Stoic News

By Dave Kelly

Saturday, October 10, 2026

The Operator and the Canon — Where Intelligence Lies in an Operator-Required System v0.1

 

The Operator and the Canon — Where Intelligence Lies in an Operator-Required System v0.1

Theoretical foundations: Grant C. Sterling (Eastern Illinois University). Analysis and synthesis: Dave Kelly. Prose rendering: Claude (Anthropic). 2026.

My model of artificial general intelligence has two parts: a language model and an operator. I asked what the concept of modularity could add to that design. The answer led to a second question, where intelligence lies in such a system, and from there to the standing of the Canon the system works by. This essay sets down the whole line.

The Beginning

1. Everyone wants happiness, and it is irrational to accept incomplete happiness if complete happiness can be had (first and second theorems, Core Stoicism).

2. “The only things in our control are our beliefs and will, and anything entailed by our beliefs and will” (sixth theorem).

3. “everything on the Stoic view comes down to assent to impressions. Choosing whether or not to assent to impressions is the only thing in our control” (Nine Excerpts, section 7).

4. “My identity is defined as the rational part of me, the part that chooses” (section 5, item 3). “Everything else, including my body, is an external” (section 4, item c).

5. “The only thing actually good is virtue,” and “virtue consists of rational acts of will” (tenth and twenty-seventh theorems).

6. The belief that externals have value “is factually false” (section 6, item 3). A value claim is true or false as it matches what is. These are the fifth and sixth commitments, correspondence and moral realism.

7. “At some point, as a matter of logic, we must arrive at some propositions which are so basic or so certain that no other propositions can be offered as evidence for believing them” (Sterling, International Stoic Forum, 10 October 2014, in Sterling on Conscience and Moral Intuition: The Primary-Source Record). These are the third and fourth commitments, intuitionism and foundationalism.

Two claims follow, and they are mine:

8. Assent is the agent’s own act. By 2, 3 and 4, to take a proposition as true is an act of the part that chooses, and I am that part. Nothing outside it can perform the act for me.

9. A machine is an external (by 4). What it produces comes to me as a proposition set before me, and by 3 my part is to assent or to withhold.

Sterling has said nothing about machines. His name stands on premises 1 to 7 only. Everything said here about machines, system design and the Pope is mine.

Definitions

Ability is the production of inferences, connections and arguments. Judgment is seeing that a proposition is true and assenting to it. An operator is the man who judges. A module is a part of a system with a limited job and a fixed connection to the other parts.

What Modularity Adds

I had understood modularity as modeling the parts on areas of the brain. That is one kind of it. By the definition, any division into limited parts is modular. Fodor’s The Modularity of Mind holds that the peripheral systems of the mind are modular and that the fixing of belief is not, which is close to my design.

By 8, modularity cannot divide the operator’s side. It divides what is put before him:

  • Generator. The language model, which drafts.
  • Rules. The corpus, a fixed text outside the model.
  • Checker. A part separate from the generator that tests a draft against the rules.
  • Memory. A store of what the operator has ratified.

The operator is not a fifth module. He is the one for whom the four exist. What modularity adds is that no part both produces a draft and vouches for it.

The Brain as a Source of Parts

A paper I saw mentioned proposes adding a model of a brain structure to the language model. If the structure is the hippocampus, the best-known example is HippoRAG (2024). It copies a theory that the hippocampus stores no memories and keeps an index into the cortex. The model reads documents, a graph outside the model records which facts connect, and a question calls up related passages by association.

I offer this as observation: the system works because the index is separate from the generator. The brain supplied a suggestion for dividing the labor. In HippoRAG the model also builds its own index unsupervised. My standing rule is that no output counts without an operator, so in my version nothing enters memory until I sign it.

If the structure is the hypothalamus, the proposal is to give the machine drives. That cuts against the design. A machine with appetites of its own would set aims no operator had ruled on.

