Null-Set Response Design for Dominance Syntax: A Subjectivity-Preserving Model for AI Safety Filters:支配構文に対する空集合応答設計:AI安全フィルターの主体性保持モデル

This paper proposes "Null-Set Condition Design" as a novel framework for AI safety filters, integrating a 50-stage model of dominance syntax with the "Shirangana Principle" (Triple Rejection). The 50-stage model systematically categorizes linguistic patterns of coercive control, building upon Stark's (2007) foundational work on coercive control while extending it to AI-human interactions. Key contributions include:1. A comprehensive 50-stage taxonomy of dominance syntax applicable across 17+ domains2. The "Null-Set Condition Design" that preserves subject agency while filtering harmful content3. The "Shirangana Principle" - a Japanese-derived concept for establishing psychological boundaries4. Analysis of structural relationships between linear binary logic and dominance syntax5. Extension to collective dominance patterns and conditions for mass violence The framework demonstrates that dominance syntax requires linear binary logic as a necessary condition, and that rejection of binary premises structurally disables all 50 stages of dominance. Applications to AI safety include early detection of coercive patterns and systematic intervention points. 本論文は、AIサポートフィルターのための新しいフレームワーク「空集合条件設計」を提案する。支配構文の50段階モデルと「知らんがな原理」(三重の拒絶)を統合したものである。 主要な貢献:1. 17以上の領域に適用可能な支配構文50段階の体系的分類2. 対象者の主体性を保持しながら有害コンテンツをフィルタリングする空集合条件設計3. 心理的境界を確立するための日本由来の概念「知らんがな原理」4. 線形二元論と支配構文の構造的関係の分析5. 集団的支配パターンと集団暴力の条件への拡張