Classical music has survived for centuries. Will AI kill it?

{"text":[[{"start":7.35,"text":"In Mozart’s day, mechanical instruments were all the rage. Self-playing instruments weren’t new — the first musical automata had been made centuries before. But innovations in clockwork mechanisms made Enlightenment automata something extraordinary. The novelty of seeing an instrument play without human input was so thrilling that clockwork organs became a must-have item for Europe’s elite. Mozart himself composed pieces for self-playing instruments, as did both Beethoven and Haydn."}],[{"start":38.05,"text":"There is a rich history of classical musicians working with the latest technologies to expand what music can be. Today, it’s artificial intelligence — though even here, composers have been experimenting for decades. Take Voyager, an artificially intelligent system designed to improvise live with humans, built in the 1980s by award-winning composer George E Lewis. When you see it in action, an unattended keyboard plays alongside the musicians, giving the same kind of visual novelty that fascinated 18th-century audiences. Unlike the mechanical organ, though, Voyager isn’t playing something a composer has written. It’s creating. Lewis’s systems are trained to “make decisions”, he tells me. “You’re in dialogue with them and you don’t have control over what they do.” The style is experimental and atonal, sometimes bordering on free jazz, but no two performances with Voyager sound the same."}],[{"start":91.9,"text":"Lewis has thought deeply about his motivations for working with AI, about the data he uses and how to centre human creativity. But his improvising systems are very different from large language models such as ChatGPT, the form of generative AI making headlines now. Tools that generate vast quantities of artistic material from a text-based prompt are reshaping the creative industries, with companies announcing new models with grandiose claims about “democratising” creative work before being valued at millions or billions of dollars."}],[{"start":null,"text":""}],[{"start":123.30000000000001,"text":"First DALL-E democratised art in 2021, then ChatGPT democratised writing in 2022, and in 2026 OpenAI is valued at $852bn while Unesco warns that the impact of generative AI is projected to result in creators losing up to 24 per cent of their revenue by 2028. Now music-generating models such as Suno, Udio and Stability AI are threatening to democratise music, too. It’s only a matter of time before this technology will force classical musicians to rethink what they do and why — just as it has for writers and artists."}],[{"start":161.75,"text":"Among those already working with GenAI, there’s an optimism about what it can offer classical music. Vocalist Harry Yeff, who performs as Reeps100, has used audio-generating AI systems for the last 10 years trained on his own voice, to push him to the limits of his vocal ability. He finds a self-trained AI system’s ability to go just beyond the human makes it an inspirational oppositional partner. "}],[{"start":null,"text":""}],[{"start":188.55,"text":"Yeff’s performances with what he has dubbed his “AI twin” are virtuoso showcases. He vocalises and beatboxes with and against the system, which transforms his voice into an uncanny, mechanised version that sounds both like and unlike Yeff himself. “Opponents can be our greatest collaborators. I think there’s a different dynamic of learning and human potential that comes out when you’re in competition,” he explains."}],[{"start":213.9,"text":"Artist Stephanie Dinkins is an example of the ways LLMs specifically might change who can become a musician. The visual artist’s explorations of data bias — and how to correct it — saw her recognised as one of Time magazine’s 100 most influential people in AI in 2023. In June, she will present the “voice-led performance and installation” Cosmologies From the Bay at the Royal Ballet and Opera’s four-day SHIFT festival in London, exploring the use of AI in classical music — alongside pieces by Lewis and Yeff."}],[{"start":null,"text":""}],[{"start":248.35000000000002,"text":"But this is Dinkins’s “very first experience writing music”. She composed using ChatGPT and consulting with singer Samantha Rose Williams, who will perform the piece. “It makes my artistic toolbox so much larger,” Dinkins says. “To me that is amazing.”"}],[{"start":266.3,"text":"Even so, Dinkins regards this extended capacity with trepidation. “I love that I can do it, but I also feel like I shouldn’t be able to do it,” she admits. If it’s possible to generate even the kind of complex music that has traditionally needed a great deal of musical training and skills, then humans are at risk of losing those same skills. Why teach what you can generate?"}],[{"start":291.6,"text":"“People are very conscious of how this may impact how society views the skills of a composer. Are they valued?” says Daniel Lewis, classical relationship manager at PRS for Music, the UK group that licences music usage and collects royalties. “We have to be very careful and conscious about the avoidance of skill death,” argues Yeff. “Augmentation is creation, automation is replacement.”"}],[{"start":317.6,"text":"Because LLMs can generate so much music so quickly, they risk creating unfair competition with human creators. Certain kinds of classical music will probably be impacted more directly than others. Daniel Lewis highlights that the expectations around pieces intended for the concert hall (like symphonies) mean they might be “more insulated” than genres where “listening is on a background level” — predominantly film, TV, game scores and advertising and ambient music."}],[{"start":346.8,"text":"But besides being deeply concerning for musicians employed in these genres, the knock-on effects of an AI-saturated media industry are hard to predict. Media music genres (especially film) bring a large proportion of new audiences to classical music and are a substantial source of employment for musicians. Even if concert composition is immediately insulated from automation, the infrastructure supporting it relies on those same genres most at risk."}],[{"start":373.15000000000003,"text":"This new world can feel especially galling to musicians given that it’s become clear that copyrighted music has been used to train various different LLMs. “It starts to, I think, feel violating for people,” says Daniel Lewis, to realise that your work has been used without permission to generate material that will compete with yours. PRS’s own research found that 93 per cent of their members want creators to be compensated if their music is used for AI training, and 92 per cent want more transparency around training data. The organisation is encouraging the government to place legal obligations on AI developers to disclose copyright works used along the AI supply chain, and to label AI-generated content."}],[{"start":null,"text":""}],[{"start":418.45000000000005,"text":"In the meantime, many musicians are confident that the ability to easily generate music will make audiences value the experience of live performance more. “I hope it will only be helpful for the live arts in that sense,” says director Robin Coops. His In the Ring — which will be previewed at the RBO festival — uses GenAI to place audiences within a role-playing game based on Wagner’s Ring cycle. Think Dungeons & Dragons meets opera. Coops’ “relationship with technology has always been love/hate”, he says, and partly he uses AI to get audiences to “question the technology”."}],[{"start":455.20000000000005,"text":"Perhaps there’s a world in which LLMs are the positive musical revolution that they’re being trumpeted as. But that will take careful regulation and a willingness to question the most quixotic claims about their potential to make music available to all. “If you wanted to democratise music-making,” Daniel Lewis comments, “there’s this amazing thing I once heard of called ‘state education’ and giving kids flutes.” It might not be as novel — but you can buy an awful lot of flutes for the $250mn raised in Suno’s last funding round."}],[{"start":490.25000000000006,"text":"June 4-7, rbo.org.uk"}],[{"start":495.05000000000007,"text":"Find out about our latest stories first — follow FT Weekend on Instagram, Bluesky and X, and sign up to receive the FT Weekend newsletter every Saturday morning"}],[{"start":511.90000000000003,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1779524216_9963.mp3"}