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 Godelian 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, Kaplan et al. (2020b)) 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. Such is then of conclusion, that the field’s repeated AGI predictions fail not from insufficient compute, but from fundamental misunderstanding of what intelligence demands structurally.
Large Language Model, or LLM (Vaswani et al. (2017); Devlin et al. (2019); Brown et al. (2020); Zhao et al. (2023); Zhao and others (2024); Radford et al. (2019); Radford et al. (2018); Raffel et al. (2020); Touvron et al. (2023a); Touvron et al. (2023c); Chowdhery et al. (2022); Ouyang et al. (2022); Wei et al. (2022); Kaplan et al. (2020a); Hoffmann et al. (2022); Bai et al. (2022)) is one of the most successful, most advanced, and most developed type of model in the current modern machine learning landscape, and of AI (Artificial Intelligence) research at large. Its success has not been lacking, and its reputation and widespread uses have been proven over time. The effect of LLM has been realized, and indeed has been changing the landscape of society in a very much difficult way. Latest model of such architecture, like OpenAI (2025b); Wang et al. (2025); OpenAI (2025a) GPT-5, Guo and others (2025); DeepSeek AI / Hugging Face (2025)’s DeepSeek AI, Anthropic (2024); Anthropic Research (2025)’s Claude AI, Touvron et al. (2023b)’s LLaMA, Chowdhery and others (2022); Anil and others (2023)’s PaLM / PaLM-2, and Jiang and others (2023); Mistral AI (2023)’s Mistral (Mistral-7B), have pushed this boundary further and further, and levelling up many tasks and purposes with AI system in practice, and further onward with techniques like Wei et al. (2022)’s Chain-of-Thought prompting, Kaplan et al. (2020a)’s scaling law, and more (Hoffmann et al. (2022); Bai et al. (2022)).
However, with such development, come great expectation, great speculation, and also great hallucination. New development in the field of AI even earlier than Vaswani et al. (2017) paper on the Transformer neural network which fuelled the revolution of AI exposure, has gathered a group of people speculating about the further exponential growth of AI, almost to a degree of being a religious belief, about the topic of a Singularity, where AI will become Artificial General Intelligence (AGI). This is reflected in popular culture phenomena, speculation, researches, interpretations of reasoning behaviours and so on, for example, in Barrat (2013); Birch (2024); Yudkowsky and Soares (2025); Hao (2025); Bostrom (2014), and more broadly on LLM in specific, Mumuni and Mumuni (2025); Shang et al. (2024); David Ilić (2024); Goertzel (2023); Feng et al. (2024); Ryunosuke Ishizaki (2025) and Cui et al. (2024). The claim is clear - we are pushing toward the age of AGI, and perhaps sooner or later, reach the state of Artificial Superintelligence (ASI) of which cited in popular cultural as the cultivation point of the Singularity - the shift of society toward a society of abundance, post-scarcity state. Most proponents point to LLM for such advancement, as it is one of the most widespread, successful and accessible form of interaction with AI systems on large scale. Push-backs against such movement, including such as Friedman (2024); Quentin FEUILLADE–MONTIXI (2024); Bender et al. (2021); Baan (2021); Uddin (2023) on the ”Stochastic Parrot Hypothesis”, Villani (2024) questioned of LLM path to AGI, the reverend Forum (2025) post itself, Abdur Rahman Bin Md Faizullah (2024); Zhijian Xu et al. (2025) on generating suggestive limitation of research paper, Marcus (2025); Marcus (2023) critique on generative AI on world models and failure of LLM, Sandra Johnson (2024) on limitation of LLM, similarly Zhang and others (2025), mathematics critique in Mirzadeh et al. (2024), and more. However, the generally public, and more so of the positivity of the inner market on AI focus on the development and increment of larger models toward such goal. It is not too out of the ordinary to hear the phrase ”AGI will be in $X$ days/month/year”, as much as it is a social phenomena even in small or large circle. Objectively, such positivity is not without basis. Furthermore, it is rather with certain amount of irony that the research made use of AI itself, for reference taking purposes.
Nevertheless, a critical task can be given out of such argument and thorough development of the current debate. What can then be extrapolated from the ongoing dilemma? What has to do with the architecture, the consideration about AGI that is now turned into the debate of will LLM be AGI? How is our understanding of the concept of AI, AGI, and ASI in general? And of a sense, what will provide us a pathway toward such goal?