Google Labs has published a very interesting blog regarding using neural networks that were trained to recognize objects to instead paint other objects.
We train an artificial neural network by showing it millions of training examples and gradually adjusting the network parameters until it gives the classifications we want. The network typically consists of 10-30 stacked layers of artificial neurons. Each image is fed into the input layer, which then talks to the next layer, until eventually the “output” layer is reached. The network’s “answer” comes from this final output layer.
They call this "inceptionism" and the results are more than a little bizarre. The following slides show the results of different neural networks "painting" the thing they were trained on even though the source is unrelated, or even random, data. The underlying mechanics are quite complex, but imagine you are seeing how a neural network "sees" the world.