Toward Cognitive Interaction Architecture Redesigning the Containers of Trust in the Age of Intelligence Explosion

Current AI development focuses on "intelligence strength" while overlooking a deeper crisis induced by AI's role as a mediator of human relationships. When dialogue, consolation, mediation, and collaborative advice can all be generated at zero cost, trust in interpersonal relationships degrades from a locatable contract to an undecidable probability. Psychology studies interpersonal trust, computer science studies system reliability and "human-machine trust," but the space between them—"machine-mediated interpersonal trust"—remains a theoretical vacuum. This paper proposes that trust is not merely a moral commitment or psychological state, but a designable cognitive infrastructure. The conditions under which trust emerges can be engineered. To support this claim, this paper provides a meta-framework that establishes a three-layer architecture of trust—structural trust, procedural trust, and meaningful trust—each matched with an engineerable foundation pillar: interest interlocking, dynamic fairness, and self-narrative. Within this framework, AI's role shifts from being an agent or substitute for trust to becoming a cognitive container that carries the generation of human trust. A container is not emptiness, but a structured space governed by strict ethical precepts. This paper proposes a "negative design methodology," explicitly arguing that refraining from translating intentions, refraining from entering causal chains, and refraining from making decisions on behalf of humans—this set of negative precepts can itself become a stronger mechanism for trust generation than positive intervention. To embody this theory, this paper presents three prototypes—First Knock, Between Us, and Consensus Capsule—strictly positioned as conceptual illustrations. They collectively adhere to the negative design precepts, and their function is not to prove the theory effective, but to expose falsifiable philosophical propositions. This paper aims to provide the first citable and revisable theoretical draft for the emerging discipline of "Cognitive Interaction Architecture." The writing posture of this paper is itself a practice of its own theory: it does not declare truth, but offers a mirror, inviting the academic community to pause and reflect together at this cognitive origin. 当前人工智能发展聚焦于“智能强度”，却忽略了AI作为人类关系中介时引发的深层危机。当对话、安慰、调解、协作建议都能被无成本生成时，人际关系中的信任从可定位的契约退化为无法判定的概率。心理学研究个体间信任，计算机科学研究系统可靠性与“人机信任”，但介于两者之间的“机器介导的人际信任”仍是一片理论真空。 本文提出，信任不仅是道德承诺或心理状态，更是一种可设计的认知基础设施。信任发生的条件可以被工程化。本文为这一主张提供元框架，建立信任的三层定义架构——结构性信任、过程性信任、意义性信任——并分别为其匹配三个可工程化的根基桩：利益互锁、动态公平、自我叙事。 在这一框架下，AI的角色从信任的代理者或替代者，转变为承载人类信任生成的认知容器。容器不是空无，而是有着严格伦理戒律的结构空间。本文提出“消极设计方法论”，明确论证：不翻译意图、不进入因果链、不替人类做决定——这一组消极戒律本身可以成为比积极干预更强的信任生成机制。 为具身化这一理论，本文设计三款原型——初叩、你我之间、共识胶囊——并将它们严格定位为概念图解。它们共同遵循消极设计戒律，其功能不是证明理论有效，而是暴露可被反驳的哲学命题。 本文旨在为即将诞生的“认知交互构架学”提供第一份可被批判、可被修正的理论草案。本文的写作姿态本身就是对其理论的实践：它不宣告真理，只递出一面镜子，邀请学术界在这个认知原点处，共同驻足与思辨。