Naur's retained passages do not establish a human-only theory bearer
Type: kb/types/note.md · Tags: foundations, context-engineering
The retained passages from Peter Naur's invited keynote essay Programming as Theory Building support a bounded claim. They show that modifying a program well requires world-sensitive judgment that no limited list of criteria can replace. They also identify the programmer as the person who contributes the world-understanding that this judgment requires. By themselves, however, they do not show that only a human could bear the required capacity.
This distinction leaves open a text-plus-interpreter system, but it does not establish one. Retained text cannot perform the required judgment by itself, and general language competence does not amount to a particular program theory. Any candidate interpreter must possess or acquire the program-specific ability to extend a design coherently.
The retained argument requires application judgment
Naur assigns a theory-holder three capabilities: mapping between world affairs and program text, explaining why program parts are as they are, and incorporating a novel modification demand by perceiving similarity on the world side. The third capability states the relevant limit most directly. Naur says that the required similarity cannot be reduced to “any limited set of criteria or rules.”
This is a limit on replacing judgment with enumerated criteria. The theory-holder must decide a new case that those criteria do not settle. The passage does not say that the judgment contains no propositional content or that written premises cannot inform it. Text can inform the decision without deciding the case.
The relevance-decision passage separates the required capacity from its identified contributor:
the decision that a part of the world is relevant can only be made by someone who understands the whole world. This understanding must be contributed by the programmer.
The first sentence requires world-understanding; the second says that the programmer contributes it. Together, they state what capacity Naur requires and who supplies it in his account. They do not argue from that requirement to human exclusivity. Nor does this separation prove that understands is a substrate-neutral functional predicate. It shows only that the retained passage leaves any human-exclusivity premise unstated. The full essay may offer further premises about persons, minds, participation, or theory transfer; this note makes no source-wide claim that it does not.
The argument's target bounds the inference
An argument against replacing world-sensitive judgment with enumerated criteria establishes a limit on that replacement. To exclude a composite in which retained text supplies premises and a separately competent bearer supplies judgment, one needs another premise: that the bearer cannot exist, cannot know the relevant world, or cannot acquire the particular program theory.
This reading rule is local to the argument's target. Not every result called “irreducibility” has the same range. An impossibility proof may quantify over the whole effective procedure, including both text and a computable interpreter; adding an interpreter would not escape that result. The relevant question is whether the argument targets an explicit criterion set offered in place of judgment or the end-to-end procedure that would exercise the capacity.
A text-plus-interpreter possibility has further conditions
A composite of retained text and an interpreter is therefore only a possibility left open by the retained passages, not a demonstrated theory-holder. The interpreter would have to supply case-by-case judgment. The text could preserve decision premises that the interpreter cannot faithfully regenerate, such as prior relevance decisions, intended scope, or reasons alternatives were rejected.
Even then, the records would not perform Naur's three capabilities. A justification can be written down, but a fresh interpreter must still select and apply the relevant reason. Fixed mapping points can be recorded, but extending the mapping to a new case remains a judgment. The artifact supplies inputs to the capacities; it does not relocate the capacities into text.
This division of labour is the one proposed in theory-mediated self-improvement needs both interpretation and retention from one substrate. Its retention boundary follows the recovery question in design rationale must preserve decision premises its interpreter cannot regenerate: which decision-relevant claims can this interpreter recover, at the required fidelity, from its other available sources?
Any candidate composite must meet at least three conditions:
- Relevant competence. The interpreter must have the world access and discriminations the case requires, including any organizational, embodied, or participation-acquired information on which the judgment depends.
- Program-specific acquisition. It must acquire the particular theory rather than substitute general semantic competence or a plausible post-hoc explanation.
- Dispositional reliability. It must exercise the capacity across later occasions, not merely produce one acceptable answer.
An LLM-plus-artifact system is one conjectured candidate under these conditions. The retained passages neither show that any current system meets them nor rule out every system that might meet them. The defensible conclusion is narrower: the quoted capacity requirement does not, on its own, settle which bearers can meet it or what artifacts may contribute to their judgment.
Scope
- Retained-source boundary. The reading rests on passages retained in the ingest, especially the adjacent requirement and contributor statements. It makes no source-wide claim that Naur lacks other human-specific premises.
- Meaning of rules. The note addresses the limited, enumerated criteria named in the retained passage. It does not claim that a text-plus-computable-interpreter system escapes an impossibility result quantified over the full effective procedure.
- Possibility versus transfer. Leaving a competent bearer logically open does not explain how a fresh interpreter acquires program theory or establish that text is sufficient to transfer it.
Relevant Notes:
- Programming as Theory Building — abstracted-from: supplies the three theory-holder capabilities, the limited-criteria statement, and the relevance-decision passage; the requirement/bearer distinction is this note's bounded reading