The Offline Optimum: Why Sleep is a Computational Necessity in Complex Brains

The evolutionary persistence of sleep in complex nervous systems remains a biological paradox. While its restorative functions are well documented, they describe what happens during sleep rather than why a complex organism must repeatedly enter a state of profound vulnerability and unconsciousness to perform them. Here, I propose the Computational Necessity Hypothesis: that sleep is an unavoidable consequence of a non–modular, high–dimensional neural architecture optimized for flexible behaviour. In biological neural networks, learning is distributed and overlapping; optimizing synaptic weights for a new task inevitably creates interference with prior capabilities ("catastrophic interference"). I argue that sleep provides the necessary "offline" state for a global optimization process – traversing network configurations to find a system–wide equilibrium that cannot be computed during real–time sensory–motor processing, i.e. wakefulness. This framework generates testable predictions about stage-specific cognitive deficits and representational drift across waking and sleep.