Creativity: Generative Artificial Intelligence

Creativity means creating something that is new and useful. Bending, Blending, and Breaking are three different strategies to foster creativity. These approaches can question, alter, and combine existing ideas to create new perspectives and innovations. Deep Dreaming generates images by using a pre-trained neural network to optimize the input image. A target function is defined to alter the recognized patterns and features in the input image. The resulting image contains enhanced patterns and features, giving it a unique, dream-like appearance. In Style Transfer, the artistic style of one image is combined with the content of another image to create a new image. Here, content and style information are separated and recombined through a pre-trained neural network. Generative Adversarial Networks (GANs) consist of a generator and a discriminator network and are used to create deceptively real images or videos, known as Deep Fakes. The generator creates new candidate images or videos, while the discriminator tries to distinguish real from artificially generated ones. Over the course of training, both networks iteratively improve in their respective tasks. Diffusion models can generate images through a process known as Denoising Score Matching. By learning to reverse a simulated diffusion process, they can create entirely new images from noise. They can also be combined with language models to generate images from text descriptions.