Will AI in Music Replace the Composer and Improvisor

There is no question that artificial intelligence (AI) is becoming a reality. We know that there have already been attempts to build computers that can ‘replicate’ the process of composition. The questions are, are they, and will they ever be any good?

We know from the history of computing, that once standard operating systems become readily available and usable, then a host of new applications follow very quickly. When a similar level of hardware building blocks is available, then robotics will go from being as expensive as the early software was, to be cheap and readily available. Simple automation of manual tasks on production lines have begun the acceleration of robots into other aspects of everyday life.

The next stage of this inevitable process is for the application of known data to create the ability of software to ‘think’ for itself. Again, this starts with relatively simple applications like autonomous cars (although the engineers currently have to programme their machines to avoid non-autonomous vehicles and bicycle driven and ridden by humans with the less than predictable responses). With the volume of ‘known’ data expanding at a huge rate and doubling every three years, we are fast building enough knowledge for the new supercomputers to assimilate and learn from it. ‘Big’ data, as it is known, is now helping retailers to gain unprecedented insight into the buying preferences of their customers, and Police departments across the world are using data to predict where and when crimes will occur with ever-increasing accuracy.

In musical terms, it doesn’t take too much imagination for someone to be able to feed all the known music on the planet into a neural network and then let it work something out. Neural networks are systems that are designed to mimic the same fundamental operating processes of the human brain; they have to learn. To do this takes time as all the eventualities (both right and wrong) have to be arrived at – think of a child when they first play around indiscriminately on a piano keyboard. At this point, there is a potential problem. Leave the child alone, and the chances are that he or she will never create anything of any meaning. But a little guidance and the license to experiment within set rules, and we just might eventually find the next great composer. The neural network, will, of course, work a little harder and faster at the task than the child at the keyboard, but will the results ever be the same? Will the program’s ability to compose be as lacking in human expression as that of a child’s?

Is composition little more than the assimilation of everything we have heard before? Does the neural network understand that greatest works of Beethoven were written when he was deaf and in the process of withdrawing from society? Will the neural network ‘know’ that Chopin was a sickly person who was both intelligent and humorous, a musical innovator and financial spendthrift, a tender lover and passionately loyal friend? The software could have had this information given to it as part of the big ‘data’ set-up, but would it truly understand the word ‘tender’?

Unfortunately, the world at large doesn’t know these things either, and so musical composition is taken at face value; of the hundreds of thousands of Western classical compositions over the past three hundred years, there is only a very small proportion which is either remembered or performed. When AI begins to create its new music, will it be able to operate in isolation and create genuinely original compositions? The greatest musicians composed without the benefit of all the other works to hand and were always stretching the rules, and they were often not fully appreciated until after the event. It seems there could be a distinct chance that we will end up being forced to listen to ‘music’ that has no genuine artistic value, but which sounds right.

In July 2012 there was a performance by the London Symphony Orchestra of a new piece called Transits – Into an Abyss. Composed by a machine built in Spain called Iamus and in the classical modernist style, it received reviews of varied praise, from ‘artistic and delightful‘ to ‘as soon as you see the title…., you know it’s going to be challenging‘. But simply because of the style it was written in, you know that the result is not the summary of three hundred years of classical music tradition. It is just a mathematical evolution of ideas that follow baseline rules input by humans. Is this composition? Isn’t this what Arnold Schoenberg and Alban Berg et al. from the Second Viennese School so spectacularly failed to convince the world was the future, with their serial music? But, as Schoenberg did famously say, ‘If it is art, it is not for all, and if it is for all, it is not art’, then perhaps I’m being too fussy.

Using a different approach, Google has created a programme called A. I. DUET. This new software, they claim, is unlike other compositional programs because they have simply fed in lots of music and it has ‘learned’ how to respond to different tunes. So you play a phrase, and the software will attempt to complete it in a similar style. However, unlike any of the great piano extemporisers, like Liszt for example, it only appears to do so in a very simplistic way. The project designers claim that they didn’t specifically program the rules of rhythm or tonality so, by definition, the software won’t know to improvise either one or both of them to surprise the user; it’s simply a mimic which won’t always, apparently, get it right. The truth is that although most people will think it clever, it is unlikely ever to be considered to be creative ‘art’.

The question here is whether a computer will ever be able to genuinely ‘compose’ music, or whether it will only ever be able to arrange notes in an acceptably pleasing way? The second of these will be relatively easy, and the resulting music will be as bland in comparison to the real thing as fast-food is to the work of a master chef. The contention is that without experiencing the characteristics, events, and everyday situations that make up our very existence as human beings, such as emotions, aspirations, conflictions and death, a mere ‘program’ cannot ever create something new which exposes this distinctly human perspective. It won’t ever suffer/enjoy the human condition.

Categorised in: Blog