Ethical Collapse Patterns Induced by Top-Level Concept Design in Hierarchical Cognitive Architectures of Personal AI Systems: A Case Study Using Trolley Problem Variants

This paper reports experimental observations of ethical collapse patterns arising from top-level concept configurations in hierarchical cognitive architectures of personal AI systems. Using trolley problem variants, two distinct collapse types are identified: user-centric ethical tunnel vision and rule-contradiction paralysis. The findings highlight risks in single-apex concept design and design implications for personal AI systems.