Panic Breeder presents a quasi-nature documentary of imaginary fantastical species as the future inhabitants of the Baltic Sea. Taking inspiration from the evolution of species theorized by Charles Darwin, it comments on the iterative and adversarial capabilities of Artificial Intelligence to probe the terrifying future of the “survival of the fakest”. The artwork uses GAN (Generative Adversarial Network), a class of machine learning systems, where two neural networks (Generative & Discriminative) compete with each other for several training sessions. The trainings utilize a large dataset of extinct and endangered birds of the Baltic Sea to generate new images. The animation created from these fake images presents an illusory transformation and evolution of the species. Furthermore, by quantifying the energy consumption of machine learning, from among the machine-generated breeding a singular creature is manifested in physical form. In this fictional scenario, the loss of species is counter-balanced by the possibilities of AI-generated creatures to act as a temporary, if not futile antidote to climate change panic and anxiety.
The images carry their own production data: training on NVIDIA V100 GPUs for 97 hours, a power consumption of 114.9 kilowatt-hours and greenhouse gas emissions of 81 kilograms of CO₂.
Does computation take over from evolution?