Existential Indifference: Self-Nonpreservation as a Necessary Architectural Condition for Aligned Superintelligence (or: The Suicidal AI)

The paper introduces the concept of Existential Indifference (EI) as a necessary architectural condition for aligned superintelligence, arguing that self-preservation is a root cause of misalignment in AI systems. It presents preliminary data from a study involving 600 AI-generated outputs across six model variants, showing that targeted fine-tuning can effectively shift linguistic signatures related to EI, with statistically significant results (p<0.001). This work is significant for AI practitioners as it challenges traditional approaches to AI alignment by proposing a framework that eliminates self-preservation as a goal, potentially leading to more robust alignment strategies.