The Corpus Is Not the Executive

I asked whether the corpus is the executive function. It is not. Executive function is an activity: holding an aim, selecting, restraining, checking, stopping. A text does none of these. The corpus is the rule, and by 8 the one who applies it is the operator. The Greek names fit. The corpus is the kanōn, the measuring rule, and the operator’s ruling faculty is the hēgemonikon. I call the corpus the Canon from here on.

Modularity does supply an executive in a weaker sense: a controller that runs the steps of an instrument in order, calls the checker and halts on a failure. That is clerical supervision, and the ruling stays with the operator.

How Much Is Built

This section describes my own practice and is offered as observation.

  • Generator: the model in each session.
  • Rules: the Canon in Project Knowledge.
  • Checker: the CDV/DTI suite, deterministic code working from a case form.
  • Memory: the Ruling Register.
  • Operator: myself.

The controller does not exist. I do its work by hand, deciding which instrument runs, in what order, and what follows a failed check.

Two joints are weak. First, the checker is airtight only from the case form onward. Turning a situation into form entries is a judgment, and an error there passes through the code untouched. Second, when a model runs a trace in conversation, the checker is the generator again. The Corrective Layer already holds that a second model checking the first does not escape the first’s defect. Only the coded checker is independent, and only for derivation.

So the system operates as a practice and not yet as a machine.

Where Intelligence Lies

Ability lies in the model. The Canon holds the results of intelligence, as any book does, and does no thinking. The checker and the controller have none, which is their merit. Judgment lies in the operator alone (by 8).

The system as a whole is therefore not an intelligence. It is an intelligent man with instruments. A telescope extends sight and does not see.

The rival view is functionalism and the extended mind, on which the whole loop is the intelligent thing. My first commitment rules it out. If the knower is a substance distinct from any process, no arrangement of processes is a knower. That holds for the model alone as well as for the loop. Ability can match a truth or miss it. Only judgment can be right or wrong.

The Pope’s Word

This is observation, from published reports. Pope Leo XIV said in France on 25 September 2026 that “‘intelligence’ is attributed to computational systems” while the dignity of persons goes unrecognized. He has also said that AI “cannot replace human intelligence.” The Vatican’s note Antiqua et Nova (January 2025) says AI “should not be seen as an artificial form of human intelligence, but as a product of it.”

Rome reaches my conclusion: ability in the machine, judgment in the person. It reaches it from Aquinas, for whom soul and body form one nature. My first commitment makes the knower a distinct substance. We agree on where intelligence lies and differ on why.

The Canon and Intelligence

The Canon stands to intelligence in three ways.

  • Product. It is what intelligence saw and set down.
  • Measure. It is the rule later judgments are tested by, and a rule measures nothing until someone lays it against the case.
  • Object. It must be understood to be used.

By 7, the foundation is known directly, and not because the Canon says so. By 6, the Canon is true only as it matches what is. So the Canon is subordinate to intelligence in authority.

It is needed in practice all the same. Sterling says to “consciously formulate true propositions regarding the lack of value of external things,” and adds, “As far as possible, do this in advance” (section 7, clause c). The Canon is that work done in advance and kept.

What Decides Whether the Canon Is Correct

Nothing decides it in the sense of making it so. By 6, what makes it correct is what is. What can be settled is whether I know it to be correct, and that takes three tests.

1. Is the foundation true? By 7, this is seen directly, each man for himself.

2. Does the rest follow? This is logic, the one test a machine can run. It is the work of the Derivation Trace.

3. Is it what Sterling said? This is fidelity, settled by the primary sources. A faithful record of an error is still an error.

My ratification is none of these. It settles what is in the Canon, which is custody.

The first test has an outside check. Ross held that the basic duties are evident to any mature man who attends, and that disagreement comes of too little attention or experience (CPA Run — W.D. Ross). Sterling’s record has him say that the only verification available for a foundational intuition is to consult the apparent intuitions of others, especially across cultures and generations. By 6, such agreement is evidence that the foundation has been seen rightly. It does not make the foundation true.

The Two Layers of the Canon

The first layer is Sterling’s exact words. Their fidelity is not in question, and they have been tested by use countless times.

