The Illusion of Compute Power: Ontological Missteps of Contemporary AGI Paradigms
Year: 2026
Type: preprint
Abstract: Global AGI research has consolidated a self-authorizing consensus: compute power, data, chips, energy and mathematical formal deduction are regarded as core increments toward artificial general intelligence. Yao Qizhi’s AWM architecture, Wang Xingxing’s compute-resource-constraint thesis, and Ma Xiao’s AI-driven automated mathematical-proof route jointly constitute the core pathways of this convergent paradigm. Drawing upon the Substance-Function-Telos meta-axiom set and truncation-presumptuousness criteria, this paper delivers the following verdict: the convergent-static ontology is merely one valid instance ontology rather than a meta-origin. The contemporary AGI research track is rooted in the usurpation of this instance ontology: convergent presuppositions are treated as the sole benchmark of reality, and computational tools are stacked exclusively at the Function dimension, while the ontological substrate (Substance) and civilizational-evolution directional tendency (Telos) are absent. Compute power, data and hardware are all downstream derivatives generated within the paradigm, not first-order premises for AGI breakthroughs. Using the three-question metric derived from Substance-Function-Telos, this paper anatomizes the inherent boundaries of the three above-mentioned pathways and identifies their shared root pathology of reductionist truncation: these approaches can only harvest fruits within the predefined domain of the convergent-ontology basin, yet can never autonomously switch ontological presuppositions and reach the ontological origin of carbon-silicon-coupled AGI. The bottleneck of AGI has never lain in insufficient compute supply, but in ontological-level mismatch: confusing an instance ontology for a meta-origin. Without dismantling this usurpation, greater compute-scale expansion will only deepen involution and move further away from genuine general intelligence. This paper can be summarized in one sentence: the present-day research track is not answering the AGI question; it is evading meta-cognitive cross-ontology questions by deploying ever-larger compute power within the basin of a single instance ontology.
Author: Li, dongnan
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225344741