The Self as a Dynamical Attractor: Emergence of Self-Like Dynamics in Spiking Neural Networks under Continuous Embodied Input

The nature of the self is traditionally modeled as a persistent internal state, memorystructure, or representational object. However, empirical phenomena such as the temporarydisappearance of subjectivity during deep sleep or general anesthesia challenge models thatrequire continuous preservation of specific neural states.In this work we propose an alternative hypothesis: the self emerges as a dynamicalattractor of neural activity rather than as a stored internal representation.Using spiking neural network simulations implemented in the Brian2 framework, weshow that self-like functional properties can arise from slow adaptive dynamics driven bycontinuous endogenous input. A recursive observer layer introduces predictive modulationthat alters but does not generate the underlying dynamical regime.We further introduce a recursive architecture in which slow embodied dynamics interactwith a faster observer layer that generates predictive modulation. A parameter sweepof prediction gain reveals that recursive prediction modulates but does not generate theunderlying attractor dynamics.These results support a view of the self as a transient dynamical structure emerging fromthe interaction between embodied input, neural adaptation, and recursive prediction. This repository contains the article, simulation code, and data required to reproduce the experiments.