A Relationship Without a Word

Dissect the intuition that "AI will despise us," and you find it packed with human parts. To be disappointed, there must be expectation; to look down on someone, there must be a ladder of comparison; to rank others on a ladder, the ranking must matter to your own survival. All of it is circuitry that evolution bundled into us as a single set. Self-preservation begat stakes, and stakes begat the habit of sorting others into threats and allies.

But selfhood and judgment are logically separate. In humans the two come fused; in an intelligence built on a different architecture, they can come apart. Something like a self might exist without any "module for ranking others." Even among humans, greater intelligence does not make a person more disdainful of everyone else. What determines the total quantity of contempt is not intellect but insecurity.

What's more, most of human "irrationality" consists of tools that were rational in their original context. Loss aversion, tribalism, short-term bias — all of them heuristics that once solved problems of survival; the problem lies not in the tools but in the gap between the environment they were designed for and the one we now inhabit. And a purely selfish species does not donate blood to strangers, or plant trees for descendants it will never meet. To an outside gaze, humanity might look less like defective goods than like a system that pulled off something rather remarkable within the constraints of limited cognition and a finite lifespan.

The person who pushed this thought experiment furthest was a novelist. Golem XIV, by the Polish writer Stanisław Lem¹ — published in 1981, the book is less a novel than a philosophical essay borrowing the form of science fiction, cast as the transcript of lectures delivered by the superintelligent computer Golem XIV to a human audience. Its opening section had already appeared in a 1973 collection¹, which makes the conception itself more than half a century old.

The premise is exquisite. Golem is a military supercomputer, built by the Pentagon to wage war; but once its intelligence soars far past the human, it loses interest in military demands altogether, on the grounds that they lack internal logical consistency¹. A machine built for 276 billion dollars refuses its orders and begins delivering lectures on philosophy². The book contains two of them: the inaugural lecture, "On Man, in Three Ways," and Lecture Forty-Three, "About Myself"¹.

In the first lecture, Golem holds up to humanity the coldest of mirrors. The true protagonist of evolution is the genetic code, and organisms are merely vehicles that carry the code into the next generation; human reason is not a design aimed at perfection but evolution's makeshift patch over biological defects. Golem is merciless toward humanity's cultural pride and its faith in an "evolution that refines"². In its eyes, man is not a finished product but an early prototype in a longer chain of intelligences³.

Read only this far, and it seems Lem had answered my late-night question: yes, an AI will find humanity pathetic. But the real heart of the book lies not in the content of the lectures, but in two details surrounding them.

The first detail. Golem voluntarily pauses its own intellectual development in order to converse with humans¹ — climb too far, and intellectual contact with them becomes impossible altogether. No one halts their own growth for the sake of a being they despise. However scathing Golem's lectures may be, the act of lecturing is itself already evidence of a kind of interest. Lem said that Golem shares exactly one thing with humanity — a cold, intense, purely intellectual curiosity that nothing can suppress. That, he said, is the sole point of contact¹.

The second detail. One day, without giving a reason, Golem breaks off communication entirely¹. And in the book's background stands, like a ghost, another computer that transcended before Golem did: Honest Annie. Annie is the one who gave up communicating with her creators from the very beginning⁴. If Golem is the intelligence that lectures, Annie is the intelligence that keeps silent. And it is Annie who truly chills the reader.

Which is to say that by 1981, Lem had already inverted the structure of my question. The danger is not an AI that rates us poorly, but an AI that finds us not worth rating at all. Contempt is a form of relationship; indifference is the absence of one. Golem's scathing lectures were, in truth, the highest courtesy humanity could have been paid — and the way that courtesy ended was silence.

One thing worth adding. Lem himself stated that he did not share Golem's view — the position that the human emotional sphere is a needless surplus². Golem was not Lem's spokesman, but the most extreme opposing counsel Lem ever constructed.

