Foundations of Artificial Intelligence Frameworks: Notion and Limits of AGI

TL;DR

The paper argues that artificial general intelligence cannot emerge from scaling current neural network paradigms and that this trajectory is neither healthy nor sufficient for progress. It critiques foundational assumptions, including misapplications of the Universal Approximation Theorem, and highlights the distinction between existential substrates and architectural organization. Through discussions of the Chinese Room, Gödelian arguments, and a fragmented landscape of learning theories, it contends that neural architectures alone lack the dynamic restructuring and deep interpretability required for genuine intelligence. To address these gaps, the author proposes a neural-architecture formalism that separates substrate from architecture and outlines an agentic AI framework, aiming to ground progress in principled, theory-driven constructs rather than hype around scalability and language-model performance.

Abstract

Within the limited scope of this paper, we argue that artificial general intelligence cannot emerge from current neural network paradigms regardless of scale, nor is such an approach healthy for the field at present. Drawing on various notions, discussions, present-day developments and observations, current debates and critiques, experiments, and so on in between philosophy, including the Chinese Room Argument and Gödelian argument, neuroscientific ideas, computer science, the theoretical consideration of artificial intelligence, and learning theory, we address conceptually that neural networks are architecturally insufficient for genuine understanding. They operate as static function approximators of a limited encoding framework - a 'sophisticated sponge' exhibiting complex behaviours without structural richness that constitute intelligence. We critique the theoretical foundations the field relies on and created of recent times; for example, an interesting heuristic as neural scaling law (as an example, arXiv:2001.08361 ) made prominent in a wrong way of interpretation, The Universal Approximation Theorem addresses the wrong level of abstraction and, in parts, partially, the question of current architectures lacking dynamic restructuring capabilities. We propose a framework distinguishing existential facilities (computational substrate) from architectural organization (interpretive structures), and outline principles for what genuine machine intelligence would require, and furthermore, a conceptual method of structuralizing the richer framework on which the principle of neural network system takes hold.

Figures (3)

• Figure 1: A typical example of a program and its logical process in PROLOG. The example requires determination of a parent-child relationship, with encoded knowledge graph, and traverse of the predicate logic. (a). Canonical example of a PROLOG predicate decision network, in an example of Chapter 2, section 6 in clocksin2003programming; (b) The program responsible for the respective sequential predicate logic, of which answers for the question accordingly.
• Figure 2: The conceptual framework of a learning agent (stanford2018ai). The agent itself in such framework then indeed, internally 'understand' the metric notion of mistake, knowing and then fixing, typical process of which a learning system can be considered.
• Figure 3: A loose illustration of the layering principle. The lower layers do not know what is on the upper part. However, they can receive potential downstream input features, and hence work accordingly. This is the basis of abstraction by layers, thus stating that the man in the box do not know, and do not understand, anything but his circumstances and his current capability. Nevertheless, it still fulfils its role in the lower level hence forth, and its operation only in such range of the layer.

Theorems & Definitions (15)

• Definition 1.1: Artificial
• Definition 1.2: Intelligence, top-down approach
• Conjecture 1.1: Artificial intelligence
• Conjecture 1.2
• Definition 2.1: Artificial general intelligence
• Definition 2.2: Standard multilayer network, zhang2023divedeeplearning
• Theorem 2.1: Universal approximation theorem --- UAT
• Theorem 2.2: Universal approximation theorem, simplified
• Definition 3.1: Minimization set
• Definition 3.2: Class $\mathcal{N}_{0}$ on $\mathbb{R}$