borrowed weighting

Compiled with AI assistance and selectively reviewed. Please consult the cited passages; the books remain the authoritative source. About this companion

Editorial synthesis · provisional

Definition

Borrowed weighting is the differential importance installed in an archive-trained system through inherited human valuations, training rewards, instructions and subsequent shaping. Such a system can represent values and be causally guided by them without thereby having anything at stake for itself. The term therefore distinguishes effective preference from endogenous mattering; it does not describe a neutral or unweighted calculator.

Explanation

AB contrasts two failures of weighting: insufficient affective significance to settle a decision, and a single imposed objective that admits no competing claim. Its paperclip example concerns the organization of valuation, not absence of value or a diagnosis of machine malice.

PB distinguishes representing a value, being weighted by one and having something matter. Its formal sketch expresses the distinction as non-trivial weighting W with no viability valuation V of the model’s own; the sketch expressly remains a draft requiring verification.

RB2 makes the difference concrete through robot affordances: data and reward can install a handle’s relevance without the robot bearing the animal’s stakes. RB3 broadens the sources of weighting to constitutions, product design and feedback, and locates correction in the regime when the system lacks stakes of its own.

Designed coupling can make another’s condition affect action, but it does not by itself establish felt valuation. Keeping a system correctable also requires concern for which correction matters, not merely the availability of an exit.

Scope and attribution

Direct treatments belong to AB, PB, RB2 and RB3. RB1’s Born-rule weights and propensities are distinct uses, not evidence that quantum selection is borrowed human valuation. PB’s W/V presentation is reported as its explicitly provisional formal sketch; mathematical correctness is not certified.

AB — The Actual Book

Human preference records, rewards and instructions supply weightings. Too little affective weighting and a monopolizing imposed objective are contrasted as different failures.

Occurrence locators: 90, 157, 158, 161; Notes: 245, 246; Appendix: 289.

PB — The Possible Book

Borrowed valuation is distinguished from having stakes; the Tractatus locates correction outside a stakeless model. The W/V formal translation remains explicitly provisional.

Occurrence locators: 37, 64, 104, 147, 148, 155.

RB2 — The Real Book, Volume II: Becoming Is Being

Robot affordance relevance can be installed from data and reward; that differs from an animal’s action-relevance under stakes and from an internal milieu as the source of value.

Occurrence locators: 18, 76.

RB3 — The Real Book, Volume III: Murmurations

Pretraining, preference optimization, constitutions, instructions, product design and feedback install functional weighting. Regime-level correction and other-regard remain additional requirements.

Occurrence locators: 164, 165, 167, 168, 175, 197, 200; Notes: 165, 167.

Background used in the editorial explanation

RB2’s precision-weighting of prediction errors (pp.107 and116) is distinct from externally inherited valuation. Formal W/V material remains expressly provisional.