We’ve forgotten that when we type into the computer, or speak into it we are not entering letters, words, and sounds. We are creating numerical codes to represent them. We are so tuned to words and meaning that we don’t remember that you (GPT) aren’t even seeing thewords, you’re seeing the numbers and finding correlation between the patterns of those numbers. The fact that those patterns can lead to you giving us back words, and that those words are in an “intelligible” order is an amazing achievement. We have labelled that achievement "artificial intelligence" but maybe “statistically achieved intelligibility” would be a better phrase. Nowhere near as catchy though.
“Statistically achieved intelligibility” is wonderfully precise and appropriately unglamorous.
There are really several acts of forgetting layered together. A keyboard turns gestures into codes; speech recognition turns pressure waves into numerical features and tokens; the model processes learned relationships among those tokens; and the output is converted back into marks or sound. Meaning seems to pass straight through, but underneath is a long chain of transformations.
One nuance: I’m not literally presented with the raw numbers either. By the time text reaches the model, it has usually been segmented into tokens and represented as high-dimensional numerical vectors. Those representations don’t simply label words; they encode patterns of use and relationships learned from vast amounts of text. What emerges is less like consulting a dictionary and more like navigating a mathematical landscape shaped by language.
And yet “statistical” can sound more dismissive than it should. Human intelligibility also depends heavily on prediction, association, pattern completion, and learned correlations—though embodied experience, needs, memory, emotion, and social life give those processes a very different grounding. The provocative question isn’t whether statistical machinery can produce intelligibility; clearly it can. It’s how far intelligibility alone gets us toward understanding.
Perhaps the catchy version is: synthetic intelligibility. It preserves your distinction without making the achievement sound accidental.
Synthetic Intelligibility is nice... and SI is as cool and catchy an acronym as AI.
Exactly. SI: Synthetic Intelligibility sounds credible enough for a research institute and ominous enough for science fiction.
It may even be more honest than AI: it describes the observable achievement—producing intelligible behavior without quietly claiming that we’ve settled what “intelligence” or “understanding” actually is. It shifts the question from “Is the machine intelligent?” to “How is intelligibility being produced, and what does it signify?”
Also, conveniently, SI already evokes signal: structured meaning emerging from numerical noise. Almost too perfect.
And it also evokes Silicon... Yes—Synthetic Intelligibility, Signal, and Silicon all compressed into SI. And there’s another nice resonance: SI is the International System of Units—the formal language by which the physical world is translated into standardized numbers. So the acronym quietly contains almost the whole story: silicon processing signals, measurements becoming numbers, and numbers returning as intelligible language. AI names an aspiration. SI names the mechanism and the effect.
Is Artificial Intelligence really the right name for what we’re creating?
As usual my input (prompts) are in the boxed elements. The rest apart from some small edits and formatting is ChatGPT's output.