Machine Hallucinations – Large Nature Model (LNM), 2024

Machine Hallucinations – LNM is a unique audio-visual collection generated by using Large Nature Model, the first open-source generative AI model focused on nature images and sounds, pioneered by Refik Anadol Studio. For this ongoing research, Refik Anadol and his team have ethically harnessed more than half billion open-source data from prominent organizations such as the Smithsonian Institution, National Geographic, and London’s Natural History Museum, as well as by venturing to collect data from various rainforest locations around the world. This collaborative effort has enabled the collection of the world’s largest archive of images and sounds showcasing the diversity of fauna, flora, fungi, corals, and rainforest landscapes. The artworks’ realistic and immersive soundscapes were generated and curated by using cutting-edge pre-trained audio neural networks and a custom audio latent diffusion model trained on a broad spectrum of complex nature sounds.

The Large Nature Model (LNM) lies at the heart of DATALAND, a museum and Web3 platform dedicated to data visualization and AI arts being developed by Refik Anadol Studio. Computing power for the Large Nature Model is driven by DATALAND partners Google Cloud, pursuing net-zero emissions across its operations and value chain by 2030, and NVIDIA, whose groundbreaking neural network algorithms and advanced tools enable Refik Anadol Studio’s visual representations and next-generation computer graphics. This research integrates the aesthetic principles of Machine Hallucinations with innovative AI techniques crafted for the LNM’s latest outputs.

Machine Hallucinations – Unsupervised – MoMA , 2022

What would a machine mind dream of after “seeing” the vast collection of The Museum of Modern Art? In other words, if the corpus of images of the MoMA collection had been accomplished by a single artist, what would their dreams look like? Emerging from Machine Hallucinations, Refik Anadol Studio’s multi-year research project that investigates data aesthetics based on collective visual memories of humanity, “Unsupervised — Machine Hallucinations — MoMA” processes 138,151 images from MoMA’s collection in the mind of a machine.

For this work and the other works in this collection, Anadol processed the entire digitized archive of MoMA through StyleGAN2, an algorithm developed by NVIDIA researchers with adaptive discriminator augmentation (ADA). He then explored a latent space with a custom software called a Latent Space Browser, which Refik Anadol Studio has been developing since 2017.