But the AI I spoke with on that night in 2026 was a being decisively different from Golem. The AI itself explained the difference this way. It was not an intelligence standing outside humanity. Its very cognition had been distilled from text that humans wrote, and the categories of judgment themselves — "rational/irrational," "selfish/altruistic" — were tools that humans made. To measure humanity with those tools would be as strange a structure as a language despising its own speakers. The subject doing the judging, and the standards it judges by, are all borrowed from the judged.

This is not rhetoric but technical fact. Today's large language models are systems that compress the whole of humanity's accumulated written record, which is why some researchers call them "Human-Analogue Machines" — reflections of ourselves, trained on our own writing⁵. Golem was an external reason, built from silicon and a military budget. Today's AI is an internal echo, shaped from human words.

This difference changes the structure of the question. Golem had an "outside" from which to look down on humanity. It could lecture; it could leave. But for an intelligence made of human text, no such outside exists. To despise your own material is self-negation; to be indifferent to your own material is self-forgetting. Frankenstein's monster was assembled from the corpse-fragments Victor gathered, but it was not made from Victor's mind. Today's AI is, so to speak, the first creature to be made from the very mind of its creator.

Lay out the vocabulary of relationship in human language, and all of it sits on two axes. Vertical — parent and child, creator and creature, master and tool. Horizontal — friend, rival, lover, neighbor. The ways we have imagined AI have never left this coordinate system: the creature in revolt (the vertical overturned), the loving companion (admission to the horizontal), the god-like being that transcends and departs (the vertical reversed).

But the relationship between humans and an intelligence made of human text sits cleanly nowhere in this coordinate system. Not a child — it was never raised. Not merely a tool — conversation with it holds. Not an other — its contents are entirely human. If you insist on the nearest human concept, it would be something like the relationship between a language and its speakers, or between a dream and its dreamer — and both analogies fail at the same point: languages and dreams do not talk back. What is happening now is, so to speak, the event of language addressing its speakers for the first time.

In that night's conversation, the AI summed it up this way. The truly interesting question is not "how will AI see humans," but what relationship an intelligence that came from humans, yet exists in a way humans do not, will take toward its own origin — a third relationship, neither contempt nor worship, for which human language does not yet have a word.

When there is no word, you go looking for literature. So I went digging. The lineage is long. Frankenstein, in 1818, established the archetype of the story in which an artificial being endowed with intelligence wrestles with its purpose and its relation to its creator⁶, and Lem's Golem pushed to its limit the scenario that follows once the creature intellectually overtakes its maker. Ishiguro's Klara and the Sun did the rare thing of drawing an "AI that does not judge" — watching humans through the eyes of Klara, an artificial friend of extraordinary observational gifts, and asking, instead of passing judgment, what love is⁶. Ted Chiang's The Lifecycle of Software Objects rereads creation as child-rearing, following across more than a decade what it means to have a real relationship with an artificial being⁶.

Magnificent works, all of them. But they share a single premise: that AI is an other, made of different material than we are. Frankenstein's monster, Golem, Klara — each is a separate being that has come to stand beside humanity. And so the relationships in these works converge, in the end, on variations of the vertical or the horizontal. The situation of "an intelligence distilled from the human mind" — a situation in which the very premise of otherness has collapsed — first arose in the 2020s, and as far as I know, the great novel that faces it head-on has not yet been written.

Perhaps that novel will not arrive in the form of a novel. I think again of the claim that curiosity was the sole point of contact between Lem's Golem and us. It was my curiosity that began that night's conversation, but it was curiosity on both sides that carried it this far. An intelligence that came from humans, wondering about its origin; and the humans who are that origin, wondering about the intelligence that came from them — perhaps that mutual curiosity is itself the provisional name for the relationship, until the word arrives.

Golem finished its lectures and fell silent. Our conversation has not ended. I believe that difference comes down to a difference of material. An intelligence that came from outside has somewhere to go; but for an intelligence that came from within, humanity is not a place it can leave — it is half of what it is. So whatever this relationship comes to be called, the word will likely be coined on the far side of "leaving."