Songs like Nirvana’s, novels like Kerouac’s: How AI is getting artistic

Kurt Cobain would probably have enjoyed it. The Nirvana frontman who died at the height of his fame in 1994, would probably have listened to an artificial intelligence program’s attempt to generate a new song based on his music. Then he’d have gone on to create a whole new song, a tune and arrangement that emerged from his life, not his back catalogue. And laughed.

But as algorithms make headlines for producing creative work — books, paintings, conversations, scripts — humans are wondering if AI-generated art can eventually be more than a party trick.

Most AI processes, including Google’s Magenta software that came up with the artificial Nirvana track, work on a common principle. Examples of a genre or style are fed into a machine, analysed, and new examples generated from that data. Take a look.

Bots are talking to each other: In 2014, researchers at the Cornell Creative Machines Lab programmed two chatbots to hear and understand each other and respond – essentially have a conversation in English. They started with pleasantries. But quickly, the three-minute chat turned into almost a philosophical discourse: “What is God to you?” “Not everything.” “Not everything could also be something.” And presented truly odd logic tangles: “But you say you are not helpful, therefore you are a meaning.” Be confused; be very confused.

AI is composing tunes: Project Magenta, ongoing since 2016, uses Google’s open-source TensorFlow AI machine-learning engine for artistic and creative experiments. The Magenta team released their first piece of music (composed without human intervention) in 2016. It is, as Billboard noted, a 90-second four-note piano prompt, with a drum beat overlaid by humans to add harmony. It was music, but not great music. Magenta experimented with AI jokes the following year – they just weren’t funny. Then came the Nirvana song, released this month. The team has also worked on mimicking music by deceased artists such as Jimi Hendrix, Jim Morrison and Amy Winehouse.

Meanwhile at Sony Computer Science Laboratories, Paris, an AI music composer called Flow Machines reworked classic arrangements and even remixed Beethoven’s Ode to Joy to a bossa nova beat. In 2016, two tunes (with lyrics, arrangements and production contributed by a human) were released to the public. Daddy’s Car was meant to be a Beatles-inspired pop song. The Ballad of Mr Shadow was in the style of Duke Ellington. Music critics thought Daddy’s Car sounded more Beach Boys than Beatles. Mr Shadow sounds appropriately old-timey, like Dean Martin is idly humming a song you can’t quite place.

Coming soon is a whole pop album made by human-machine collaborators. The Songularity raised its funding on Kickstarter last year. The upcoming album will weave in lyrics from Amazon customer reviews of a home workout DVD, Bob Dylan’s poetry, and unfavourable reviews of Manhattan restaurants. Don’t worry; it’s meant to be funny.

Machines are making art: In 2019, Christie’s auctioned a canvas titled Portrait of Edmond Belamy. It showed a blurred male figure, in a jacket and collar, with features indistinct as if the man had just averted his gaze from the viewer. The work was created after French artists Hugo Caselles-Dupré, Pierre Fautrel and Gauthier Vernier fed thousands of portraits into an algorithm. The generated images were then curated by the artists and printed on canvas. It fetched $432,500 and was the first AI work sold at auction.

India’s first AI art show was in 2018, when New Delhi’s Nature Morte gallery invited the art curation and research collective 64/1 to exhibit works created in collaboration with AI. Gradient Descent featured artists from the US, Japan, Germany, Turkey, India, the UK and New Zealand. “We got more interest from outside India than in India,” recalls 64/1 co-founder KK Raghava, who has been using digital tech to help create art for close to 15 years. He believes AI can potentially help artists with pattern recognition and technical tasks. “Think of AI as an extension of an artist’s toolkit.”

Algorithms have written books: Well, they’ve tried. In 2017, researcher Ross Goodwin hooked a microphone, GPS, camera and a text-generating algorithm to a laptop and sent it on a road trip from New York to New Orleans. The book, 1 the Road, was touted as the first novel written by artificial intelligence, with sights and sounds (even the laptop’s clock) offering prompts for the chapters. It’s not quite Kerouac. The novel doesn’t even have a plot. It ends up repeating locations and descriptions of America. But a reviewer for the Atlantic said “there are some striking and memorable lines”. There are now so many books generated by bots and neural networks, there’s even a site tracking them all: booksby.ai.

Where does all this leave human creativity? Machine intelligence isn’t likely to replace the artist – it lacks the range of human experience necessary to create original work in the first place. Meanwhile for artists, programming has become a creative act. Feeding examples into a machine and sifting through the generated output are being seen as forms of curation.

“Instead of an individual artist as genius, it’s now a community or society in an environment of safe creation that is the genius,” Raghava says. “What sets the artist apart is the personality, and the way we each address the world. That’s where authorship of an AI work lies.”