Stepping into Holly Herndon and Mat Dryhurst’s exhibition The Call at the Serpentine North Gallery the first work you see is The Hearth. A display of GPU fans, which play a role in cooling the graphics cards that “make fast and large-scale computation by AI models possible”, whizz and whirr, creating song that’s almost sweet in the way music made from any appropriated object can be. The instrument, as it’s called, is said to represent the layers of nodes in a neural network with its repeating blocks and cables, which rise at the centre above a brass engraving that shows a child blowing a lur. The angled panels between the blocks of fans use the ancient wind instrument as a motif, alongside designs which abstract the fans to the point of them being flowers and radial designs that appear like light shining or signals reaching.
Discussions of AI often revolve around the practicalities and implications of the technology, trying to unpack what consequences it will have for work, consumption of media – and even truth. Other times, the question of AI is thrown forward: the reality is not yet known, and therefore speculated upon. The Call takes a different approach to AI. Religious imagery and spiritual sentiment are used to portray a future where generative AI is a fully-fledged creative tool, data collection is an inevitability and we are but subjects succumbing to computation.
The exhibition uses AI models trained on choirs from across the UK who sang from a book of hymns, exercises and a recording protocol. The engraved brass panel in The Hearth protects a backup of one of these choral models, called Linked Diffusion. This model has been used to create choral music that is played from speakers around the interior wall of the gallery. This music isn’t sung, it is generated, and the inhuman voices begin to sound like the machine-generated, ethereal voices used on the album Models by Lee Gamble. Yet, while Gamble’s voices expose an eerie sense of the unknown that lingers in black boxes, Herndon and Dryhurst’s voices fill the empty gallery perimeter in a way like music that fills a church as if from a divine source.
In the Serpentine’s South Powder Room The Wheel provides more insight into how the data used by the AI models was generated and collected. The Wheel is said to be partially inspired by the microphone rig that was used to capture the choral singing, with a hanging orb in its centre representing the ambisonic microphone that was used to record the choirs and their environment. Religious imagery continues here with the shape of The Wheel inspired by chandeliers, which can symbolise divine light. A frieze shows winged choristers. Herndon is depicted with golden hair, evoking a gilded halo. A panel depicts the recording setup. Like much religious art, the depictions create a visual story from which to learn, but with its visible screws and chain quick links, The Wheel veers towards a pastiche of the ecclesiastical. It almost seemed a shame to read about and experience the depth of work that went into creating the sonic aspects of the show, only for the fabrication of this piece to not live up to the liturgical objects with which Herndon and Dryhurst draw comparison.
Hanging from the The Wheel is the book of the hymns and exercises which you can read as you listen to snippets of a choir working through its content playing in the space. The lyrics are of indoctrination:
Submit yourself to visions of new forms.
Submit yourself to something more.
O’er wires it comes to us.
O’er time we have to learn to trust.
And worship:
There is no end to what we are.
It’s endless you provide.
There is no end we’ve come this far.
We’re endless with you, our guide.
There is no end to how we feel, held on and now you’re real.
Endless with you by our side.
Endless with you as our guide.
The appeal to the divine seems at odds with Herndon and Dryhurst’s use of AI in elsewhere as well as how they’ve acted as (not uncritical) spokespeople for the technology: Herndon ‘birthed’ an AI baby called Spawn which was used on her album Proto; their search engine tool Have I Been Trained? lets you see if your work is the data set used to train Stable Diffusion, a deep-learning model used to convert text to images; and last year the duo released xhairymutantx, an AI image generator trained on images of Herndon, which in turn acts as an example of – and experiment in – how public data is a malleable. The fourth work on display, The Oratory, seems to fit into the advocacy role the artistic duo have assumed. It is a curtained room with a white and gold relief depicting humans climbing a mountain path. A white microphone is in front of the relief. When you sing into it, one of the duo’s choral models will respond with harmonised voices. If The Hearth’s fans eased you into the shows with their innocent whirring, The Oratory reveals the potential of these models as a way to make music – if you’re brave enough to let it.
Some years ago I listened to an episode of Herndon and Dryhurst’s Interdependence podcast with K Allado-McDowell about their novel Pharmako-AI, which was co-written with OpenAI’s older Chat-GPT3 large language model. I was surprised that the podcast didn’t go into greater depth about the use of language in the book or its editing process – or why one should even care for what a machine has said. In my view, the fundamental task of a writer is to pair their imagination and technical skills to persuade or seduce the reader into reading. To assume and not consider whether a writer, even a writer that’s a machine, should be read reveals an infatuation before the turn of a page. It’s a thought I had walking around The Call, and as I watched a screen where Dryhurst says AI models ‘want’ to be interacted with and experienced. I fear a god has been created where an artist should be heard.