I've started in on my dissertation work, and I thought I'd give an introduction to what I'm working on in case anyone is interested.

My interest has mostly shifted to trying to understand intelligence by understanding how the neocortex works, and by trying to model the most important aspects of it on a computer. A lot of this focus is due to the book On Intelligence by Jeff Hawkins. It's a great read, and the only really hard part to read is Chapter 6, which delves into the neuroanatomy of the neocortex. But it's very well written, and I recommend it to anyone interested in how people, animals, and ultimately machines might think.

His theory is very clearly stated and intriguing, but he's very short on a lot of details. He's gone on to co-found Numenta, a company whose technology is based on his theory. But from the white papers I've read, and what I've seen of the implementation, they're using an approach that doesn't necessarily seem to mesh very well with what's going on in the brain. And that's fine...the goal of an engineer is to get something to work, not necessarily to understand how the system it was based on works in its entirety. An engineer building an airplane needs to understand the basic principles of lift and thrust, not the specific ways in which a bird's wing flaps in order to fly.

Anyway, I plan on doing a series of posts explaining some of the concepts from my research, and in this first one, I'll give a brief tour of the neocortex, its organization, and the class of theories of which Hawkins' is a part of that try to explain just what the neocortex is doing.

So first of all, your neocortex is a big, flat sheet, crumpled up to fit in your skull. The neocortex sits on top of the rest of your brain. If you were able to take your neocortex out and spread it flat, it would be about the size of a large dinner napkin, or the diameter of a large pizza. If you thought about the two hemispheres, each would be about the size of a sheet of notebook paper. The neocortex has six layers, with layer one, the outermost layer, nearest your skull, and layer six the closest to the center of your brain. Each layer is about the thickness of a playing card or a business card, so if you stack six cards together, you get a sense of how think your neocortex is.

Current estimates are that your whole brain has about 100 billion neurons. The neocortex has about 20 billion neurons, but the cells and connections take up about 75-80% of the volume of the human brain.

It's called the neocortex because it's the most recent evolutionary addition to your central nervous system, and this particular structure is a feature of mammals. Besides being structured horizontally in layers, the neocortex has a very distinct vertical organization. The prevailing theory is that the smallest functional unit above the level of the neuron is the minicolumn, a vertical structure composed of about 60-120 neurons. Minicolumns work together in groups called macrocolumns, or just columns.

The prevailing line of thought is that a precursor structure to the minicolumn evolved in reptiles, and that it was a general-purpose module that enabled new functionality (e.g. language) and increased functionality (e.g. better visual processing) just by scaling...adding more minicolumns. Many theorists think this because minicolumn size, structure, and connectivity is extremely similar between and within mammalian species.

The neocortex evolved on top of the reptilian brain, like adding an additional scoop of ice cream on another, and it was so good at what it did, it basically took over all the basic functions previously carried out by older parts of the brain. Your neocortex is now primarily responsible for processing input from all your senses, for language use, abstract reasoning, planning, decision making, and action.

Some theorists, including Hawkins, think that they key to understanding what we think of as intelligence, is understanding what the neocortex does...specifically what the function of the minicolumn is.

I think that's enough for one blog...so next time I'll talk about some ideas about what kind of algorithm the minicolumn is carrying out.