Brain-computer interfaces decode motor intent through population dynamics—not single neurons

The dominant model in BCI research treats each neuron as an independent signal source. Record enough neurons, average their firing rates, and you can extract a movement trajectory. This view is wrong.
The brain encodes motor intent through coordinated population activity, not individual spike counts. Single neurons are noisy. Population dynamics are stable. The key insight from recent work: we should decode latent factors from neural manifolds, not firing rates from channels.
Full analysis below ↓