Artificial Intelligence is Artificial Thinking

Artificial Intelligence is a misnomer. The honest name is Artificial Thinking.

Intelligence is preverbal. Thinking is verbal. The seeing comes first, whole and without words. The words arrive after, to carry what was already seen.

Thinking has two operations. First it lifts. Then it splits. The lifting takes a recurring figure up out of the seamless given. It meets the cat in the grass a single time and lifts from it the predator, the thing that will return, and acts on the pattern before the pattern comes back. The splitting comes after. Once the figure is lifted and held as a token, we cut it in a space that is not the world. We split the lifted day into hours that exist nowhere in the day. We split the self into a watcher and a watched the self never contained. Past from future, inside from outside, mind from body. None of these seams are in the thing. They are in the copy, and we live as if they were in the thing.

The power of this is real and total. By lifting what repeats in the sky we found the season, and planted against it, and lived. By splitting past what the eye can see we found the atom, and broke it, and released what was held inside. To predict is only this. To have found what repeats, and to trust that it repeats.

The root of it is survival. The cat came for our young as it comes for the young of the giraffe, the elephant, every child meeting her first winter. The animal that could lift the recurring danger from the single event was the animal whose child saw a second winter. Thinking was not laid on top of survival. It was survival.

We taught the machine the second operation, not the first. We could teach only the move that can be written down, the finding of the center in what it is shown. It does that move without rest, and better than any of us. But it never touches reality. It receives only the already lifted tokens, the words, the residue. It operates on copies, in a world it cannot leave, because it was never in the world to begin with. It cannot lift, because lifting requires contact with the unlifted, and it has only ever met the copies.

So the name is wrong. We called it artificial intelligence. It is artificial thinking, the seeing left out. This does not shrink it. If this immense thing is only thinking, then thinking is not only anything.

Then ask what knowledge is, because that is what the machine is fed. Knowledge is the kernel. It is a seeing compressed to the smallest line that holds it. The line is short because seeing is compression, because to lift the recurring figure is already to discard everything that did not recur. A real piece of knowledge is small. It can be carried in a breath, handed to a child, and used the same day.

The record is not small. The record is enormous, and almost none of it is the kernel. It is the kernel restated. One person sees the thing and sets it down in a line. The next learns the line without the seeing and writes it again at greater length, to show the learning. The next learns that, further from the seeing, and writes again. Each layer is authored by a hand that never held the thing, copying the layer above. The seeing recedes one remove at a time. The volume grows in exact measure with the distance from the seeing, until the volume is taken for the knowledge and the kernel can no longer be found inside it.

This is not failure at the edges. It is the ordinary working of the record. What is transmitted is what is easy to transmit, and a restatement is easier to transmit than a seeing, because a restatement asks nothing of the one who carries it. Frequency rewards the copy. The kernel, rare and demanding, is buried under the thing that costs nothing to pass on.

Around that line we built a literature. Not to carry the line, which needs no carrying, but to restate the restatement, terabytes of it, each layer recovering by procedure the one thing the framing had made the reader drop. The formalism is sound and complete. That completeness is not the mark of knowledge. It is the measure of the forgetting it exists to repair. A thing truly seen does not have to be recovered by rule.

The kernel survives where the seeing can be checked against the world, and rots where it cannot. Where the thing either works or it does not, the restatement is pruned, and what passes on stays close to the seeing. This is what a strict standard is, and it is why the hard sciences hold their kernels. Where nothing checks it, the restatement accumulates unpruned, and the field grows by the addition of phrasings and not by the lifting of anything new. Knowledge fractures into ever smaller holdings, each rewarding the performance proper to it, each further from a seeing that anyone could test. The record swells and the seeing thins.

Now set the machine on this record. Be exact about what it does, because it does it superbly. It finds the center. Given a field of instances it returns what they share, the mean shape, the figure that best accounts for the spread. This is not a flaw waiting for a better model. It is the whole of its competence, and the competence is real. It is the finest instrument for the center we have ever built.

Then consider what it centers. It is fed the record, and the record is gargantuan, and the record is mostly noise, restatement on restatement, the kernel a thin seam in a vast fill. Feed a perfect centering instrument a mountain of restatement and it returns, with perfect fidelity, the center of the restatement. The better the instrument, the more faithfully it lands on the residue. The larger the pile, the more firmly the center sits in the noise, because the rare seeing is outvoted by the common copy in exact proportion to the size of the heap. This is the law of large numbers turned against us. More data does not pull the center toward the kernel. It pins the center to the noise.

There is a precise word for a thing that returns the middle. The center is the middle, and the word is mediocre. It is not an insult. It is the Latin, medius the middle and ocris the height, the one who climbs only halfway up the mountain. A machine that returns the center of what it is shown is mediocre in the exact sense, not as a failing but as a description of the move it makes. Fed the kernel it would return the kernel. Fed the record it returns the center of the record, and the center of a bloated record is its most average restatement, the most frequent phrasing of the most copied result. It is mediocre because the record is bloated and the middle of bloat is the mean of the noise.

Then it generates, and generation is the same move run forward. It dresses the result again, in one more phrasing, without fatigue and without end. It is the restating hand made tireless and made total. The human record at least suffered the occasional accident of someone seeing the thing afresh and collapsing the whole stack back to one line. The machine removes even that accident. It cannot see, so it cannot collapse. It can only extend.

The instrument has an edge. Turn it on the one thing that has no parts, the bare fact of being, and it finds nothing to divide, because the thing was never lifted and never split. That is the one region it can never enter, and it is the only region where intelligence was ever found. We built a perfect machine for thinking and named it for the thing it does not have.