The Self as an Emergent Adaptive Filter: A Spiking Neural Network Study

The nature of the self is traditionally modeled as a persistent internal state, memory struc-
ture, or representational object. However, empirical phenomena such as the loss and recovery
of subjectivity during deep sleep and general anesthesia challenge models that require contin-
uous state preservation. In this work, we propose and computationally validate an alternative
hypothesis: the self is not a stored entity but an emergent adaptive filtering process that arises
during ongoing sensory interaction.
Using spiking neural network (SNN) models implemented in Brian2, we demonstrate that
self-like functional properties emerge selectively in response to temporally correlated inputs
and collapse under decorrelated stimulation. Through a series of controlled computational
experiments, we show that (i) organized network dynamics depend on temporal correlation
rather than signal energy, (ii) apparent memory effects arise from altered adaptive dynamics
rather than stored representations, and (iii) functional identity can be restored after complete
loss of activity, despite the absence of preserved instantaneous state.
We further introduce a formal decomposition of system state into fast and slow components,
providing a principled account of how self-like behavior can disappear and re-emerge while
maintaining functional continuity. The results support a view of the self as a transient, input-
dependent adaptive filter rather than a persistent internal object.