Every neuron in primary visual cortex has a preferred orientation — the angle of a bar of light that makes it fire most strongly. But it doesn't fire only at that angle. It responds over a broad range: a cell preferring 90° still fires substantially at 70° or 110°. A single cell's response is ambiguous. From its activity alone you can't determine the stimulus.
The population solves this. When a stimulus appears, each neuron contributes a weighted vote in the direction of its preference. The sum of those votes — the population vector — points at the actual orientation with a precision no single neuron achieves.
Move the stimulus. Watch all eight neurons respond simultaneously. The individual responses are broad and overlapping. The decoded vector is not.
The key result: the decoded estimate is almost always more accurate than the response of even the best-tuned neuron for that stimulus. Widen the tuning curves — each neuron becomes more ambiguous, but the population vector barely degrades. Narrow them — the simulation looks more like a labeled-line code, and the advantages of population coding diminish.
Georgopoulos, Schwartz, and Kettner (1986) described the population vector in motor cortex: individual neurons in M1 have broad directional tuning, but the weighted sum of their preferred directions accurately predicts the direction of arm movement. The same principle operates in V1, olfactory bulb, the hippocampus, and elsewhere.
The representation is the pattern across the population, not any single neuron's activity. There is no place where the orientation is stored in isolation. The gap between "this neuron fired" and "the stimulus was 45°" is not filled by one cell — it is filled by the relationship between all of them.