The Eightfold Principles of Knowing-Equilibrium: Authority Bounded Inward, Sight Extended Outward — A Synthesis of the H-DKPS Series on Physician-Governed Clinical AI
Year: 2026
Type: preprint
Abstract: The grand synthesis of Volumes I and II of the H-DKPS series on physician-governed clinical AI (知衡八法). It reports no new experiment. The question. Volume I of this series asked what must be true of a clinical system in which a probabilistic component proposes and a deterministic component decides, if it is to be safe and useful to the patient in front of the clinician. It answered at a single decision. Volume II asked what must be true of everything around that decision — the reasons behind it, the measurements under it, the rules over it and the people beyond it — if extending what the system can see is not to extend what it may do. This synthesis reads the two volumes as one argument. Three layers. The series serves a clinical philosophy, negentropic rebalancing medicine, which treats illness as growing disorder across interlocked physiological systems and treatment as the restoration of order and balance across all of them rather than the correction of one number. It does so through an architecture, the Human-Centric Deterministic-Kernel Probabilistic-Shell (H-DKPS), in which a deterministic kernel is the only path to an output, and through one deployed implementation. The philosophy is not tested here. Within the system, "negentropy" names one advisory projection, labelled as mechanistic and not as demonstrated efficacy, and one envisaged population assessment that has deliberately not been built. The argument. Volume I bounded authority inward. Its four principles — 知止, 知行, 知變, 知己 — correspond to four kinds of property: safety over a decision, bounded liveness, safety over a trajectory, and validation at the human interface. Each part began where the previous kind of property ran out. Volume II extended sight outward, in four directions. 知然 checks the justification of each output against the physician's causal graph. 知物 checks the inputs: what may enter, what may not be manufactured, and what must travel with every value. 知度 checks the system's account of itself against requirements written by others. 知眾 governs what may be said about groups. None of these adds a ground on which the system may refuse a physician's order. The one outer mechanism that constrains the physician is the production form of the bounded override that Volume I had already argued for. What runs through all eight. Authority is placed by position. Every bound is reported with where the risk went next. Every silence is a finding — and in Volume II the same suspicion was turned on the checks themselves: a first causal rule that could never fail for the most connected agents, a mutation that passed by accident, a gate that changed what a daily calibration meant. Every mechanism is reported in its actual state — written, wired, running or correct — because these are four different claims. Scope. No part of the series reports a clinical outcome. Volume II used no record of patient care. Not peer reviewed.
Author: Chiu, Lu-An
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7222821035
Cites:
Cited by:
Related to: