MAI is an interactive installation highlighting the water consumption of AI in Microsoft’s data centres. It visualises the environmental cost of AI, showing users that each AI response consumes 3 litres of drinkable water, urging a rethink of our digital footprint and its hidden impact.
How did my research start?
As an autonomous artist and a graphic designer, I am interested in emphasising on topics which are not widely discussed. Topics which would make people rethink their behaviour and which have been kept under the radar. One of the latest developments happening is made by a company named SoulMachines, which is creating the closest version to a digital human. They are mimicking the human neural systems and brain, striving for the creation of authentic human reactions. This inspired me to critically examine the process of humanising AI and to research the costs to the planet. I have designed a digital version of myself, linked to the knowledge of the latest ChatGPT and programmed it to be able to have a normal human conversation. This AI-driven persona is programmed not only to answer questions but also to educate users on drinkable water consumption in data centres by presenting relevant facts. Simultaneously, the installation visually represents the water used for generating each response by pouring it into a container.
What is the consumption?
During my research, I discovered that ChatGPT consumes approximately 500 ml of water for every 5 to 50 questions answered. For image generation, the energy consumption is significantly higher, with text generation using 0.042kWh and image generation requiring 1.35kWh—a 32-fold increase. This translates to about 3.2 litres of water used for generating a single image, based on a proportional calculation.