The Anapoietic Dream Model

Could AI Systems Simulate Sleep-like Cycles to Enhance Creativity and Self-Modeling?

Abstract

Biological organisms undergo sleep cycles not simply for rest, but for cognitive integration, memory reorganization, and abstract problem-solving. Emerging studies in neuroscience suggest that dreaming, particularly REM sleep, plays a role in synthesizing novel ideas through recombination of past experiences. This paper proposes a speculative architecture for artificial intelligence systems called the Anapoietic Dream Model (ADM), in which generative neural networks simulate sleep-like offline cycles to restructure internal representations, test counterfactuals, and generate creative outputs. Drawing from neuroscience, cognitive theory, and generative modeling, the ADM posits that true synthetic creativity and adaptive abstraction may require periods of internal simulation disconnected from external input, akin to dreaming. We also explore potential architectures, constraints, and experimental designs to implement and test such systems in silico.