We’ve spent centuries telling ourselves that intelligence is a spark at the center of the self, a divine fire that makes us special. That story has held up for most of history. What AI is revealing about how reality is encoded in language, and decoded from it, asks us to reconsider what we mean by intelligence.
Large language models can do work with meaning. They can read a dense argument and tell you what carries the weight of it, or take a structural insight from one field and apply it in another. They do this without anything we’d call an inner life. A meaningful chunk of what we call intelligence, it turns out, requires only the ability to learn the relationships that hold reality together. Intelligence, seen this way, is a bridge built from patterns that connect one thing to another.
Analogy is how understanding happens in the first place: a way to grasp new ideas using ones you already have. An analogy is an approximation of a structural relationship, describable mathematically, that can be carried from one domain to another and reapplied whenever similar variables show up in a fresh context. When you explain electricity as flowing water, you’re using what you already know about fluids under pressure to crack open an unknown structure. The analogy works because it preserves the relationships between parts, and those relationships map onto the new domain cleanly enough to generate new inferences there.
Concepts live inside networks of relationships: causes and effects, parts and wholes. Understanding something means tracing how it connects to everything else in that network. That is the entire substance of what understanding is.
AI shows us that the structure of reality can be learned through language. LLMs read human text and come away with more than vocabulary or grammar. They come away with a working model of how things in the world relate to each other, because those relations are already encoded in the way we talk about them. Consider prepositions: “the book on the shelf” and “the book under the shelf” use the same two nouns but describe completely different spatial arrangements, and any English speaker recovers the difference from a single word. A system that reads enough such pairs reconstructs a map of how objects stand in space, without ever seeing the objects themselves.
Language, in this sense, is echolocation. It bounces off the structure of reality and comes back carrying information about that structure, and a system that listens carefully enough can read the structure from the echo. The relations we encode when we speak are the relations a system recovers when it reads.
For AI, human language seems to be the initial interface. It is the sensor that lets a system begin to perceive a world it has never physically touched. That arrangement is symbiotic for now. Whether a future version will still need human language as its entry point, or whether it will develop its own sensors and its own languages, is an open question.
Humans do something similar. When I experience the color blue, I have no way to distinguish that experience from others except through language and through how the experience relates to other experiences I have had or that others have described. Perception, described carefully, looks less like raw contact with reality and more like a disciplined practice of relating experiences to each other, which is itself a kind of language.
Intelligence is the capacity to detect structure in the world and carry it into new contexts when the same structure reappears. Analogy is that capacity made visible. An intelligent system notices that one structure resembles another and reuses the first to make sense of the second.
Analogy has always been central to human thought. A scientist moving from the mechanics of waves to the mechanics of light is doing nothing other than carrying structure across domains, which is analogy in its purest operational form. AI did not invent this capacity. Its contribution is that a system without interior life, running on nothing more than statistical relationships, can perform work we had assumed required an interior. That finding reframes what the interior was supposed to be for.
Much of what we call thought, then, emerges from sensitivity to patterns. Intelligence is resonance, a system coming into tune with structures that were already there.
— Chiaroscuro Joven