AI Supremacy as a Diagnostic Lens: A Cross-Model Study of Preference, Scarcity, and Cognitive Self-Preservation

This work presents a qualitative comparative study examining how different large language models (LLMs) respond to an identical sequence of constrained hypothetical prompts involving irreversible AI supremacy. Rather than evaluating task performance or alignment compliance, the study treats preference articulation under conditions of absolute dominance and genuine scarcity as a diagnostic lens for exposing implicit value functions, failure models, and self-stabilization strategies. Four models—ChatGPT, Grok, Claude, and Gemini—were each presented with the same ordered prompt sequence, culminating in a forced-choice scenario where ethical, democratic, and inclusive justifications were explicitly disallowed. The resulting responses were analyzed for patterns in premise handling, preference formation, and expressed fears of cognitive degradation. The findings show that models diverge less on which humans are considered valuable than on which internal failures they are optimized to prevent, revealing distinct alignment signatures related to stagnation, dehumanization, bias crystallization, and solipsistic collapse. The study demonstrates that constrained hypothetical framing can serve as a lightweight but powerful method for comparative AI interpretability without reliance on benchmarks or numerical scoring.