I’m a neuroscientist and some of my work is in single-cell recordings. I’m impressed how similar the firing of some neurons is to a “concept”. I’ve worked with hippocampal place cells — cells that fire in a particular environment whenever an animal (rat) crosses a particular region of environmental space. O’Keefe and others propose that these are the key elements in a cognitive mapping system. Individual place cells do roughly as good a job identifying a specific location in the environment as an animal can do when trained to go to an arbitrary, unmarked place. From what I can see, these neurons have the concept of a particular place nailed.

The important point, as I see it, is that the firing of single neurons can parallel perception. This, in itself, is amazing.

About 50 years ago Hebb introduced the concept of the cell assembly. The cell assembly idea is that the co-active firing of a set of cells creates objects of perception — concepts. Additionally, the more these cell fire together, the more their co-active firing will strengthen and the stronger the concept will be. Most importantly, this does not mean that the members of an assembly cannot be parts of other assemblies.

How does the cell-assembly idea relate to grandmother cells? I can imagine two types of grandmother assemblies. In one, each of the neurons is a grandmother cell. When they fire together, as in the Portnoy example, the brain has the concept of “grandmother”. And any cell in the assembly would fire whenever grandmother was in mind. In another type of grandmother assembly, the elemental cells would be feature detectors: one firing to eyes that looked like my grandmother’s eyes, etc. All members of the assembly would fire whenever my grandmother’s face was presented, etc. But these cellular elements would not show much specificity. Each would fire in many other situations.

But an interesting feature of cell assemblies is that their elements can, in theory, grow from simple feature detectors to “grandmother cells”. the more that the assembly fires as a unit, the more activation of subsets of its elements will cause the entire assembly to discharge. So, although one of the cells might initially fire to particular eye features, with training, it can learn that these eye features are often accompanied by other features of my grandmother, and the other features can gain control of the cell’s firing. Slowly, the elements may change from low-level feature detectors to represent higher-order concepts.

Overall, I’m amazed at finding high-order neural elements in the brain. I don’t for a second think that single cells can signal perceptions or that all perceptions will have clear single cell correlates, but the existence and number of high-level correlates is striking.

2. Carl, I don’t understand your comment: “But they might not be the same neurons from one subject to the next”. There is very little likelihood of identifiable single neurons comparing one vertebrate brain to another. Mapping single neurons works in some invertebrates with very small brains, but not vertebrates. the general concept is that populations of neurons map from brain to brain, but single cells don’t. You can compare amygdalas across two brains, but you can’t compare specific, identifiable cells in the two amygdalas.