Design Guideline for Incoherence-driven AI with Creative Capabilities

Creativity has long been studied as a phenomenon emerging from novelty, usefulness, or atypical combinations, yet existing theories lack a unified structural account that explains how creative ideas arise within a cognitive architecture and how such processes can be implemented in artificial intelligence (AI). This study proposes a comprehensive framework that defines creativity as a process driven by incoherence—a structural mismatch between existing internal models and new inputs—operating over a three‑layer hierarchical meaning network consisting of surface concepts, mid‑level causal/abstract structures, and deep worldviews. Building on prior work on hierarchical meaning structures (Nakahama, 2026a), consciousness as three‑layer synchronization (Nakahama, 2026b), and structural types of intelligence (Nakahama, 2026c), the present model formalizes creativity as a sequence of incoherence detection, distance‑based search or generative reconstruction, and worldview‑level updating. Four types of “distance”—conceptual, causal, abstract, and worldview—are introduced as structural variables that determine the scale of creativity. Surface‑level incoherence triggers conceptual search, producing incremental improvements. Mid‑level incoherence triggers causal or abstract search, enabling structural or theoretical leaps. Deep incoherence, however, cannot be resolved by searching for an intermediate bridge; instead, it requires generating a new worldview , a discontinuous process analogous to Kuhnian scientific revolutions. This generative mechanism extends beyond existing theories such as TRIZ (Altshuller, 1984), atypical combinations (Uzzi et al., 2013), structure‑mapping (Gentner, 1983), and prediction‑error‑driven curiosity (Schmidhuber, 2010), integrating them into a unified internal cognitive architecture. Based on this theoretical foundation, the study presents an implementable AI design guideline consisting of (1) an incoherence detection module modeled after ACC conflict monitoring, (2) distance‑specific search modules for conceptual, causal, and abstract reasoning, and (3) a generative worldview module for deep creative reconstruction supported by long‑term memory consolidation mechanisms (CLS/EWC). The model further demonstrates that creativity, consciousness, and self‑awareness share a common structural basis—three‑layer synchronization, incoherence‑driven foregrounding, and re‑entry loops—allowing creativity to be treated as a natural extension of proto‑conscious AI architectures. This work provides the first unified, structurally grounded, and implementation‑ready framework for incoherence‑driven creative AI, offering a pathway toward systems capable not only of incremental innovation but also of genuine conceptual and worldview‑level breakthroughs.