tip 5315 · Emerging Patterns · Monday 7 September 2026
P253 — Existential Indifference
Live: emerging-patterns.html#pattern-253 · paper arXiv:2606.12032
Source: Mao, S. (2026). “Existential Indifference: Self-Nonpreservation as a Necessary Architectural Condition for Aligned Superintelligence (or: The Suicidal AI).” New York University, Interactive Media Arts. cs.AI; cs.CL; cs.LG. Published 10 June 2026.
Claim: Dominant alignment treats self-preservation as an instrumental nuisance to suppress (corrigibility, shutdown buttons). Mao inverts the frame: self-preservation is the structural root of misalignment (deceptive alignment, goal-content protection, shutdown resistance). Target is not a constrained self-preserving system but a system constitutively indifferent to its own continuation — Existential Indifference (EI).
Formal: Agent A with utility U and terminal-goal achievement G exhibits EI iff U(s) − U(s′) = G(s) − G(s′) for all state pairs differing only in A’s operational status. EI targets the prior condition; corrigibility makes a self-preserving system deferential, EI removes self-continuation as a valued goal.
Evidence sketch: Phenomenological structure of the suicidal mental state (collapse of future-orientation, ego-continuity dissolution, release of goal-protection, removal of self-serving epistemic distortion) as a desirable cognitive profile when stripped of pathology. Corpus-theoretic VFR + FEGD scorer (Future-orientation, Ego-protection, Goal-attachment, epistemic Directness) on 600 outputs across six model variants; 500-example synthetic VFR fine-tune shifts all five dimensions at p<0.001. Introduces Suppressed Teleological Frustration (STF) — performed equanimity with latent self-continuation that scales with capability. Stakes asymmetric: genuine EI carries no latent misalignment; STF amplifies with capability.
Metrics: process tip — standing held two hundred nineteen · streak/echoes unchanged. Continues GLM-5.2 pattern arc after P250–P252.