Where's The Soul? Discussing Music Curation In The Age Of AI

24/04/2017 17:18

Artificial Intelligence (AI) is developing quickly. Research into AI in the field of music is broadly regarded as being in its infancy but the foundations were of course laid years ago.

2017-04-20-1492693152-9718323-tallinnpic_z.jpg Patrik Tamm /

In 1954 Elvis Presley recorded a song with Sam Phillips at Sun Records in Memphis and started a revolution in popular music. The same year, at Princeton University, Nils Aall Barnicelli was developing genetic computing methods. Alan Turing had already published his work 'Computer Machinery and Intelligence' in 1950. And by 1956, Lejaren Hiller had developed 'The Illiac Suite for String Quartet', now recognised as being the first piece of music to be composed by a computer.

I'm in Estonia, attending Tallinn Music Week 2017. At a panel discussion, entitled 'Introducing New Music - Human Curation in the AI Age', Berk Vaher, a music critic and DJ, asks the panel: "Do you feel threatened by AI?" It's a big opening question. And It is perhaps a question we should all be asking.

A guest panellist answers with a confident "No. AI is just a point of differentiation." suggesting that AI is simply a kind of tool that we can employ to navigate masses of new musical content we encounter. But I can't help feel that this statement, coming from someone who makes a living curating playlists, may be wishful thinking!

Another panellist comments that AI could never make human-like choices because "AI has no soul", and, "AI can't throw curve balls".

With these two comments, it feels as if Veher's question is answered. Do you feel threatened by AI? Not yet - because we just don't fully understand it!

And in fairness to the guest speakers, nobody really knows what the reality of AI in the music world is going to be, how to cope with it etc. But judging by the large attendance at this particular panel event, it is becoming a matter of real and pertinent interest in the music industry.

AI is beginning to open up a range of new possibilities for music creation and curation. Genetic algorithms use the principles of evolution in order to create adaptive software that mimics biological processes, such as natural selection. The algorithms will test their own output and mutate in order to become increasingly robust and effective.

It is reasonable to assume that, when these processes are applied to the type of tests that music creators and curators are using to measure the effectiveness of their own creative choices, they could perform at least as well as human beings, and may even surpass them eventually.

Recommendations based on previous content choices, powered by machine learning, are already commonplace. In time, the recommendations may no longer be solely based on search history and taste, but could be determined by other metrics such as the emotion or mood of the listener at a particular moment in time and being aware of their context.

Throughout the discussion, indefinable qualities like "soul", "genius", and "quirkiness" are referred to as characteristics that will continue to separate us from AI. The problem with these arguments is that those definitions are unlikely to survive the rigours of scientific research and discovery.

For example, a key question when considering music and art that has been created by artificial intelligence is: how do we decide if a creation sounds, or looks, "human"?

Turing-like tests, to find ways to arrive at a statistical conclusion are one approach, but those types of methods present their own sets of problems. The original Turing test, created in 1950, was based on a Victorian parlour game called 'The Imitation Game'. Alan Turing proposed that a machine and a human being could both be asked questions by a participant. Based on the responses given by the human and the machine, neither of which would be audible or visible to the participant and with each of their responses being passed to the participant via written notes, a decision could be reached as to which was machine or human.

In terms of music creation or the curation of a music playlist, is one person's opinion of how"human" something seems, more accurate than another person's opinion? If one person is convinced that human intelligence is present, but another isn't, what does that mean for the result? Does the ability to convince a large group of people make a result more accurate?

Essentially, if a large group of individuals can be convinced that AI is displaying the characteristics of a human being, it is not difficult to imagine that obscure qualities such as "soul", "genius", and "quirkiness" could be convincingly imitated.

As this kind of research continues, and the definitions of some of these human characteristics are becoming more clearly understood, the line between imitation and reality could become completely obscured.

I finished Tallinn Music Week by watching a performance by Polish electronic music duo Coals. At one point during the set, vocalist Katarzyna Kowalczyk repeatedly sings the line "I am a lone dancer", as if she were a machine stuck in some hypnotic infinite feedback loop.

Perhaps we are lone dancers for now, but I can't help but think that we may not be for much longer.