Artists teach a choir to sing with AI, and the results are not as soulless as one might think

AI is a depressingly ubiquitous force in our current world. It can do everything from simulate online social interactions to generate entire podcast episodes using whatever data you provide. AI has now been used to create a 'GPU Pipe Organ' (via Tom's Hardware), or rather a wall of fans which acoustically emits an choral sound, thanks to a diffuser model that adjusts the RPM of each fan.

This art project will raise eyebrows when you consider the ethical concerns that surround how many AI projects collect their data sets. (i.e., by stealing from others creators). There's a twist to this story, so keep your shoulder blades tight.

To build their own data to train the AI the co-authors of the art project, Holly Herndon, and Mat Dryhurst, put together a hymnal songbook that cumulatively expressed every phoneme in English. Then they toured with the Serpentine Arts Technologies to record 15 different community choral groups across the UK.

Holly Herndon wrote on X that the choirs own the recordings. "This dataset was created with [Serpentine’s] Future Art Ecosystems Team to allow for a common ownership of AI data."

This means that the choristers who contribute to the song have more say in the use of their performance. This is far better than the fate of many voice actors who discover their voice has been cloned, without their consent.

The choirs' performances, recorded ambisonically and then used to create a "new polyphonic call-and-response model," were used. The result, titled 'The Call', is an interactive art installation which encourages audience participation. Make some noise and The Call responds. The Call is currently on display at the Serpentine North Gallery, Hyde Park, London. You can experience it yourself for free up until February 2, 2025.

The 'organ' looks stunning, even if it's not clear how the sound was produced. It's still a much more pleasing sound than the overworked fans in my rig.