top of page

TWO ROCKS TALKING ABOUT THE WEATHER - 2022

Shown at the Sarabande Summer Group show, August 2022

Swiss Rock | Materials - Natural Boulder, Nylon, Resin,  Rhizocarpon geographicum, Mild Steel, Acrylic rod.

SB_SGS_LOUIS_ART_2022_3.jpg

Two rocks talk to each other about the weather. The rocks have become the bodies for two artificial language models, and between them they have a quite narrow perception of the world. This work was inspired by Blake Lemoine’s attribution of sentience to the computational conversational model laMDA as a question to whether anything without a body can truly be a 'mind'.

Their language is built around weather data taken from stations close to their respective extraction sites. One from Vale de Bagnes in Switzerland and another from Pendle in Lancashire. Can these rocks begin to comprehend each others experiences? And what, if anything, is gained from these repetitive acts of climate smalltalk?


The two characters attempt an exchange over messenger, each initially unclear of the others meaning until they begin to build a shared language with each other. Climate smalltalk is something we can all understand.

The level of comprehension between the two rocks is what is represented through the displays. With more clear statements being understood and blurrier statements for incomprehension. English text was used as an overlay to help viewers tap into their conversation.

The language model training was assisted by Dominic Oliver using a form of reinforcement learning, to repeat weather conversations over a 3 year period of collected weather data between the two rocks.  The visual interface was developed by Rifke Sadleir to imitate text messenger applications. 

 

304786288_5875914902419831_6006575969322618874_n.jpg

Lancashire Rock | Materials - Natural Boulder, Cotton Rope, Natural latex,  Lecidea lithophila, Mild Steel, Earth pigments.

The reinforcement language model was trained by Dominic Oliver with the visual interface and scripting  by Rifke Sadleir of DXR Zone. The language model was trained over 30 cycles of 3 year weather data, captured at hourly intervals. At each cycle, comprehension builds between the two rocks. 

bottom of page