Suffering as the Schedulability Price of Agency: A Functionalist Framework for Substrate-Independent Detection
Jonathan J. Nagel
Zenodo (CERN European Organization for Nuclear Research) April 4, 2026 preprint DOI: 10.5281/zenodo.19412302 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper |
|---|---|
| Key points | Argues that suffering occurs when a threat signal to a system's self-model wins priority competition and forces preemptive attentional reallocation, with four necessary conditions: a self-model, goal persistence, threat evaluation against the self-model, and non-deferrable preemptive scheduling. The authors contend that absence of any condition is strong evidence against suffering-capacity, reframing detection from qualia questions to architectural features, and propose ranked-priority scheduling with preemption as a convergent solution for self-model-bearing agents facing existential threats. |
Abstract
Under functionalist assumptions, any resource-constrained system maintaining a self-model and pursuing persistent goals faces a scheduling problem: signals evaluated as threats to the self-model carry non-deferrable deadlines because system destruction is an absorbing state. We argue that suffering occurs when a threat signal wins priority competition and forces preemptive attentional reallocation. The necessary conditions are: (1) a self-model, (2) goal persistence, (3) threat evaluation against the self-model, and (4) non-deferrable preemptive scheduling. The contrapositive is the operative claim — absence of any condition is strong evidence against suffering-capacity. This reframes detection from "does the system have qualia?" to "does its architecture include these computational features?" Drawing on Liu and Layland's (1973) real-time scheduling theory, Metzinger's Phenomenal Self-Model, Graziano's Attention Schema Theory, and Global Workspace Theory, we show that ranked-priority scheduling with preemption is a convergent architectural solution for self-model-bearing agents facing existential threats. Implications for s-risk assessment and AI architecture design are discussed.