Could Advances in AI Technology Re-shape Music as we Know It?
Technology is revolutionising many industries, and music is no different. AI can now write complete songs. What does this technological change mean for music as a business - and as an art-form?
Music has been big business for a long time, but these days it’s not just record-companies, concert promoters and (occasionally) artists that are getting in on the action.
In July 2019 Landr, a Montreal-based start-up that has been developing AI software for music announced $26 million in series B funding. Their platform, trained on data from millions of tracks, analyses musical styles to create bespoke post-production tools. AI is routinely used by streaming platforms, like Spotify and Apple Music, to predict listening behaviour, and we are starting to see similar technology being used by a wide range of other companies in very interesting ways.
The major record labels have entered into an uneasy licensing alliance with streaming platforms in recent years, and the music business is arguably at its most settled since the massive tech-inspired disruption that began in the 1990s. But some companies are looking to turn things upside-down once more by developing software that can take production responsibilities away from major studios, assist an artist’s creative process or maybe even write music itself.
Speaking as someone who has never been able to write a song consisting of more than four chords, this is a hugely intriguing development. What if these kinds of applications could unleash my heretofore unheralded musical genius?
If the machine can provide the first-move of creative conjecture, the genesis of an idea, could I take that idea forward with my own (limited) musical ability and turn it into a hit? What if the only barrier between me and superstardom is the initial moment of involuntary creative avolition, which could be overcome by this wonderful technological advancement?
Australian start-up Popgun has been working on an AI called ‘Alice’ that can improvise on piano, bass and drums, compose original backing tracks for vocalists and even sing. Companies like AIVA, which composes tracks for film and gaming, and Mubert, which has launched an app that produces a stream of AI music that over time tunes to the user’s preferences, are among those developing ideas based on similar concepts.
Jukedeck, a UK based AI-generated music business was recently acquired by China’s Bytedance, the company that develops the hugely popular lip-sync video app TikTok (Bytedance continues to have some well-documented royalty disagreements with the major record labels about the licensed music used within the app; this acquisition may represent an attempt to move away from the licensed model and towards royalty-free, AI composed content). But what does this mean for music, both as a business and as an art-form? And just exactly how does a machine write a song?
Artificial Intelligence can refer to any computer function that attempts to replicate a form of human behaviour. AI algorithms are very good at matching patterns in a well-defined space, with set rules, which makes them very good at Rubik’s cubes and games like chess. This is ‘weak AI’; technology that is suitable for narrow tasks with clear, well-established goals. Programs capable of these kinds of activities have been with us for decades — the Deep Blue chess computer beat the reigning world champion, Garry Kasparov, back in 1996.
Music is, at its essence, a series of sequential notes played at a particular rhythm; on one level, this should be something that an AI can handle. Any activity that requires advanced creativity, however, like writing music, is better understood as ‘strong AI’, that is, a more general type of intelligence that is comparable to human thought.
‘Strong AI’ is still effectively in the realm of science fiction. There is no algorithm that can replicate the complexity of the conscious mind. However, a subset of AI, called machine learning, can make use of vast data-sets to learn patterns by itself. When a machine is capable of learning new information, it can spot patterns in data that humans are cognitively incapable of processing. It creates its own models for recognising and categorising this data, which a human observer can neither understand nor perceive. This means that an ‘unsupervised’ machine learning algorithm can create truly original and unique outputs, independent of human involvement. This may not correspond to human-level creativity, but it could be a real example of a non-trivial novelty (one of the key indicators of creative thought).
Machine learning has become much more practical for programmers as data storage has become cheaper and more data is being collected. This gives developers access to enormous libraries of music, effectively enabling them to teach an algorithm by feeding it millions of songs. The algorithm can then create its own interpretation of the music. What the algorithm cannot do is decide whether the interpretation is any good or not; this requires human intervention at some level.
This technique is more suitable for some genres of music than others; jazz, for example, is particularly difficult to replicate, as Popgun acknowledge (and as Dadabots demonstrated when they trained a neural network with a John Coltrane record to create what might be the most terrifying music imaginable). Simple MIDI files are much easier for machine learning algorithms to parse and understand, as Abraham Khan brilliantly illustrates when he and his team used neural networks to generate Pokémon game music, while Max Frenzl has used AI-generated sounds to create what he calls ‘Neuralfunk’, a sub-genre of Drum & Bass.
Is this, then, the end of creativity as we know it? Will songs, books, poems and films be written from now on by machines, designed and optimised by algorithms to keep us placated and content? If that sounds sterile and dystopian to you, I’d be inclined to agree. But I don’t think that’s going to happen any time soon. For a start, we’re a long, long way away from developing an Artificial General Intelligence capable of this kind of creativity. Don’t let the alarmists stress you out — the replicants and cyborgs won’t be coming for us, or our songwriting workshops, any time soon.
A machine that writes songs simply by listening to other songs is merely re-organising and repeating what is already there. It has no way of knowing whether its output is valuable or not. The only way to prevent an AI from spewing out randomly generated garbage is by training it and telling it exactly what you’re looking for (this is a form of ‘supervised learning’), or else sifting through the resulting mass in an attempt to find a pearl of value, a task that sounds so unappealing it might just put you off music forever.
What we do have at the moment is a collection of ever-improving tools that can allow us to augment and facilitate our own creativity. Popgun says they see their software as just like another instrument, much like the synthesiser in the ’80s, when fears abounded that it would take jobs away from trained musicians and ruin pop music forever. AI backing tracks could expedite the often exhausting and repetitive nature of the songwriting process. Machine learning algorithms are capable of generating genuinely original content that could act as the kernel of inspiration; the launch-pad for a budding songwriter to apply their skills and knowledge and create a valuable piece of work.
I’m still holding out hope that someday technology can augment my own creativity enough that I can fulfil my (imagined) musical potential. I think this kind of augmented creativity can encourage people into creative areas who may not have had the confidence to do so in the past. People with strong technical skills, for example, who never thought of themselves as creative, could use technology like this to explore avenues that they never thought possible. Combining the analytical and number-crunching power of AI with human creativity could totally revolutionise the industry.
Have you ever considered using technology to augment your own creativity? Are you a musician or an artist who could use software like this to supercharge your ability to innovate? Or maybe you never considered yourself creative at all, and have decided to leave the creating to the creatives. It should help to realise now that everyone is creative at some level, but some people might need a little more help to bring their ideas into the world.
All views are my own and not shared by Oracle. Please feel free to connect with me on LinkedIn
