The new issue of the German music journal Positionen – titled Unberechenbar, meaning both ‘incalculable' and ‘unpredictable' – carries in its masthead the symbol of two hands: a declaration that the magazine is ‘made by people, not by AI'.
In his editorial, Julian Rieken traces the mark back to the oldest evidence of human creativity: hand stencils in the Cueva de las Manos, a gorge in Argentine Patagonia, some up to 13,000 years old. More than 800 survive, mostly left hands, since the right hand held the bone that blew ochre pigment across the rock.
It is one of the small ironies of the present, writes Riecken, that hands are exactly where image generators fail most reliably. A hand has 27 bones and 34 muscles; AI, working by statistical pattern alone, has no sense of why it can move as it does.
Rieken traces the gap between imitation and understanding through art history, from flat medieval hands to Leonardo's dissections and Masahiro Mori's concept of the uncanny valley. Behind the glitch, he adds, is a story of exploitative labour: the content moderators in Nairobi, Manila or Bogotá who sort AI's training data for very little pay.
The failure, Rieken suggests, is perhaps less a bug than a reminder that ‘creativity is more than the statistical repetition of images'. The craft of writing, writes Rieken, with ‘all its detours, doubts and quirks', is not a flaw to smooth away but the precondition of any criticism worth reading – a small, deliberate refusal of ‘the logic of mere efficiency'.
Between hysteria and dystopia
Irene Lehmann starts her essay with an advertising poster at a railway station, on which a vacuum-cleaner brand promises that AI can remove invisible stains. From there she turns to the productivity paradox facing companies that have poured money into the technology. If staff spend their days sorting ‘slop' out of AI-generated text, what's the point?
Lehmann is sceptical of the mythology surrounding AI – the term coined by computer scientist John McCarthy in 1955 to make a funding pitch sound more exciting. Joseph Weizenbaum's 1964 chatbot ELIZA – named after the protagonist of G.B. Shaw's take on the Pygmalion myth – already showed how readily people form attachments to unfeeling software.
Citing error rates of around 30% in general-purpose language models and an MIT study in which students writing essays with generative AI showed markedly less brain activity, Lehmann argues that the usefulness of AI is itself a myth.
The hope is that AI will turn us into technological centaurs, giving us superpowers as hybrid beings. Instead, drawing on science-fiction writer Cory Doctorow, Lehmann argues we risk becoming ‘reverse centaurs', a sliver of human perception harnessed to ends beyond our control. Her own prescription is neither doom nor hype, but ‘a pragmatic, sober view', steering between ‘hysteria' and ‘dystopia'.
A manifesto for staying strange
Ken Ueno frames his manifesto around a piece of film history: the benshi, live narrators of Japanese silent cinema, made obsolete overnight by the ‘talkies'. Akira Kurosawa's brother Heigo was a benshi who led the doomed strike against that change; later, he took his own life. Ueno uses the story to skewer the current debate around AI in music education, which he finds strangely narrow: composition students worry about being undercut by AI-generated music, but few ask what creation even means once a style can be modelled.
His central move is to stop examining what AI produces and look instead at its datasets, since its failures are diagnostic. He cites a project by the Hong Kong-born musician Chun Chau, which used AI to set Cantonese texts to music. Chun repeatedly showed that the technology mangled the tones of this famously tonal language – exposing not a technical glitch but the politics of the archive itself: which languages and histories are legible to a model, and which are sidelined.
Ueno borrows the idea of ‘opacity' from the Martinican philosopher Édouard Glissant: the right to exist without being fully translated or classified by outside systems. His own performance work aims at exactly that. In one piece, he submerges his face in water and holds his breath for two minutes, activating a shared bodily unease he calls ‘physiovalence', something that no dataset could contain. Art's task, Ueno concludes, is not to compete with AI but to begin ‘where legibility ends'.
The sound of the cloud
Julian Rieken also contributes with a reflection on the soundscape of AI, which he insists has a ‘sound'. Although the industry likes to sell AI as weightless, frictionless, as a cloud, it is in reality a low industrial scream of turbines, chillers and fans that never stops.
More than 3,000 data centres already operating in the US, another 1,500-plus under construction, and almost 40% of American households now within eight kilometres of one. Meta is building a facility in Louisiana big enough to cover half of Manhattan; Google is putting $6 billion into India and another £1 billion into a data centre north of London, whose smaller cousin in Essex is expected to emit as much CO2 as 500 short-haul flights a week.
Beneath the audible roar, Rieken traces a second, unheard layer: infrasound leaking from cooling halls into nearby bedrooms, which residents link to headaches, nausea and insomnia. Borrowing the composer R. Murray Schafer's idea of ‘sound imperialism' – the right to be loud always belonging to the powerful – Rieken suggests that AI's true signature may turn out to be acoustic rather than visual: a planetary hum that future historians read as a record of how hard, and how violently, we made machines think.
Review by Cadenza Academic Translations
