When they met as choir boys, aged eight, Patrick Stobbs and Ed Newton-Rex didn’t imagine that their friendship would one day lead to the creation of a song-writing robot.
Ten years later, the co-founders of Jukedeck, which provides artificial intelligence (AI) that can compose and adapt music, reunited as students at Cambridge University. Their enterprise began with Newton-Rex’s paper on the science of music – what it is that humans like about it and why. He was, says Stobbs, “getting into the chemical reactions that take place in the brain and the devices composers use to grip the listener”.
Then, while visiting his girlfriend at Harvard, Newton-Rex gatecrashed a computer science lecture. “Something clicked and he thought, ‘OK, computers are fairly good at calculating rules and probability, doesn’t music composition work in a similar way; isn’t it just essentially a set of rules and probabilities and therefore is it possible for a computer to write music?’” says Stobbs. It was on the plane home that Newton-Rex started writing Jukedeck’s first algorithm.
After securing £125,000 from Cambridge University, Newton-Rex spent several months locked away in his room developing the idea. Stobbs was working at Google when Newton-Rex eventually showed him the programme and encouraged him to leave the tech giant to co-found the company in 2014. While Jukedeck now face competition from a number of startups, including New York-based Amper Music and Brisbane-based PopGun, using AI to create music for commercial purposes was new at the time. “There were groups of academics around the world that were doing little bits of what we’re doing […] But no one was building a commercial application.”
Today the company has customers in 169 countries, who have created more than 500,000 tracks. It’s predominantly used by businesses – including Coca-Cola, Google and the Natural History Museum – that need background music for online video, but composers have also been known to use the tool to extend music they’ve written for clients.
Jukedeck writes music using artificial neural networks – computer systems inspired by the neural networks of the human brain. The team feeds these networks musical data to analyse. “We then build up that digital understanding, or representation, of a composer’s brain and use that to create original tracks of music,” says Stobbs.
Customers can set parameters for their music, including the genre, instruments, length of track and its speed. Within around 20 seconds the system composes the track, Jukedeck produces it and it is given to the customer as an MP3. Users can download it to use, or, if they don’t like it, create another. Individuals or businesses of fewer than 10 people pay $1 (£0.78) per track, but if they credit Jukedeck, they can use the track for free. Larger businesses pay around $22 (£17.20) per track or can sign up to a monthly subscription model. The service is both quicker and cheaper than using a traditional, human composer.
“What we’re really offering is the ability to customise music on the spot,” Stobbs adds. “The first thing we did is look to provide a source of custom music to amateur video creators.”
The involvement of popular YouTubers, including LEMMiNO, MarcianoPhone and Today I Found Out, has helped the company grow. “These people are inherently social, they make content and share it with their friends. They add soundtracks to their videos and then mention us in the credits … we got hundreds of thousands of hits overnight,” Stobbs says.
In 2015, Jukedeck accepted a £2m investment from Cambridge Innovation Capital, and has since won competitions including TechCrunch Disrupt London in 2015 and an Innovation Lion award from the Cannes Lions International Festival of Creativity in 2016. The London-based team now numbers 19 and most of the recruits have advanced musical training – a fact Stobbs says, sets them apart from their competitors. With DJs, producers, instrumentalists and singers on staff, the work of prototyping ideas into code is often followed by regular jam sessions in Jukedeck’s office basement.
But with worldwide fears about the impact AI will have on jobs, has Jukedeck faced pushback from musicians? “There are certainly people who hear that AI is going to be writing music and painting pictures and are suspicious of that,” Stobbs says. “Interestingly, when people start to really think about what [our tech] makes possible, they warm to it and engage with what we’re doing, particularly musicians. Because really what this is, is a tool with which music can be created more easily.”
Stobbs doesn’t think AI will disrupt the creative arts as much as it will other sectors, partly because AI can’t stand in for musical stars’ fame or performing talents. “You go to a Justin Bieber concert not just to hear the notes but because you follow his story and you go to see him visually. You engage with the full context of Justin Bieber, not just the music that he sings.
“One of the reasons people are interested in this problem [of producing music swiftly and inexpensively using AI] is that it gets to the heart of something that until recently people weren’t really thinking about in terms of machine learning, which is creativity,” he adds. “There’s been a realisation that not only can AI be useful in fields such as finance and transportation, but it can enable humans to be more creative and in different ways. And that’s what we’re really focused on.”
Sign up to become a member of the Guardian Small Business Network here for more advice, insight and best practice direct to your inbox.