When you remember eating a biscuit at your grandmother's kitchen table, the memory feels unified — the biscuit, the table, the kitchen, the grandmother, all fused into a single experience. The intuition is that individual neurons store the complete scene, binding content to context in a single representation.
Bausch, Mormann, and colleagues at the University of Bonn recorded from individual neurons in epilepsy patients performing memory tasks and found the opposite. The brain maintains two separate neural libraries. “Content neurons” respond to specific objects or people regardless of the situation — the biscuit neuron fires for biscuits whether you're in the kitchen, the office, or the garden. “Context neurons” respond to the situation regardless of what's being viewed — the kitchen neuron fires for the kitchen whether you're seeing a biscuit, a dog, or a face.
The libraries are separate but coordinated. When retrieving a memory, activity in content neurons predicts context neuron responses within milliseconds. The biscuit neuron fires, and within a few milliseconds the kitchen neuron responds — as if activating the content pulls up the associated context through a learned association between the two populations. Pattern completion runs between the libraries, not within them.
This differs from rodents, where individual place cells combine spatial and content information in single neurons. Human memory, apparently, keeps the what and the where segregated. The advantage is flexibility: the same content neuron can participate in unlimited contextual memories without requiring a unique cell for each combination. The biscuit neuron works in any kitchen. The kitchen neuron works with any object.
The separation is the efficiency. A unified code would need n × m neurons for n objects in m contexts. Two separate libraries need n + m. The human brain solved combinatorial explosion by refusing to combine.