Beyond static perception: Animals, neurons and synapses move to compute efficiently

Neural computation is not performed by static circuits processing signals corrupted by noise, but by actively moving biological structures that use motion itself to sample, encode, and predict the world. Neurons are often modelled as filters that transmit information through chemical and electrical signals constrained by noise and bandwidth. However, this static view contrasts with biological reality, in which dynamical processes operate across scales, from whole-animal movements to motion at the subneural level. Here, we combine recent experimental observations with biophysically realistic modelling to show that neural information processing, beyond electrochemical signalling, is dynamically shaped by motion across biological scales, from morphodynamic ultrastructural changes to whole-body movements. In this framework, ultrafast mechanical adjustments in cellular and synaptic structures interact with retinal, eye, head, and body motions to accelerate encoding and enhance precision. Adaptive variability in self-motion-coupled morphodynamic sampling arises from changes in response waveforms, latencies, refractoriness, and ultrastructural dynamics. This variability improves signal fidelity and extends spatiotemporal resolution, enabling neurons to generate reliable, high-speed representations with minimal delay. Thus, through active sensing, animals continuously enhance the speed and reliability of sensory representations. This perspective, in which self-generated sampling motion across multiple scales enhances encoding and perception, offers new insight into how the brain achieves efficient, predictive, and noise-resistant computation while providing a foundation for future experimental tests and biologically inspired AI designs. We first explain how motion-coupled encoding improves neural performance, focusing on edge-coding in early sensory systems. We then extend these ideas more speculatively, proposing how motion-coupled sampling may have influenced the evolution of neural architectures at multiple levels, potentially contributing to efficient predictive cognition.