There is a kind of knowing that lives in the body. A carpenter feels when a joint is tight before the mallet strikes. A physician notices the stillness in a patient's posture before any symptom is spoken. A parent hears the particular quality of silence in a room and knows, without seeing, that something is wrong.
This intelligence — embodied, contextual, deeply human — is conspicuously absent from every large language model we have built. And I think we are only beginning to understand how much that absence matters.
The Tyranny of the Verbal
We have built our artificial intelligences in the image of our most explicit, legible self. They consume text, produce text, reason through text. Their entire ontology is verbal. Ask them about joy, and they will give you a definition stitched from millions of sentences about joy. But they have never felt the particular weight of a room when everyone is laughing at the same time. They have not experienced the release of tension in their own shoulders when a worry resolves.
Some researchers argue this does not matter. If the outputs are indistinguishable, what difference does the substrate make? But I think this conflates performance with understanding. A parrot may produce sounds that resemble speech without grasping what speech is for.
Context Is Not Compression
A human conversation is not merely an exchange of propositions. It is a dance of gazes, hesitations, vocal timbre, shared history. When two old friends talk, most of the meaning lives in what is not said — in the references they do not need to explain, the emotions they can read in each other's breathing.
Current AI models attempt to reconstruct this context through attention mechanisms and ever-larger context windows. But context in human interaction is not a compressed history of previous tokens. It is a lived, felt continuity. It has texture, temperature, gravity. You cannot tokenize the weight of a glance.
What Remains Ours
None of this is to dismiss the genuine wonder of what we have built. AI systems can synthesise knowledge across domains, spot patterns invisible to human perception, and serve as patient, indefatigable collaborators. These are real capabilities with real value.
But the quiet intelligences — the felt sense of a situation, the moral courage to act without certainty, the creativity that emerges from constraint and mortality — these remain stubbornly, beautifully human. They are not ineffable mysteries, but they are anchored in a mode of being that is fundamentally different from statistical prediction.
Perhaps the most important question is not whether AI will surpass human intelligence, but whether we will recognise the value of the intelligences it cannot replicate. In our rush to measure everything by what machines can do, we risk forgetting what only we can be.