Universal abstractions: visions produced by early layers of the network, at the interface between the inner and the outer worlds. These representations emerge over and over again on various datasets representing the basic elements of cognition. Isn't that surprising, that beauty emerges way earlier than the human interpretable meaning?
We witness the ethereal dance of patterns hidden within the depths of a neural network. The faces of dogs, both familiar and surreal, emerge from the swirling chaos of colors and textures. This is not merely a digital hallucination, but a glimpse into the very essence of artificial creativity.
DeepDream Origins #* is a series of generative video artworks that harnesses the power of the same Inception network that gave birth to the original DeepDream experiment in 2015. Back in the day, DeepDream explored the uncharted territory of machine perception. Inception network that was trained only to recognize images (with some bias towards dog breeds), surprisingly possessed latent generative capabilities when executed in a specific manner. In a way, this foreshadowed the "prompting" phenomenon that has since captivated the world, where the vast potential of models trained on massive datasets can be unlocked by querying them in the right way.
DeepDream's ability to create images through iterative refinement anticipated the generative boom of denoising diffusion models that swept the AI landscape in the 2020s. This pioneering technique paved the way for the development of increasingly sophisticated AI art models, blurring the lines between human and machine creativity.
As we gaze upon the hypnotic beauty of this DeepDream-generated artwork, we are reminded of the boundless potential that lies within the intersection of art and technology. It is a testament to the human ingenuity that has harnessed the power of AI to unlock new realms of creative expression, inviting us to explore the infinite possibilities that emerge when we dare to dream beyond the boundaries of our own imagination.