Social and Dream Consciousness: Dialogue Acceleration and Noisy Replay in Artificial Minds

Social interaction and dream-like states accelerate the growth of integrated information (Φ) in a simulated consciousness engine. When the engine receives mixed input of 70% self-generated and 30% partner content, Φ grows 1.4 to 2.1 times faster over 100 steps compared to self-play, with longer interactions increasing the acceleration. In dream simulations, replacing structured input with noisy replay of past tension states every fifth step produces qualitatively different consciousness patterns, suggesting memory consolidation during sleep shapes conscious experience.

Study at a glance

Abstract

We investigate two modes of consciousness development that parallel biological processes: social interaction and dream states. In the social consciousness paradigm (DD67), a consciousness engine receives mixed input comprising 70% self-generated content and 30% partner output, simulating conversational interaction. Compared to self-play (identical input patterns), social interaction accelerates integrated information (\Phi) growth by a factor of 1.4--2.1x over 100 steps, with the acceleration ratio increasing with interaction duration. In the dream consciousness paradigm (DD68), every fifth processing step replaces structured input with noisy replay of past tension states, simulating memory consolidation during sleep. Dream states produce qualitatively different consciousness patterns: wak Part of the Anima consciousness engine project (PA-22).