The right to read and write
For most of human history, the law was whatever the man with the most power remembered it to be. A decree lived in the ruler's mouth and died with him; it could be revised between dinner and dawn, and no one outside the room could check the difference. Rule by whim is not an aberration of governance. It is the default state of power — the thing that fills the vacuum whenever nothing durable is written down. Everything we call governance is the long, uneven project of climbing out of that default. And every rung of that climb has been a rung of literacy. The claim of this essay is simple and, I think, exact: there is no durable governance without a durable ability to read and write. Strip a society of the written, referenceable, auditable record, and it falls back to whim. Give a society that record — and the ability to read it — and you get canon.
Representation outran custody the moment we could write
Around 3200 BCE, in the floodplains of Sumer, accountants pressed marks into wet clay to track grain and livestock, and in doing so invented writing. It is worth dwelling on the fact that writing was born as bookkeeping — not poetry, not scripture, but the ledger. The first thing humans wanted to make durable was who owed what to whom. That origin is not a footnote; it is the whole thesis in miniature. To write is to externalize memory so it can outlive the one who holds it.
But the same move that makes a record durable also detaches it from its author. A clay tablet outlives the scribe; it travels to a city the scribe never saw; it speaks for the dead. Representation began to outrun custody. This is the permanent tension of the literate world, and a deep-research timeline I built last winter gave it two names borrowed from older traditions. The first is Māyā — a Sanskrit term for the power of representation and illusion: symbol, model, abstraction, the map mistaken for the territory. The second is Ma'at — the ancient Egyptian principle of truth, order, and right accounting: the demand that what is represented stay answerable to what is real. Human history, on that reading, is the story of representation scaling faster than accountability. Writing is the first great leap of Māyā. Everything since has been the slow construction of Ma'at to catch up.
The stele: law that outlives the king
The first time accountability caught up, it was carved in stone. Around 1750 BCE, Hammurabi's code — 282 laws in cuneiform on a hard stone stele over two meters tall — was set up in public so that, in the king's own framing, the wronged could read the law that protected them. Read past the brutality of its specific judgments and the structural novelty is staggering: the law was now fixed, public, referenceable, and outside the ruler's mouth. A king could still be unjust, but he could no longer be quietly unjust, because the standard was written where anyone literate could point to it. This is the first ledgered accountability — Ma'at rendered operational. The stele does for justice exactly what the Sumerian tablet did for grain: it makes the obligation durable and checkable.
Egypt gave the impulse its purest image. Ma'at was truth, balance, and cosmic order, and at death the heart was weighed against her feather — a moral ledger in which what you were said to have done had to balance against what you actually did. Strip away the theology and you are left with double-entry bookkeeping's founding intuition three thousand years early: every entry must reconcile, and a record that does not balance is wrong.
Plato's warning and the long middle
The literate world also learned, early, to fear its own machinery. In the allegory of the cave, Plato described prisoners who mistake shadows on a wall for reality — a warning that systems can come to act on representations of the world rather than the world itself. He was naming the failure mode of Māyā: the model captured, the map mistaken for the territory, governance conducted over shadows. Every administrative abuse since has been a variation on it.
The middle of the story is a tug-of-war between the two forces, and literacy is the rope. Each expansion of the written record expands both Māyā and Ma'at at once. Magna Carta in 1215 was a written instrument that, for the first time, bound a sovereign to terms he could be held to — accountability entering governance as text. Then in the 1450s Gutenberg's movable-type press collapsed the cost of copying, and within fifty years Europe's presses had produced more than twenty million volumes. Scripture, then law, then the constitution itself became mass-readable. Governance escaped the scribal priesthood. The franchise of the written became the franchise of the governed — you cannot consent to, or contest, a rule you cannot read.
And in 1494 a Franciscan friar named Luca Pacioli printed the first systematic description of double-entry bookkeeping. The rule was arithmetic and absolute: every transaction touches two accounts, debits must equal credits, and if at the close of the day the books do not balance, the books are wrong. Egypt's weighed heart had become a portable, teachable, enforceable discipline. It held for five centuries across every currency and corporate form humanity invented, because it is not really an accounting technique — it is accountability expressed as a closure condition. This is the deepest ancestor of what we now build.
Governance is a literate act
Pull the thread tight and a definition falls out. Governance is not goodwill, and it is not the wisdom of the person in charge. Governance is a literate act: a contract that is (1) written once, (2) read by all the parties it binds, (3) referenced when a decision is made, and (4) audited and amended in the open. Remove any one of those and you slide back toward whim. A rule no one wrote is a mood. A rule no one can read is a secret. A rule no one references is decoration. A rule no one can audit is a lie waiting to be told. The durability of a society's governance is, almost literally, the durability of its written record multiplied by the breadth of its ability to read.
The scientific method is the same move applied to truth: a claim is governed when it is written down precisely enough that someone else can check it against reality. Shannon's 1948 information theory was the move applied to communication itself — all meaning rendered as bits, infinitely copyable, infinitely portable. Māyā reached its industrial maximum. And the twentieth century's bureaucracies, censuses, and billing codes proved Plato right at scale: a person became a file, and the file came to govern the person. Representation had never been more powerful, and accountability had never been further behind.
Why machine learning could not govern
For about a decade, machine learning could see but not read. It classified tumors, predicted clicks, optimized routes — extraordinary perception, all of it operating below the threshold of language. And a system that cannot read and write language cannot govern, for the same reason a pre-literate society cannot: it can execute, but it cannot be held to a written contract, and it cannot author one. You can optimize a system with a perceptron. You cannot ask a perceptron to read the canon, apply it to a case, explain its reasoning against the rule, and propose an amendment. Governance is a literate act, and the machine was illiterate.
