Saved in:
Bibliographic Details
Main Author: XU, Judit
Format: Recurso digital
Language:English
Published: Zenodo 2025
Online Access:https://doi.org/10.5281/zenodo.18042124
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866901222379749376
author XU, Judit
author_facet XU, Judit
contents <p><span lang="EN-US">This work presents a theoretical and architectural analysis of Artificial General Intelligence (AGI). Rather than proposing an implementable system, training method, or optimization strategy, the paper advances a set of <strong>architectural impossibility claims </strong>concerning the conditions under which AGI, as defined in this work, cannot arise.</span></p> <p><span lang="EN-US">AGI is defined here not in terms of task generality, benchmark performance, or autonomous goal pursuit, but as <strong>open-ended semantic generativity</strong>: the capacity of a system to generate novel conceptual structures beyond any fixed objective space, and to continually transform its own representational framework without terminal convergence. Under this definition, the paper argues that a broad and dominant class of contemporary AI architectures—those organized around global optimization toward fixed objectives, complete internal alignment, and convergence-oriented cognitive closure—are structurally incompatible with AGI.</span></p> <p><span lang="EN-US">The contribution of this work is explicitly <strong>negative and foundational</strong>. It does not claim empirical validation, performance relevance, or near-term applicability. Instead, it aims to <strong>constrain the conceptual and architectural design space </strong>of AGI research by making explicit the structural commitments under which general intelligence becomes impossible, regardless of model scale, data volume, or computational resources.</span></p> <p><span lang="EN-US">The paper is intended for readers interested in the conceptual foundations of artificial intelligence, including researchers in AI theory, cognitive science, philosophy of mind, complex systems, and AI safety. It is made publicly available to encourage careful conceptual discussion and critical engagement on the architectural limits of current AI paradigms.</span></p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18042124
institution Zenodo
language eng
publishDate 2025
publisher Zenodo
record_format zenodo
spellingShingle Architectural Thresholds for Artificial General Intelligence: Why Optimization, Alignment, and Closure Systematically Preclude General Intelligence
XU, Judit
<p><span lang="EN-US">This work presents a theoretical and architectural analysis of Artificial General Intelligence (AGI). Rather than proposing an implementable system, training method, or optimization strategy, the paper advances a set of <strong>architectural impossibility claims </strong>concerning the conditions under which AGI, as defined in this work, cannot arise.</span></p> <p><span lang="EN-US">AGI is defined here not in terms of task generality, benchmark performance, or autonomous goal pursuit, but as <strong>open-ended semantic generativity</strong>: the capacity of a system to generate novel conceptual structures beyond any fixed objective space, and to continually transform its own representational framework without terminal convergence. Under this definition, the paper argues that a broad and dominant class of contemporary AI architectures—those organized around global optimization toward fixed objectives, complete internal alignment, and convergence-oriented cognitive closure—are structurally incompatible with AGI.</span></p> <p><span lang="EN-US">The contribution of this work is explicitly <strong>negative and foundational</strong>. It does not claim empirical validation, performance relevance, or near-term applicability. Instead, it aims to <strong>constrain the conceptual and architectural design space </strong>of AGI research by making explicit the structural commitments under which general intelligence becomes impossible, regardless of model scale, data volume, or computational resources.</span></p> <p><span lang="EN-US">The paper is intended for readers interested in the conceptual foundations of artificial intelligence, including researchers in AI theory, cognitive science, philosophy of mind, complex systems, and AI safety. It is made publicly available to encourage careful conceptual discussion and critical engagement on the architectural limits of current AI paradigms.</span></p>
title Architectural Thresholds for Artificial General Intelligence: Why Optimization, Alignment, and Closure Systematically Preclude General Intelligence
url https://doi.org/10.5281/zenodo.18042124