Rise of the machines

To process an image, for example, the lowest layer is fed the raw images. It notes things like the brightness and colours of individual pixels, and how those properties are distributed across the image. The next layer combines these observations into more abstract categories, identifying edges, shadows and the like. The layer after that will analyse those edges and shadows in turn, looking for combinations that signify features such as eyes, lips and ears. And these can then be combined into a representation of a face--and indeed not just any face, but even a new image of a particular face that the network has seen before.