What does a neuron do? A new model for Neuroscience and AI

Efficient coding theories have elucidated the properties of neurons engaged in early sensory processing. However, their applicability to downstream brain areas, whose activity is strongly correlated with behavior, remains limited. Here we present an alternative viewpoint, casting neurons as feedback controllers in closed loops comprising fellow neurons and the external environment. Leveraging the novel Direct Data-Driven Control (DD-DC) framework, we model neurons as biologically plausible controllers which implicitly identify loop dynamics, infer latent states and optimize control. Our DD-DC neuron model accounts for multiple neurophysiological observations, including the transition from potentiation to depression in Spike-Timing-Dependent Plasticity (STDP) with its asymmetry, the temporal extent of feed-forward and feedback neuronal filters and their adaptation to input statistics, imprecision of the neuronal spike-generation mechanism under constant input, and the prevalence of operational variability and noise in the brain. The DD-DC neuron contrasts with the conventional, feedforward, instantaneously responding McCulloch-Pitts-Rosenblatt unit, thus offering an alternative foundational building block for the construction of biologically-inspired neural networks.