No single neuron knows which way you’re about to reach — each one just fires most for its own favourite direction and a little for the neighbours. But add up all their votes, each pulling toward its preference by how hard it fires, and the crowd points exactly where you’re going. Meaning from a mob. Slide the true direction and watch the vote track it.
Each neuron has a preferred direction and a cosine tuning curve: it fires at rate max(0, cos(θ_stim − θ_pref)). Alone, any one is ambiguous. But the POPULATION VECTOR — the sum of each cell’s preferred-direction unit vector weighted by its firing rate — points at the true stimulus. The instrument builds it from 12 cells live. A fail-loud self-check throws unless the decoded angle matches the true direction to within a few thousandths of a radian — Georgopoulos’s finding (1986) that intention is read from the crowd, not the cell.
Idealised noiseless cosine tuning with evenly-spaced preferences; real motor cortex has noise, non-uniform tuning, and needs many trials (and richer decoders now exist). The cosine model and the population-vector recovery are exact.