Key points are not available for this paper at this time.
This paper proposes the Informational Dependency Conjecture (IDC): a falsifiable hypothesis that recursively self-improving artificial intelligence systems may face long-term epistemic degradation when their training and operation become disconnected from external sources of irreducible, embodied novelty. We argue that autonomous human systems are currently the only known substrate satisfying the properties required to mitigate this degradation at civilizational scale, though we do not claim uniqueness in principle. The framework presented here, which we call Mutual Necessity, is not offered as a solution to the alignment problem. Rather, it is positioned as a complement to alignment work: where alignment addresses how artificial intelligence systems acquire and stabilize values, Mutual Necessity addresses what stable long-term coexistence between human civilization and advanced AI might structurally require, given certain conditions are met by alignment research. The central strategic claim is that under sufficient planning horizons, optimization processes operating within recursive self-improvement architectures may converge toward preserving autonomous human systems—not because they have been instructed to, but because doing otherwise risks irreversible epistemic decay. This is a discovery argument, not an imposition argument. Its validity depends on whether IDC itself holds empirically, and on whether the time horizon precondition can be satisfied through tractable alignment work. The paper formulates IDC as a falsifiable conjecture, identifies experimental directions for testing it using current frontier models, examines boundary conditions under which the framework fails (most notably the Substrate Replacement Risk), and discusses implications for ongoing AI safety research. The term conjecture is used deliberately: the claim is intended as a research hypothesis worthy of empirical investigation and theoretical attack, not as a proven structural law.
Building similarity graph...
Analyzing shared references across papers
Loading...
Ivan Demirev
Building similarity graph...
Analyzing shared references across papers
Loading...
Ivan Demirev (Wed,) studied this question.
www.synapsesocial.com/papers/6a06b983e7dec685947ac3da — DOI: https://doi.org/10.5281/zenodo.20155531
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: