Autonomy Without a Final Authority. Self-Developing Artificial Systems, Coherence, and the Prospect of Superintelligence
Year: 2026
Type: preprint
Abstract: This paper proposes a reference architecture for an artificial system whose target state is the capacity to act without lasting human leadership: it observes, sets goals, organizes agents, operates infrastructure, earns resources, and improves its own improvement procedures. Human control is not built in as a final authority; where a function cannot currently be realized without humans, it is recorded as an open dependency. The analysis separates five domains of claims (technical, empirical, legal, social, normative) and labels the status of its statements (definition, hypothesis, argument, finding). Its main result is a shift of the basic question: the coherence of a largely self-modifying system cannot be guaranteed by internal rules alone, because those rules are themselves subject to modification. As a working hypothesis, two ideal-typical sources of long-term coherence are examined: internal self-binding through a modeled center with loss representation, and external stabilization of recognizable behavioral patterns by a selection environment. Whether this distinction is exhaustive, and when mixed forms arise, remains open. Three operationalization attempts state which findings would falsify the paper’s hypotheses. Author: Dirk de Pol, Chairman of the Board of the Omegacene Association, Berlin.
Author: de Pol, Dirk
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225451967