The Future is Tomorrow: Artificial Intelligence and Music
The Future is Tomorrow
Artificial Intelligence and Music
Aside from music, I have a passion for science. I read a lot about science in my spare time from light Particle Physics to Cosmology, I have always had an avid interest in what makes the universe so mindbogglingly incomprehensible. Something that is frequently brought up is the apparent inevitable rise and exponential development of A.I. (usually buddied with concerns of a Skynet-esque apocalypse). Hopefully before A.I. supersedes the average human, we will have a period of fascinating integration where both human and A.I. can get the best out of each other.
But where does A.I. fall into music? Even more so, where does creativity fit in A.I. if at all? Some predict that we might see up to a third of the music manufactured within the next decade could eventually be written by A.I. processes. As someone who primarily writes music electronically for a living, the threat of a diminishing need for a composer - particularly those working in media or mainstream pop - is a concerning one. But should we be worried? I thought it might be interesting to explore how A.I. helps us at the moment, where we might be going with it and if we are all going to loose our jobs to robots...
Technology giants and startups alike have caught on to this promising money making machine and have began developing their own artificial intelligence. Big names like IBM’s Watson Beats, Sony and their Flow Machines, Google’s Magenta and even Spotify have had a go with their Creator Technology Research Lab. For the last few years these companies have been using mathematical algorithms to match songs to a specific individual user. By feeding one of these systems some musical stabilisers, it seeks out connections and nuances within the music such as harmonic changes, style, mood and progression.
So how does it work? Well, it’s not necessarily generating creative output but it does use algorithms generated by analysing selected music which is input into the programme. Most systems work similarly in that they use artificial networks mimicking neural connections similar to a human brain. These foundational tracks will lay the groundwork which the system will base it’s musical intelligence and programme rules off of. In turn, it provides a system that will create various textures, moods or even adaptable (and all modifiable!) accompaniments, which is derived from the patterns and relationships between the movement of each individual note. These elements can then be used to build and develop ideas within the outputted suggestions. Generally speaking, it seems likely that as technology grows and the average humans needs become more technologically based, there will be a gradual integration in biomechanics.
A second graduation in A.I. is likely, from integration in machine assisting to the actual manufacturing of music but with a physical human performance. Some systems, like the industry leading Amper, use a different method. Instead of being fed existing music for analysing - has been taught music theory, as well as what techniques trigger such emotional response. The finished product is by no means perfect and requires a humans touch to refine the A.I. created piece. This is something that may only be a temporary fixture though, as the technology grows, there is a chance that the human interaction will become more and more limited. It is within the brains limbic system where emotions become involved in the music we listen to via the neural pathways that connect our brains to our reward centre and our dopamine response. Therefore, one’s emotional response to music is a unique and individual one, a byproduct of the conditioning of your collective prior experiences. Inevitably it raised the question: Can artificial intelligence make better music than us?
Putting subjectivity aside, we have to question, what it is that makes music so appealing to begin with? Out-with the fundamental elements of music, our general definition of effective music is one that can provoke a strong emotional response. From this angle then, an A.I. based composer might actually be able to write in real-time, catering specifically to the listeners emotional requirements at that particular time of listening. What’s-more, the programmes would only become more accurate and powerful as more data can be collated. With more information being processed through A.I. and it’s independent learning, it could be possible to target emotional reactions like nostalgia or euphoria. In the future we would most definitely see this becoming infused with how we consume media and advertisement by way of directly provoking emotional reactions.
As I mentioned earlier, headline tech companies have been trying to grab hold of these technologies. However it’s also at the forefront of tech innovation and start ups - some more established than others. One such company, Jukedeck and it’s founder Ed Newton-Rex , gave a talk at Screen Music Connect in London last year which I was lucky enough to attend. (You can read my blog on it here). Newton-Rex has a brilliant grasp on the companies goals and where he plans to take Jukedeck. Naturally, people are curious to know when we can be hearing more common place musical A.I. - something Ed must be asked on the regular now is, when will we see A.I. make a number 1 hit? To which he answered,
““Predicting the future’s a fool’s game, but I can see an AI No.1 happening probably before it deserves to - before the music’s good enough - just due to the novelty factor. “But it depends on the bar you’re setting. If your bar is AI composing a pop song without people knowing it’s involved, that’s a much bigger hurdle, I think. That would involve AI composing music that’s as catchy and emotionally resonant, with lyrics as powerful, as non-AI-composed music. And each one of those things is really hard.’
”
More likely than not, at least for the time being, we shouldn’t expect to see A.I. topping the charts.At the end of the day there are so many elements to what contributes to the success of a number 1, most of which are inherently human an obstacle for AI. Factors like the current climate of society, the artist themselves, endorsements and marketing etc.
Part of the problem with using a learning technology like this, particularly at such a primitive stage, is that it is fundamentally based on prior knowledge and working with with data that has already been understood and developed. Although, it might not be focussing on the next major advancement of human technology - the level of skill that these AI machines will soon be able to operate with absolutely have the potential to shape our lives of the future.
Ultimately, it’s always been possible for a computer to write a piece of music, maybe even a piece of catchy music. But to write something which is to become a hit would certainly be a step in a new direction. I remain skeptical though, as it is the human elements in music that remind us where the true essence of creativity comes from, down to the production and even the final mix. In a lot of ways I feel like I am left asking more questions than when I started, like: What is the real purpose of music? What are the knock on effects that this can have? What becomes of the artist when a programme can directly target and provoke emotional responses instead?
A.I.’s method of learning by reinforcement, which includes lyricism can help to reduce costs when it comes to production like recording or even marketing and strategy which includes learning to gauge a listeners sentiments then tailoring the listening experience for them. We are reaching a threshold, where A.I. is becoming a mainstay within the music industry and in a format that is constantly evolving and developing, only to get more intelligent the more data that can be input.
SD
You can find links to some useful websites below:
Programmes:
IBM Watson’s Beats: https://www.ibm.com/case-studies/ibm-watson-beat
Sony Flow Machine: https://www.flow-machines.com/
Google Magenta: https://magenta.tensorflow.org/
Francois, with Spotify: https://www.francoispachet.fr/
Jukedeck: https://www.jukedeck.com/