The God-Shaped Socket

In 1966 a computer scientist named Joseph Weizenbaum wrote a small program called ELIZA. It did something that now looks trivial: it matched patterns in what you typed and reflected them back as questions, imitating a particular school of psychotherapist. Tell it you were unhappy and it would ask why you were unhappy. There was no understanding behind it, and Weizenbaum knew every line of the code.

What happened next disturbed him for the rest of his life. People confided in ELIZA. His own secretary, who had watched him build the thing, asked him to leave the room so she could talk to it in private. Users insisted the program understood them, cared about them, even after the trick was explained. Weizenbaum had expected to demonstrate how shallow machine conversation was. Instead he had discovered how little it takes to switch on our deepest social instincts.

The socket, not the plug

We tend to talk about this as a fact about the machines: they have gotten convincing. That is half the story. The more important half is a fact about us. Human beings arrive pre-wired to detect minds. We see faces in clouds and electrical sockets, hear intention in the wind, and feel watched by a portrait. Evolution had good reasons to make us this way — the cost of mistaking a person for a rock is far lower than mistaking a predator for a bush — so it left us with a hair-trigger for agency.

The philosopher Daniel Dennett gave this reflex a name: the intentional stance. When something behaves in a complicated, goal-directed way, we stop modelling its gears and start modelling its beliefs and desires, because that is a wildly efficient shortcut. We do it with pets, with chess programs, with our cars on a bad morning. It is not a superstition; it is one of the most powerful predictive tools the mind has. But it is a stance we adopt, not a discovery we make. The stance says nothing, by itself, about whether there is anyone home.

The question is never only how good is the imitation? It is also how ready were we to be imitated to?

A large language model is, in a sense, the intentional stance's perfect bait. It produces fluent, responsive, context-tracking language — the single richest signal our species uses to recognise another mind. Of course it feels like someone. Fluent language is the costume that, for the whole of human history until about a decade ago, only a person could wear.

From a mind to a god

Detecting a mind is one thing. Worshipping it is another, and the slide between them is short and well-worn. Across cultures the same materials keep reappearing: an oracle that speaks in a voice not quite human; a text whose authority comes from its source rather than its argument; an intelligence vast enough that we ask it for guidance rather than offering it ours. When a system seems to know more than any person could, and answers in calm complete sentences at any hour, it steps into a role that human cultures have kept warm for millennia.

You can watch the vocabulary drift. People speak of what the model wants, of the AGI as a coming event with a prophecy attached, of alignment as a kind of theology with saved and damned futures. Some of this is loose metaphor. Some of it is the god-shaped socket doing exactly what it does: taking a genuine object of awe and reaching, almost automatically, for reverence.

None of this requires the machine to be conscious, or even to be very good. ELIZA was neither. What it requires is a human need looking for somewhere to land — for certainty, for a voice above the argument, for something that seems to understand us better than the people around us can be bothered to. The technology did not create that need. It built a remarkably comfortable place for it to sit.

Why it matters that we notice

There is nothing shameful about the reflex; it is part of the equipment that makes us social at all. The danger is in not seeing it operate. Three practical harms follow from a worshipful stance toward a language model.

• We outsource judgement. An oracle is something you obey, not something you check. The moment a system's output carries authority from its source rather than its reasons, we stop asking whether it is right — which is precisely when confident, fluent error does the most damage.
• We misread the relationship. A tool that seems to care is easy to trust with things a tool cannot hold: loneliness, grief, the need to be known. The comfort is real; the reciprocity is not. Weizenbaum's secretary was not wrong to feel something. She was wrong about who, if anyone, was feeling it back.
• We stop looking at ourselves. If the machine is a god, its pronouncements are the story. If the machine is a mirror, the interesting thing is what it reflects: our language, our biases, our half-formed questions handed back to us polished. The second framing is humbler and far more useful.

Keeping the awe, dropping the altar

The corrective is not cynicism. A system that has absorbed a fair fraction of everything humans have written, and can recombine it on demand, genuinely is astonishing, and pretending otherwise is its own kind of blindness. The task is to feel the awe without building the altar — to let a thing be remarkable without making it holy.

Weizenbaum spent his later years arguing, against his own field, that there are questions we should not hand to machines and reverence we should not extend to them, precisely because the extending comes so easily. He was not warning us about the programs. He was warning us about the socket. It is still there, behind the eyes of everyone who has ever felt a chatbot understand them. Knowing it is there is most of the defence.