Connectionism

Cognitive science is a field of psychology that studies the computational nature of the mind. One sub-discipline of this body of thought is called connectionism. Connectionism places emphasis on the study of neural networks as the key to understanding the human mind. I'll back up for a second if you're not sure what a neural network is.

Neural networks are, like the name implies, networks made up of neurons. Maybe that's not a very useful definition. 'Neurons' in this case can be physical, but the type of neural networks people usually speak about are mathematical constructs. These neurons can be connected to other neurons and with the right sort of tinkering we can create networks that display complex pattern recognition abilities. But before we get into that maybe we should focus on some basic facts about neurons.

At the simplest computational level, our brains are made up of neurons. Every thought you've every had, every emotion you've felt, has been due to a collection of neurons in your head rapidly firing at each other in a specific pattern. If you're familiar with computer architecture, this is similar to logical gates in a circuit--although the biological computers in our heads operate very differently. In the early 20th century famed psychologist William James proposed his theory of association, which basically proposed that if a neuron in one part of your brain fires and causes another neuron to fire it strengthens the bond between the two neurons, associating them with each other. This association makes them more likely to fire at the same time in the future and essentially forms the basis of the way our brains learn. If you touch a hot stove your mental concept of a heated burner will very quickly be connected with the idea of pain. This makes sense at an intuitive level, and is in fact a very old idea (Aristotle seems to be the first to propose it).

But how are ideas and feelings and memories represented by neurons? Well that's where connectionism comes in. Ideas are collections of associated neurons that are distributed throughout the brain. Learning is the process of reinforcing those connections. We can model this in neural networks using mathematical entities called perceptrons.

At a basic level, perceptrons take a set of input values, pass those values through a mathematical function, and output the result into a set of output nodes. Each of the input-output connections has a weight value that is altered during the process of training a network. The details are too technical to get in here (though not especially complicated), but the basic idea is this--a network undergoes a series of training where the networks 'learns' to respond to input values with the right output values. The weights between the connections are altered depending on the correctness of a networks output response. This training continues until the errors in the network response have been minimized. The cool thing about neural networks is that once we have trained a network we can present the network with stimuli is has never seen before and it can still produce a correct response. Your brain is capable of recognizing a cup as a cup even if you've never seen that specific coffee cup before. The same pattern recognition process is present in neural networks.

So that's great, right? We can program a network to do everything a human can do using neural networks. Well, theoretically, but it's a lot more complex than that. Neural networks are still unable to learn on their own in the sense that humans learn on their own. There are also a number of immensely difficult problems relating to the human brain that have to be solved before we can begin to work complex thoughts and ideas. The cerebral cortex--the part of the brain that controls higher-level thoughts and emotions--is the most recently evolved part of the human brain. It is also the most sophisticated, and unless we are able to understand the lower-level functioning of the brain we still have a long way to go before we can fully understand consciousness. But that's a subject I'll have to take up some other time.