Family culture essay

Terraform Table

Commission by:


The V&A Museum in London commissioned Tellart to create an installation for The Future Starts Here, a major exhibition exploring the implications of rapidly evolving technology on our world. Tellart's Terraform Table is a machine-intelligent sand table that invites visitors to shape landscapes and seascapes with their hands. It playfully explores the notion of remodeling the Earth's oceans, coastlines and mountains in response to climate change—or even for terraforming other planets.

As a designer on this project, the main challenge was to design "with" the algorithm and create new design processes and toolsets that allowed for co-creation and custom datasets. 

Commissioned for the Planetary section of the exhibition, this speculative project addresses the ethical and environmental question, “Should we shape the Earth and other planets for human use?” As we contemplate missions to Mars, the Terraform Table poses the ethical question of whether humans have the right to colonize space and its pristine celestial bodies.

As visitors form hills and valleys in the sand, a machine learning algorithm reads the relative height of the sand and generates artificially composited rivers, lakes, forests and snowy peaks. These are projected onto the sand in real time, giving the impression of a photorealistic 3D landscape that responds to the visitor's touch. The projected imagery of land and water is derived from thousands of satellite images of Earth that the algorithm was trained to correlate with altitude data.

While the resulting landscapes appear continuous and real, they are actually created through the intelligence of a predictive model. A single artificial coastline could contain qualities extracted from hundreds of locations distinct as California, the Persian Gulf and the Japanese Archipelago—a synthesis of “Earthness” intended to spark a new and emotional connection with our planet.

We trained a machine learning algorithm on thousands of real satellite Earth images and corresponding elevation data. Over time, the algorithm “learned” to correlate the height of different landforms (image on left) with their appearance from orbit (right), in order to generate an artificial satellite image (middle).

Tensorflow, Pix2PixHD, python, spout, openFrameworks


Selected press: