Connectionist methods

Connectionist methods attempt to mimic the essential behavior of the brain using artificial neural networks. An artificial neural network is an abstraction of the networks of biological neurons that appear in the human brain. An artificial neural network is comprised of a large set of artificial neurons that are densely interconnected and have a simple mapping from input to output. While an artificial neural network is not engaged in the same kind of electrochemical interactions as a biological neuron, it shares similar macroscopic behaviors.

Connectionism was founded with an ambitious vision of creating deep, human-like intelligence (and commonsense). Today, it is primarily seen and applied to ‘narrow-AI’ problems of image recognition and other learning of non-linear functions.

Hugo de Garis’ China-Brain project stands out as a dramatic return to the original ideals of connectionism. The project is attempting to train thousands of high-speed artificial neural networks (using custom hardware) and assemble the networks into a larger super-network to drive the intelligent behavior of a robot. So far, the system is capable of recognizing limited objects but work is ongoing in order to build sophisticated behaviors.

There are other, less grandiose, attempts to apply artificial neural networks to challenging problems of general-purpose intelligences. Numenta Corporation’s Hierarchical Temporal Networks (HTM) are hierarchical layers of nodes that can perform spatio-temporal reasoning and learning; recurrent neural networks are being successfully applied to increasingly difficult learning problems in work such as that of Graves and Schmidhuber ; and Steven Thaler’s Imagination Engine is used to generate creative and novel inventions by adding noise to a trained neural network.