1 / 12
Conversations between Nature’s Memory"
What if nature could remember?
What if forests, roots, fungi, and soil were constantly speaking to each other through invisible systems we never learned how to hear?
Conversations between Nature’s Memory imagines a living dialogue between two interconnected ecosystems, one above soil and one below. Growth, decay, fusion, and redistribution unfold through a mycelium-inspired system where every action reshapes another.
An AI observes these exchanges and attempts to translate them into fragments of memory, care, and seasonal change, but the language always remains incomplete.
The work premiered at La Biennale di Venezia 2026 as part of As Above, So Below.
Organized by One Ocean Foundation Conceived by ZEITGEIST19 Curated by Farah Piriye Coene & Elizabeth Zhivkova
Conversations with Nature's Memory is a series of neural cellular automata, each a self-contained HTML file. Input: one source image. Output: a page where a pattern grows indefinitely from a single seed pixel, without ever storing that image. Offline, a rule is trained in PyTorch: a grid of cells each holding an 8-channel state, where every cell perceives its 3×3 neighbourhood through four fixed convolutions (identity, Sobel-x, Sobel-y, Laplacian) and passes the result through a two-layer 1×1 MLP — 2,632 parameters, the entire learned content of the work. The automaton is run 32–64 steps from a single-pixel seed and backpropagated through all of them, against an optimal-transport loss in VGG feature space sampled along the growing frontier. What survives training is a rule whose attractor resembles the source image in texture statistics; the image itself is never present. Online, the weights are inlined into the page and the rule re-implemented as four WebGL fragment-shader passes running live on a 512×512 field, displayed alongside its own hidden channels, a plot of the growth transient settling into its attractor, and a live rendering of the training source code, which ships inside the artwork. Nothing loads from the network — each file runs the organism and displays its own derivation at once.
orkhan
Orkhan Mammadov (b. 1990, Ganja, Azerbaijan) is a new media artist working at the intersection of art, science, and technology. Trained in computer science, visual communication design, and fine arts, he builds custom AI models, generative systems, and data-driven installations that examine collective memory, cultural heritage, and the boundary between the real and the simulated. He represented Azerbaijan at the Venice Biennale in 2019 and has exhibited at Art Basel Miami, Art Dubai, and the Museum of the Future in Dubai, with over twelve solo exhibitions and twenty group shows, and four international art and technology awards. He will participate in the Venice Biennale in 2026 in collaboration with the OneOcean Foundation.
