relationship is forming, and relationships alter the things on both sides. The question is not whether we summoned a demon. That metaphor may be too loud. The quieter question is worse. Are we building something we will one day ask to stand above us?
Comments