The Analogy Engine
Ask a language model to explain something with an analogy. Then ask again, and again, on different days, in different chats. After a while you notice it reaches for the same five or ten comparisons. Memory is a library. The brain is a muscle. The internet is a highway. It does not matter how you phrase the request or which model you use. The well is shallow, and it is the same well every time.
This bothered me long before I could say why. We treat these systems as if they have read everything humans have written, which is close to true, so why does their stock of analogies feel smaller than a clever ten-year-old’s? The answer turned into one of the cleaner ideas I have about what understanding actually is, and where the machine sits relative to us.
We do not trade words
Start with what an analogy is for.
When I want you to understand something, I have a problem. The thing I understand does not live in me as words. It lives as a kind of felt sense, a semantic feeling, a shape I can turn over without naming. Words are not that shape. Words are the handles I reach for to point at it. They are lossy. When I hand you the word, you do not receive my shape. You receive your own decoding of the handle, which may be a different shape entirely.
An analogy is what I do when I do not trust the handle. Instead of passing you the word and hoping you decode it the way I mean it, I look for something I believe is already part of your experience, something that should light up a feeling close to mine, and I say: it is like that. I am not describing the thing. I am trying to recreate my internal sense of it inside you, by routing through experience you already have.
This reframes a lot. A good analogy is not a clever description. It is a successful transfer. And it succeeds only to the degree the referent I picked already means something to you. That is the whole mechanism, and it explains the thing everyone has felt: why the same analogy lands for one person and dies for another. If I explain backpropagation as tuning a guitar, it works beautifully for someone who has tuned a guitar and not at all for someone who has not. The comparison was never good or bad in the abstract. It was matched or mismatched to a particular listener’s lived semantic space.
Which leads to a small, slightly uncomfortable conclusion about communication. Good communicators do not have better words. They have a better model of the other person’s semantic feelings, and they pick referents accordingly. Eloquence is mostly accurate empathy. You are not searching your vocabulary. You are searching theirs.
Interpolation across your own experience
Once you see analogy as transfer, the cognitive move underneath it gets clearer too. An analogy is interpolation, or extrapolation, of personal experience from one object onto another. I take what it felt like to deal with one thing and I stretch that feeling across to a thing I have not dealt with, looking for the pattern that survives the jump. Finding the similar shape in different things.
Notice the load-bearing word in that sentence. Personal. The raw material of analogy is experience you actually had. The guitar analogy is available to me only because I have the bodily memory of a string going slowly into tune, the tiny corrections, the moment it locks in. I am not retrieving a fact about guitars. I am replaying a feeling and reaching it toward something new.
This is the part the machine cannot do, and it is the reason the well is shallow.
Why the machine reuses the same five
A model does not have personal experience to extrapolate from. It has handles. It has the words, in enormous quantity, with exquisitely mapped statistical relationships between them. What it does not have is the felt shape the words were ever handles for. So when you ask it for an analogy, it cannot do the human move of replaying a sensation and stretching it. It can only do the move available to it: retrieve the comparison that most reliably co-occurs with the concept across everything it has read.
That move has a strong gravitational pull toward the average. The internet has already said memory is a library a million times. So the model says memory is a library. It is not reasoning analogically. It is returning the most probable handle-pair, which is, almost by definition, the most common one. Repetition is not a bug in its analogies. It is what its analogies are. The shallowness is a fingerprint of a system that manipulates the handles without ever having held the feeling underneath them.
I want to be careful here, because this is the part where I might be over-drawing the line. The model’s relationships between words are not nothing. They are a genuine, high-dimensional map of how concepts sit near each other, learned from the residue of millions of people who did have the feelings. In a real sense the embeddings are an approximation of the semantic feeling, assembled secondhand from the traces we left in language. But an approximation of the shape, reconstructed from handles, is not yet the shape. It is a map drawn entirely from other people’s directions, by someone who has never walked the streets.
The handles carry their own history
There is one more layer that makes the human side richer than it looks, and it cuts the other way. Even our handles are not empty.
Take the word robot. It comes from a Slavic root for forced labor, the same family that gives us the word for a serf or a slave. So buried inside our everyday word for a helpful machine is an old structure: the machine as the one who does the work it is told to do, the laborer with no say. Most people have never traced that etymology, and yet the association quietly shapes how we feel about robots. We sense something about servitude and obedience in the word before we have reasoned about it.
That is analogy operating below the surface of awareness, compressed into the handle itself and inherited along with it. Our words are not arbitrary tokens. They are fossils of older feelings, and we carry the feeling forward even when we have lost the history. The model has the etymology available as a fact it can recite. What it does not have is the buried sense that arrives uninvited when a human says the word.
The bridge runs both ways
So here is the shape I keep landing on. Analogy is a bridge, and the two sides are not symmetric.
On the human side, we have the feeling and we struggle for the handle. We know exactly what we mean and we cannot find the words, so we build analogies to smuggle the feeling across to someone else through experience we share with them.
On the machine side, it has the handles in abundance and none of the feeling. It can produce fluent comparisons all day, and they collapse toward the same few, because there is no lived shape underneath to generate a new one from.
This is the mirror again, but held at a strange angle. The machine does not break in the same place we do here. It breaks in the opposite place. We are all feeling and poor handles. It is all handles and no feeling. The interesting future is not one side beating the other. It is the bridge itself, the day a system can hold something close enough to the feeling that its analogies stop being retrieved and start being made, matched not to the statistical average but to you, to what you in particular already know. Until then, the tell is simple. When the analogies repeat, you are talking to the handles. When they surprise you, and they fit, something on the other side felt the shape.
Related reading: This is the human-side companion to A Word Is Not a Concept, which argues meaning is built by analogy in the first place. The Measure of Misunderstanding makes the matched-or-mismatched idea quantitative: the gap between two people’s spaces has a size. And The Vector tells the same story from the machine’s side, where the handles live as vectors it can compute on.