Kottke on Neurons

“Our brains have Oprah neurons, Aniston neurons, Eiffel Tower neurons, and Saddam neurons that fire when we see pictures or hear the names of these people and places.”Jason Kottke

While I’m all for public interest in science, especially neuroscience, its a pity when undecided questions are reported as solved.

The issue in question is one of the neural coding of semantic information. Jason and the New Scientist article he links to describe what is known as the Grandmother Cell theory. In short, the theory argues that most distinct semantic concepts each have their own dedicated neuron which fires when we access that concept.

The problem with this theory, despite the fact that our brains are never actually this simple, is that there simply aren’t enough neurons in the right areas to encode all the possible content we might encounter. What would happen when we run out of neurons?

An alternative to the Grandmother Cell theory is the Distributed Representation theory (also called a neural network), which argues that semantic content is encoded by the specific structure of connections between neurons. This, to me, sounds much more reasonable. Realistically though (and as seemingly suggested in the article, though they don’t outright say it) is that our brains probably work in a way that combines the two theories.

One thought on “Kottke on Neurons”

  1. There’s another possibility, that’s similar to the grandmother cell idea, but that permits more content to be held in the limited number of neurons (about 100 billion) in the human brain. The idea is that instead of 1 neuron representing 1 idea (idea can mean a sensory percept, a word/concept, a desire, etc), a group of neurons (a set of about 1000 neurons is sometimes suggested) represents the idea. At first, you should say that this should REDUCE the number of ideas that can be represented (1000-fold!). But the idea is that you have ‘overlapping sets’, i.e., idea #1 may occur when a given set of 1000 neurons is active (and reverberating in order to give some persistance to the idea over some time, say a second), and idea #2 is active when another set of 1000 neurons is active, BUT the set of 1000 neurons constituting idea #2 can have many of the same neurons that constitute set #1. So, a given neuron can contribute to many different sets. Imagine that the brain had just 2000 neurons, and each idea arose from a specific set of 1000 neurons. With overlapping sets you could represent a LOT more than 2000 ideas. To read more about this type of model of cognition,
    go to http://www.columbia.edu/~nvg1/Wickelgren/
    It’s the site of Wayne Wickelgren, a cognitive neuroscientist who passed away about 5 years ago from Motor Neuron Disease. He was an amazing and brilliant person (graudated summa cum laude from MIT). I knew him when I was teaching at Columbia – he had retired from research but continued to work on theories of cognition and the brain even as his movement ability was lost (he had a special device set up so that he could press a key on a keyboard with minimal motor effort). On his site he put free pdf’s describing his ideas about neural sets (he calls them ‘webs’) and coding for ideas.

Leave a Reply

Your email address will not be published. Required fields are marked *