ҹ1000

A classical concert at the Barbican shows how AI interprets a recital

Sight Machine, an AI project at the Barbican, shows how the algorithms used in techniques like facial detection capture the performance of a string quartet
Kronos quartet
As the Kronos Quartet plays, a screen shows machines trying to make sense of the performance
Bruce Guthrie

, Barbican Hall, London

ON THE evening of 11 July in London’s Barbican Hall, the celebrated Kronos Quartet will play for the benefit of a dozen machine eyes. There will be a human audience, too, following whatever it is the machines see as their output is fed to a big screen.

This is the European premiere of US artist Trevor ʲ’s Sight Machine. Last year, during its debut at the Smithsonian American Art Museum in Washington DC, the video feed showing the AI view of the musicians faded away to leave only neon lines on a black background. This made no sense to the human audience but reflected the way the machines grasped events on stage.

The system’s off-the-shelf pattern-recognition algorithms, combined with ʲ’s own software, should offer unique insights again. “Sunny [Yang, a Kronos musician] is 94.4 per cent female” the Smithsonian audience was told. Then, as lights caught her cello strings, the machines insisted “Sunny is holding a pair of scissors”.

Kronos lead violinist and founder David Harrington has encountered AI before. “I remember when we played what might be the first piece written by a computer for a string quartet. It seemed so bizarre and kind of clunky.”

The Barbican concert turns that on its head: the music is composed and performed by people, it is the responses to it that will be artificial.

“The first time we performed Sight Machine, the Kronos guys were asking, ‘Why are people laughing at us?'” Paglen recalls. “But the laughter was about how preposterous the machine responses were.” This is the underlying theme: what can the machines see or not see – and what will they misinterpret?

Paglen wants to challenge how we view AIs. Invoking intelligence to describe them is wrong, he says. It can lead us to grant them a superhuman infallibility. “This is convenient for people trying to raise money [for AI-based applications], but it’s incredibly dangerous.”

Topics: algorithms / Artificial intelligence / Music / Software