An article in The Verge (Reference: https://www.theverge.com/2018/10/25/18023266/ai-art-portrait-christies-obvious-sold) reported that an artwork named Portrait of Edmond Belamy created using an artificial intelligence algorithm was sold for $432,500. This artwork was estimated to sell for about $7,000 to $10,000. The deep learning algorithm that was used to create this artwork is called a generative adversarial network (GAN). The unique attribute of generative adversarial networks is that two deep networks are made to compete against each other to generate something meaningful. The two networks that compete against each other and try to outsmart one another are called generator and discriminator networks.

Consider a situation where we want to generate new handwritten images of the digit five. A generative adversarial network in this case would involve a generator network that creates fake images of the handwritten digit five from simply random noise and sends it to a discriminator network. The fake images are mixed with genuine images and the discriminator network, which is trained to differentiate between real and fake images of the handwritten digit five, will try its best to successfully differentiate between real and fake images. These two networks are made to compete against each other until the generator network starts making realistic-looking fake images that the discriminator network finds increasingly difficult to differentiate between. In addition to image data, application of generative adversarial networks can be extended to generate new text or even new music. We will illustrate an application of a generative adversarial network to generate new images in Chapter 8, Creating New Images Using Generative Adversarial Networks.