Fear Exchange

Fear Exchange is an AI-powered telematic installation, first shown at the Dark Matters exhibition at Science Gallery Melbourne with Arts at CERN, that creates a window into another world, enabling distant peoples to share their fears. It provides an insight into the lives of others and a way to discover what unites and divides us. 

The project was originally installed in London, Hong Kong and Melbourne. In each location participants encounter a sculptural ear and mouth placed on a plain wall and are invited to confide their deepest fears into the ear sculpture.

These whispered fears can then be heard by participants in the paired location when they lean into the mouth sculpture.

The project invites people to share their innermost concerns, but with the safety and anonymity of distance. What does it mean to give and receive fears with people you may never meet, and may have little in common with? What new form of intimacy and understanding is generated through this encounter?

A universe of fears

Collected recordings are analysed using a Large Language Model’s embeddings and are clustered by theme creating a three-dimensional interactive visualisation of a universe of fears.

Constellations of fears

The collected fears illuminate the commonalities and differences that shape our global society, ranging from disparities in wealth, politics, and social status, to more intrinsic differences such as age and gender. Each location’s database of fears is also projected in 2D space using LLM embeddings.

The result is graphically re-imagined as fear constellations in the night sky where every location's constellations are diverse and reflect the specific concerns of that location.

A fully automated pipeline using AI

Instead of physical sensors to detect when a person is sharing a fear and a human sifting through all the recordings, we created a custom pipeline where the microphone in each location records the whole time. Then we used AI technology to transcribe the audio and we used a fine-tuned LLM model to automatically detect expressed fears in the transcripts. Using the associated timestamps of the recorded fears we automatically clipped the audio segment of interest and after a content moderation by an LLM agent, uploaded the recording to the other locations. See the pipeline schematics below.

By transforming the physical interaction problem into a language one, we took full advantage of the latest LLM technology to make the installation fully autonomous.

Design and Implementation: Theo Papatheodorou / Jesse Wolpert

Technology: Sounddevice audio // gpiozero // Firebase // Raspberry Pi // sensors // microphone // speakers // LEDs // wooden enclosure // 3D printed models // chatGPT API // custom audio processing pipeline

Credits: Soundtrack is Spirit Blossom by Roman Belov // 3D ear model by Alex Lashko CC Attribution license // 3D mouth model by Jack Evans CC Attribution license

Location: Melbourne, London, Hong Kong

Exhibited: Dark Matters exhibition, Science Gallery Melbourne / CERN Arts, Australia