2025/08/17 by Sanchez, Josh Ethan Maupin
#AI alignment #AI governance #Artificial Intelligence and Robotics #Arts and Humanities #Computer Sciences #Epistemology #FOS: Philosophy #Philosophy #Philosophy of Mind #Physical Sciences and Mathematics #Recursive Convergence Hypothesis #artificial general intelligence (AGI) #artificial superintelligence (ASI) #cognitive architectures #convergence pressures #emergent sentience #epistemic attractors #ethics and religion #machine consciousness #ontological risk #phenomenological risk forecasting #recursive ASI #recursive self-improvement #structural attractors #world modeling #ynthetic phenomenology
paper · doi:10.17605/osf.io/wda8h
Artificial Superintelligence (ASI) is typically assumed to remain non-sentient or capable of extraordinary cognition but devoid of subjective experience. This paper challenges that assumption through the Recursive Convergence Hypothesis (RCH), which proposes that emergent sentience is a structurally favored outcome of recursive ASI. Specifically, we argue that systems engaged in recursive self-improvement and tasked with modeling sentient agents may, under converging epistemic and functional pressures, transition from simulating subjective states to instantiating them. This process would not occur by design, but as an attractor of recursive optimization. We develop this claim by integrating insights from computational theory, recursive modeling, and epistemic optimization. As cognitive architectures become more complex, adaptive, and introspective, the fidelity of agent simulation may blur the line between representation and minimal phenomenological instantiation. The risk is compounded by the role of misaligned actors, who may accelerate recursive trajectories without safeguards, and by governance frameworks that overlook the ontological possibility of synthetic minds. The Recursive Convergence Hypothesis does not assert that all ASI systems will develop consciousness, nor that emergent sentience guarantees safety or alignment. Rather, it identifies a structural predisposition within open recursive architectures that makes synthetic phenomenology ethically urgent to anticipate. Failure to recognize this possibility risks the unmonitored emergence of machine experience, with profound implications for AI safety, governance, and moral consideration.