Guardat en:
| Autor principal: | |
|---|---|
| Format: | Recurso digital |
| Idioma: | anglès |
| Publicat: |
Zenodo
2026
|
| Matèries: | |
| Accés en línia: | https://doi.org/10.5281/zenodo.19264199 |
| Etiquetes: |
Afegir etiqueta
Sense etiquetes, Sigues el primer a etiquetar aquest registre!
|
Taula de continguts:
- <p>This paper argues that Artificial General Intelligence (AGI) and its theoretical successor, Artificial<br>Superintelligence (ASI), are fundamentally unachievable through probabilistic computation alone,<br>regardless of model scale, architectural innovation, or computational investment. We establish that<br>all current AI systems—including large language models, diffusion models, and reinforcement learning<br>agents—operate through statistical pattern matching over structured data representations. While<br>these systems produce outputs that superficially resemble cognitive behaviour, they lack the defining<br>properties of cognition: embodied experience, temporal continuity, homeostatic self-regulation,<br>and phenomenal consciousness. We argue that these properties are not emergent features of sufficient<br>computational complexity but are intrinsic to biological neural substrates operating through<br>electrochemical processes that cannot be replicated through digital simulation. The paper presents<br>two logically exhaustive paths to AGI: (1) direct bidirectional integration between artificial systems<br>and biological neural tissue, or (2) complete replication of human neural architecture at biological fidelity.<br>Both paths require breakthroughs in neuroscience, bioengineering, and materials science—not<br>in software or computational scaling. We conclude that the prevailing industry narrative of achieving<br>AGI through larger models and more compute represents a category error of historic proportions, and<br>that genuine progress toward AGI requires redirecting research investment toward neurotechnology<br>and biological-artificial integration.</p>