Gardado en:
| Autor Principal: | |
|---|---|
| Formato: | Recurso digital |
| Idioma: | |
| Publicado: |
Zenodo
2026
|
| Subjects: | |
| Acceso en liña: | https://doi.org/10.5281/zenodo.20247604 |
| Tags: |
Engadir etiqueta
Sen Etiquetas, Sexa o primeiro en etiquetar este rexistro!
|
Table of Contents:
- <p>Artificial intelligence is moving from isolated answer generation toward situated action. A modern AI<br>system may respond in language, call tools, inspect files, write code, execute commands, search memory,<br>ask for approval, launch tests, produce artifacts, and preserve evidence. The system is no longer only a<br>model. It is a network of possible actions.<br>This shift changes the problem of AI architecture. If a system has many possible paths, intelligence<br>is not only what it says. Intelligence is also how it chooses where to route attention, which tools to<br>activate, when to inhibit action, when to preserve memory, when to ask for human approval, and when<br>to treat an output as evidence.<br>This paper calls that underlying structure a <strong>neuro-network substrate</strong>. The term is intentionally<br>hybrid. “Neuro” points to activation, inhibition, trace, reflex, deliberation, and correction. “Network”<br>points to routes, surfaces, tools, memories, workers, registries, artifacts, and feedback loops. The term<br>does not claim that engineered AI systems reproduce biological brains. Instead, it provides a practical<br>vocabulary for systems that behave less like single functions and more like living operational networks</p>