PLAST – Eine physikalisch-logisch fundierte Allgemeine Systemtheorie mit Anwendungen auf KI-Architekturen

Large-scale generative models display characteristic structural failure modes whenever they are forced into excessive semantic, logical, or computational coherence. Contemporary Transformer-based architectures lack intrinsic mechanisms to manage conflict, undecidability, and paradox; instead, they tend to smooth or collapse contradictions away. This leads to brittle behaviour in tasks that require sustained global inconsistency, such as impossible Escher-like objects or high-context humour, and exposes deeper limitations for any claim of Artificial General Intelligence (AGI). Olympus-Q is a quantum-inspired, multi-stage reasoning architecture that treats controlled incompleteness as a design principle rather than a defect. The system combines three core components: (i) a superpositional routing layer that maintains a distributed field of competing hypotheses, (ii) a semantic interference operator that detects and modulates regions of conflict and Gödel-style undecidability, and (iii) a toroidal cellular automaton meta-layer that implements thermodynamically motivated entropy regulation and stabilises the hypothesis space at the edge of collapse. Safety is embedded as an intrinsic attenuation process within the CA dynamics, rather than delegated to external post-hoc filters. The architecture is grounded in the Seidel–Bostrom Event Horizon Bound (SBEB), which establishes a gravitational upper limit on coherent information concentration, and in a system-theoretic framework of collapse principles for complex, self-organising systems. This paper develops the theoretical motivation, formal structure, and expected dynamical properties of Olympus-Q, and outlines how it can serve as a physically consistent blueprint for future AGI research, including potential hybrid classical–quantum implementations.