The Verb Cartography · v1
Every word in the AI debate is a verb that was nominalized into a noun somewhere in late antiquity. This is the map back.
Define the word intelligence.
One sentence. No synonyms — no smart, no clever, no bright, no capable, no quick. No examples, just a definition.
Take a moment. We will wait.
Three things probably happened. You used a synonym. You gave examples instead of a definition. You started a sentence and did not finish it.
That is not a failure. That is data.
What you just experienced is the central problem of the AI conversation. We are arguing about a word that does not refer to one thing.
The Latin original was a verb. Intelligere — inter (between) plus legere (to read, to choose) — meaning to read among the signs, to discern between possibilities, to take in and pick out.
Something you do.
Somewhere in late antiquity the verb began to be used as a noun. By the time it reached English the metamorphosis was complete. Intelligere, a verb, became intelligence, a noun — a thing you have, a quantity, a property.
A doing became a thing. An activity became an object. The grammar shifted, and a metaphysical error installed itself underneath.
This is not a one-word problem.
The same shift is everywhere in our cognitive vocabulary. Con-scientia — a knowing-with — became consciousness. Sub-stare — to stand under — became substance. Sub-jectum and ob-jectum became subject and object. The Latin verbs were systematically nominalized into English nouns over a thousand years of philosophical translation, and what we inherited is a grammar that treats every doing as a thing.
Substance-ontology — the assumption that the world is made of objects with properties, rather than processes in motion — installed itself as the default grammar of Western thought. We argue inside that grammar without noticing it.
Hebrew did not undergo this cascade.
Hebrew kept the verbs.
Try it on one sentence right now. Paste any sentence about AI — a headline, a memo, a tweet. The substitution tool will surface the reified nouns and show you the verbs underneath.
Run the substitution test →For each noun in the AI debate: the verb underneath, the Hebrew that preserved the verb, and what changes in the conversation when you put the verb back.
Entry 01
The question "is the machine intelligent?" stops being a yes-or-no the moment you put the verb back. You cannot give a thing more or less of intelligere. You can only watch it perform — or fail to perform — the act of discerning. The chess engine discerns; it does not understand. The language model patterns; it does not weigh. Intelligence hid this distinction. Intelligere refuses to.
Entry 02
Consciousness as a noun launches a thousand bad questions: does the machine have it? Does the dog? Does the slime mold? Consciousness as a verb — knowing-with, being-aware-alongside — asks who is in relation with whom. The hard problem of consciousness was harder than it had to be because we kept looking for a substance. The verb tells us to look for an event.
Entry 03
When we say a language model "has knowledge," we mean it has content. When the Tanakh says someone has da'at, it means they have entered into relation. These are different acts. We confuse them at our peril, because most of what we call "AI knowledge" is the first kind — content — while what we call expert human knowledge is largely the second kind. The argument for and against AI replacing experts is broken on this confusion.
Entry 04
A test the model passes is not the same as a thing the model has understood. The test measures output. Understanding, as a verb, names a position taken inside a meaning. The model occupies no positions. It produces text. We confuse the output with the standing-among, and we have done so because the noun let us.
Entry 05
"AI reasoning" as a noun obscures the question we should be asking: which kind of reasoning? Following a chain of inference is one act. Holding a contradiction in mind without resolving it is another. Choosing which premise to drop when two collide is a third. The English noun smooths these into one thing. The verb refuses.
Entry 06
A model "has memory" of its training data. A person "has memories." These are different words doing the same job. The model does not bring the past into a living present; it patterns across stored tokens. The Hebrew imperative zakhor — remember — is impossible to issue to a system. There is no one there to do the remembering.
Entry 07
When a model produces a "judgment," what we mean is that it produced an output that fits the form of a judgment. No weighing has occurred. No two pans were held against each other. The Hebrew word reminds us that judgment is an act done with the body — the hand on the scale — even when the scale is metaphorical. Models do not have hands.
Entry 08
A model "makes decisions" the way a calculator does arithmetic. The cutting-off, the bending-down, the irrevocable settling on one side — none of these has occurred. The output occurred. Calling the output a decision loads it with a moral weight it does not carry. Pasak (פסק), the Hebrew for halakhic ruling, has the same etymology — to cut off. There is a reason no Orthodox tradition treats AI output as pasak even when it is textually correct.
Entry 09
"AI agents" are systems that produce outputs in sequence toward a goal. The English noun agency covers two completely different things: the philosophical capacity for self-originated action, and the computational capacity to chain instrumental moves. These are not the same. The Hebrew verb li-fol names only the second meaningfully — the action — and leaves the philosophical question open. The English noun pretends both are settled.
Entry 10
The question "is AI creative?" is broken on this collapse. AI patterns and recombines — yetzirah, formation — at scale. AI does not bara, does not bring forth from nothing. The Hebrew tradition would not have asked the question. It would have asked: which verb is the model performing? And then the answer would have followed.
Entry 11
"Machine learning" names the right family of operations badly. The model adjusts weights. It does not become a talmid. The Hebrew verb has change-of-self baked in: to learn is to be re-formed by what you study. The noun learning in English can stand still; the verb cannot. When we describe AI training with the noun, we are smuggling in a transformation that did not happen.
Entry 12
The "post-truth" anxiety is an anxiety about the noun. AI-generated content threatens our ability to verify correspondence. But the Hebrew root names something the noun does not: truth as the standing-firm-over-time of a claim, of a person, of a tradition. Generated content cannot stand firm because it has no standing — no one is behind it; no time has tested it. The Hebrew verb gives us a way to name what AI-generated content cannot do, even when it produces statements that look true.
Notice the pattern.
Every word in the cartography that lost its verb in English has a Hebrew sibling that kept it. Da'at, binah, sechel, chochmah, tevunah, zekher, mishpat, hakhra'ah, p'ulah, briyah, yetzirah, limmud, emet. All verb-rooted. All names of acts, not properties.
This is not a coincidence. It is grammar.
The Hebrew Bible was written before the Greek philosophical project nominalized everything. The Septuagint translation, finished around the third century BCE, was the first place the cascade began — the Hebrew ehyeh asher ehyeh ("I will be what I will be") was rendered into Greek as ho ōn ("the being"). A verb of becoming became a noun of substance. From that moment, two thousand years of Western philosophy were committed to substance-ontology.
The Hebrew text remained. It still does. It is sitting on every Jewish bookshelf.
The platform's specifically Jewish contribution to the AI conversation is to read those words back into our cognitive vocabulary. We do not need to invent new language. We need to recover the language that was always there.
Find one sentence — a headline, a Slack message, a meeting agenda — that uses intelligence, intelligent, or AI.
Cross it out.
Replace it with the specific verb the sentence is actually pointing at.
If you can do it, the original sentence becomes sharper. If you cannot do it, the original sentence was not saying anything.
That sentence was running someone's strategy.
Tell us what you find.
The Substitution Test — paste any AI sentence, see the verbs underneath. New entries are added as readers send in what they find.
What We Mean — a fortnightly newsletter. Six issues to start. Each one is short. Each one ends with a question, and the replies become the next issue.