A team from Art-A-Hack™ 2025 Special Edition - CETE-P and Fulbright, Prague Czech Republic - Ecological-Technological Entanglements

Generative AI makes things that are time-consuming and difficult for people – research, coding, translation or making image and videos – incredibly quick and easy, so much so it can seem almost magical. This is possible due to the hugely energy-, resource- and labour-intensive processes needed for AI’s training and operation. The expensive hardware that powers AI needs vast amounts of electricity and water, but AI also needs human knowledge and creativity as its inputs, and people to test and correct its outputs.

PumpGPT is an AI interface made of several connected parts – a camera, computer and projector, a water pump, microphones and sound – that participants interact with through physical handwriting and pumping. It uses the same powerful technologies that are behind OpenAI’s ChatGPT but changes the way we access them. Hand-written prompts invite us to consider ours questions more carefully, and the physical pumping needed to get and answer makes us share some of the work of generating the output, giving participants time to think about their own possible answers to their question.

Generative AI is a machine. It is vastly more complex technically than a water pump, but the principle isn’t so different. Work needs to be done to make something accessible and useful, whether its water or information. This installation aims to make the similarities more obvious. If we slow the process down, and give ourselves time to think for ourselves, the AI generated answer might not seem so magical after all.