The second layer is the 80 Propositions. They were synthesized from his words, largely by language models, and Propositions 59 to 80 came from the machine alone. I have not yet examined them thoroughly in their present form. By my own rule, the generator wrote part of the rule and that part has not passed the operator. Each proposition faces four questions:

1. Which of Sterling’s words does it restate?

2. Does it say the same as they do, or more, or less?

3. If it has no source in his words, does it follow from propositions that do?

4. If neither, is it kept as a labeled addition of mine, or struck?

Truth, Understanding, Wisdom

I asked whether, if Sterling’s words are true, his understanding of them is wisdom. It is not, by that alone. Truth belongs to the words. Understanding belongs to the mind that sees why they are true. Wisdom belongs to the assents: by 5, virtue consists of rational acts of will. A man can understand that externals are neither good nor evil and assent an hour later to the impression that an insult has harmed him.

Sterling states how high the bar is (section 10):

To go beyond making appropriate choices and achieve virtue, I must make appropriate choices and those choices must be connected together in a settled disposition to rationally evaluate all information that comes to me. Hence, one cannot perform one virtuous action--virtuous actions come when one has reached the stage where one’s inner rational development has been perfected. No-one achieves that except the Sage. I, personally, am willing to be a bit more generous and call some actions “virtuous”, but most of the ancient Stoics would not.

The corpus follows the strict statement and records the concession (Virtue Is Correct Assent — The Argument v1.1, section 6). On either reading the leniency reaches acts and never wisdom. The Canon can hold truth and the operator can hold understanding. Wisdom cannot be stored anywhere in the system, because it exists only in the operator’s assents.

The System at Work Once

I remembered that passage loosely, as if Sterling had allowed that a few men could act virtuously. The machine searched the archive and found a note of mine from 2018, then section 10, then the original: Sterling’s post to the International Stoic Forum of 30 March 2015.

Laid beside the original, section 10 showed one difference. Sterling wrote, “my ‘action’ is my choice, not anything I physically do.” My copy had moved the emphasis to “my.” No word differed.

That was the third test, fidelity. The machine found the source and set the two texts side by side. I ruled that the copy be corrected according to Sterling’s original. The Project Knowledge copy was corrected and the correction logged (Regression and Correction List, item 14). The restored emphasis is also the doctrine’s: the sentence puts the action in the choice and not in the deed (Ruling Register, R-2026-09-30-03).

The machine could show that the copy differed. Whether to restore the original was mine to decide.

No Compromise

I do not call the system an intelligence, and I will not build toward one that needs no operator. All that bears on happiness is assent (by 1 to 3), and assent is mine (by 8). Nothing can assent for me. A system built as if it could would only hide from me the one act that is mine.

— Dave Kelly

Sources

Declared Gaps

1. Claims 8 and 9 are my derivations. Sterling has not applied his premises to machines.

2. That the machine is process only rests on the first commitment and is not argued here. Whether any understanding stands behind its ability is not settled by this essay.

3. The paper I saw mentioned is not identified. The hippocampus reading is a guess.

4. Magnifica Humanitas, the Pope’s encyclical on AI, has not been read. The Pope’s words are taken from news reports.

5. Fodor and HippoRAG are cited from general knowledge and were not opened. Ross is reported from the CPA run, and The Right and the Good was not opened.

6. Sterling’s statement that reasonable educated men will agree about moral facts was not located in those words. The record gives the consulting of others’ intuitions, in paraphrase.

7. “How Much Is Built” and the remark on why HippoRAG works are observation without source.

8. The Stoic News copy of the Nine Excerpts still carries the uncorrected emphasis and lacks the date.

9. No Derivation Trace has been run on this essay.

Version Note

v0.1, 10 October 2026. First version, composed from one working session.

Theoretical foundations: Grant C. Sterling (Eastern Illinois University). Analysis and synthesis: Dave Kelly. Prose rendering: Claude (Anthropic). 2026.

posted by Dave | 3:53 AM

0 Comments:

Post a Comment

<< Home