Artificial Convergent Evolution

Leo Isikdogan

January 19, 2022

Artificial Convergent Evolution is an artistic visualization of a novel AI-based convergent evolution process, inspired by convergent evolution in nature. It results from an experiment investigating the aesthetics that emerge when generative models are isolated and then pushed towards convergence.

Convergent evolution occurs when unrelated organisms independently develop similar traits due to similar environmental pressures. For example, the eyes of humans and octopuses have similar structures and functions despite not sharing a common ancestor. Similar environmental conditions can lead to convergence, resulting in similar appearances through different evolutionary paths.

This artwork utilizes a custom-designed AI art model to simulate this process of isolation, branching off, and convergence. Initially trained on a large, generic image dataset, the model's copies are subsequently trained in isolation to specialize in disjoint sets of images. Following this divergence, the copies are then pushed together, converging on similar evolutionary paths.

This novel artificial convergent evolution process helped generative models diverge from their original training sets while converging towards similar aesthetics, resulting in visually distinct and aesthetically pleasing images.

Available on Artsy.