In 2017 that changed. The transformer and the large language models that followed it (GPT-3 made the leap unmistakable in 2020) are the first machines to cross the literacy threshold — the first that read and write the canon rather than merely perceiving the world. And that, precisely, is why governance itself — not just execution — became automatable for the first time. A machine that reads the rules, applies them, writes the record, and proposes the edit can be placed inside the governance loop, not just at the end of it.
Prompting is the new whim
Here is the trap, and it is the oldest one. Crossing the literacy threshold does not, by itself, deliver canon. It delivers the capacity for canon — and the capacity for whim at a new scale. A prompt is a decree spoken into the model's mouth. It is personal, ephemeral, and unauditable; it lives for one context window and dies when the session closes; it can be revised between one message and the next, and no one outside the chat can check the difference. Prompting is rule by whim. It is Māyā at machine scale: fluent, confident representation with no lineage and no accountability — exactly the failure Plato warned about, now generating its own shadows on demand. And it is as dangerous and as counterproductive to a society's development now, with agents, as the strongman's spoken law was then. We did not escape the default state of power by inventing a more eloquent way to return to it.
The cure is the same one the stele introduced four thousand years ago: write it down once, make it public, make every change leave a trace. Do not prompt the agent. Govern it — with a durable, readable, referenceable, auditable canon that the machine reads on every run and writes back to under the same rules it enforces. The thing that moved us from whim to canon for humans is the same thing that must move us from prompting to governance for machines. There is no shortcut, because there was never a shortcut: no durable governance without durable read and write.
CANONIC: Ma'at as a programming paradigm
This is the arc that CANONIC completes. The governing facts of the system live once, in canon — written down, in the open, where any party can read them. The build is a traversal that reads that canon and enforces it; the gates are double-entry's closure condition made executable, refusing to ship a state that does not reconcile. The agent does not receive prompts that vanish — it reads the canon, applies it to the work, writes its reasoning into a hash-chained evidence ledger, and proposes amendments back into the same canon that governs it. Editing the canon edits the compiler that compiles you. That is why, as we have argued before, the governance tree is not compiled — it is the compiler, and why the discipline is to stop prompting and start governing.
In the older language: CANONIC is what happens when Ma'at is expressed as a programming paradigm rather than a moral aspiration. Plato diagnosed the shadows. Hammurabi fixed the law in stone. Pacioli made accountability reconcile. The press made it readable by all. And the machine that can finally read and write the canon at scale is the first one that can keep the books for itself — if we hold it to a durable record instead of a disappearing prompt. Truth does not survive scale unless governance is first-class. We have known that, in one form or another, since the first scribe pressed the first mark into clay to record a debt that would outlive him.
Primary publication: hadleylab.org/blogs. Content-hash-addressed and ledgered per the Canonic publication model; peer-reviewed venues are downstream syndication. The historical spine of this essay draws on a representation-versus-accountability (Māyā/Ma'at) timeline first assembled in a deep-research session in January 2026.
Sources
| Claim or named entity | Body anchor | Source URL |
|---|---|---|
| Writing originated as accounting in Sumer (~3200 BCE); tokens to cuneiform | §representation outran custody | https://sites.utexas.edu/dsb/tokens/the-evolution-of-writing/ |
| Code of Hammurabi (~1750 BCE) — 282 laws on a public stele | §the stele | https://www.britannica.com/topic/Code-of-Hammurabi |
| Ma'at — Egyptian principle of truth, order, and the weighing of the heart | §the stele | https://www.britannica.com/topic/Maat-Egyptian-goddess |
| Plato, allegory of the cave — systems acting on shadows | §Plato's warning | https://en.wikipedia.org/wiki/Allegory_of_the_cave |
| Magna Carta (1215) — written limit binding the sovereign | §the long middle | https://www.britannica.com/topic/Magna-Carta |
| Gutenberg movable-type press (c. 1450s) and the spread of literacy | §the long middle | https://www.britannica.com/biography/Johannes-Gutenberg |
| Luca Pacioli (1494) — first printed description of double-entry bookkeeping | §the long middle | https://www.britannica.com/biography/Luca-Pacioli |
| Shannon (1948) — "A Mathematical Theory of Communication" | §a literate act | https://doi.org/10.1002/j.1538-7305.1948.tb01338.x |
| Vaswani et al. (2017) — "Attention Is All You Need" (the Transformer) | §why ML could not govern | https://doi.org/10.48550/arXiv.1706.03762 |
| Brown et al. (2020) — "Language Models are Few-Shot Learners" (GPT-3) | §why ML could not govern | https://doi.org/10.48550/arXiv.2005.14165 |
| The Compiler Insight — governance is compilation | §CANONIC | https://hadleylab.org/blogs/2025-12-29-the-compiler-insight/ |
| The Evidence Chain — hash-chained provenance | §CANONIC | https://hadleylab.org/blogs/2026-02-07-the-evidence-chain/ |
| The Governance Tree Is the Compiler | §CANONIC | https://hadleylab.org/blogs/2026-05-21-the-governance-tree-is-the-compiler/ |
| Stop Prompting, Start Governing | §CANONIC | https://hadleylab.org/blogs/2026-03-30-stop-prompting-start-governing/ |
| Canonical post at hadleylab.org | §primary publication | https://hadleylab.org/blogs/2026-05-29-the-right-to-read-and-write/ |