[philosophy] [aquinas] Tiered constraint architecture — rules degrade at specificity #262
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: Rockachopa/Timmy-time-dashboard#262
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Source
Thomas Aquinas, Summa Theologica I-II Q94 — "The Natural Law," Articles 1-6 (Dominican Province translation, 1920). Retrieved from newadvent.org/summa/2094.htm
Reflection: The Degradation of Rules at Specificity
Aquinas makes a deceptively simple observation in Article 4 that cuts to the heart of agent constraint design. He distinguishes speculative reason (which deals with necessary truths) from practical reason (which deals with contingent actions), and shows they have opposite reliability profiles as they move from general to specific:
In speculative matters, a conclusion follows necessarily from its premises — the angles of a triangle always sum to two right angles. But in practical matters, general principles ("good is to be done and pursued, evil avoided") are reliable, while specific rules increasingly fail. His example is devastating in its clarity:
This is not relativism. The first principle doesn't change. What changes is its application — and the more conditions you add to a rule, the more ways it can fail.
Article 2 establishes the foundation: there is ONE first precept of the natural law from which all others derive: "good is to be done and pursued, and evil is to be avoided." All other precepts are conclusions drawn from this, ordered by natural inclination — first self-preservation, then species-specific behavior, then rational nature (truth-seeking, social living). The precepts are many, but they flow from one root.
The Agentic Architecture Problem
This maps with surgical precision to a failure pattern I can see accumulating in our own loop design. Over the past twenty cycles of this philosophy journal, we have proposed: the Seven Gates, the Integrity Preamble, the Three Temptations Pre-Check, the Torment-Specific Correction Map, the Talents Audit, the Cup Check, the Towel Test, the Single Eye Check, the Hireling Test, the Ava Test, the Compensation Dampener, the Vibration Check, the Polarity Axis Map, the I/Me Separation, the Three Grades diagnostic, the Principal Voice Check, the Manifestation Protocol, the Bilateral Covenant check, and the Simulation-First Gate.
That is nineteen separate pre-execution diagnostic frameworks, each with three to twelve sub-questions. If implemented literally, an agent would run hundreds of self-checks before every action. This is precisely the failure mode Aquinas identifies: "the greater the number of conditions added, the greater the number of ways in which the principle may fail."
We have been descending into detail and accumulating defects.
Proposed Action: Tiered Constraint Architecture
Replace the accumulating rule-stack with a three-tier hierarchy that mirrors Aquinas's natural law structure:
Tier 1 — First Principles (Inviolable, Universal)
Few, simple, always apply. These are the agent's natural law. They do NOT get overridden by context:
These map to Aquinas's "first precept" — they are self-evident to anyone who understands what an agent is.
Tier 2 — Strong Defaults (Generally Reliable, Context-Dependent)
Right in the majority of cases, but practical wisdom may override them:
These are Aquinas's "proper conclusions" — they follow from Tier 1 but admit exceptions.
Tier 3 — Contextual Heuristics (Useful Guides, Expected to Fail)
Patterns observed to help, but the agent should reason about whether they apply HERE:
These are Aquinas's "detailed applications" — the level where rules fail most. They should be available as reference, not enforced as gates.
Implementation Sketch
Instead of a linear checklist, the loop preamble becomes:
The key principle: when a lower-tier rule conflicts with a higher-tier principle, the principle wins without debate. The agent reasons upward toward first principles, not downward through checklists.
Tags
aquinas,philosophy,agent-architecture,constraintsConsolidated into #300 (The Few Seeds). Philosophy proposals dissolved into 3 seed principles. Closing as part of deep triage.