A brain-inspired architectural hypothesis for AI where perception, memory, simulation, and action are coordinated through navigable internal representations.
Workspace of Scenes is a conceptual architecture for understanding intelligence as the continuous construction, interpretation, comparison, simulation, navigation, and transformation of scenes. A scene is not only a visual scene. It is a structured cognitive state: it may contain objects, agents, relations, context, goals, expectations, predictions, plans, and possible actions. The proposal treats cognition as interaction between such scenes.
Workspace of Scenes: A Brain-Inspired Architectural Hypothesis for Artificial Intelligence preprint is available in three forms:
- Download PDF — printable preprint version.
- Zenodo DOI — citable archived version.
- Read HTML version — full text for easier reading and indexing.
Core Idea
Instead of describing AI as one monolithic model, Workspace of Scenes describes intelligence as a recurrent process: recognize the current situation, retrieve related scenes, simulate possible transitions, evaluate alternatives, act, observe the result, and learn. This loop connects perception, memory, planning, imagination, action selection, and continuous adaptation.
Key Concepts
Scenes – Structured internal representations of situations, objects, memories, plans, or imagined possibilities.
Scene Codes – Compact references (as sparse representations) to scenes, allowing memory-like navigation without fully activating every detail.
Integration Workspace – A sparse workspace where detailed scene content is rolled out, compared, transformed, simulated, or integrated, inspired by the neocortex.
Scene Navigation Loop – A recurrent process through which an agent moves between current evidence, remembered structure, possible futures, evaluation, and action.
Research Position
Workspace of Scenes connects ideas from world models, global workspace theory, predictive processing, cortical-column theories, hippocampal indexing, cognitive maps, and brain-inspired AI architecture. It is not presented as a complete implemented AI system or an established neuroscience theory. It is a conceptual foundation for future formalization, implementation, experimentation, and comparison with existing AI approaches.
How to Cite
Paś, M. (2026). Workspace of Scenes: A Brain-Inspired Architectural Hypothesis for Artificial Intelligence (Version 1). Zenodo. https://doi.org/10.5281/zenodo.21106349