Earlier in May, in Russia’s St. Petersburg, I took part in intensive two week workshop focused on using AI in musical context. We were group of about 10 musicians/artists, 2 programmers, and Peter Kirn with Natasha Fuchs overseeing and supervising the whole process.
Peter also wrote a nice article on the whole lab, which illustrates it much more holistically than I do. Here it is:
What I offer here are my two cents on the topic. Spoiler alert; AI is not here to rule the world yet.
One of the things I came to realise during the workshop was, that compared to human, AI is still very much primitive even if incredibly efficient system. This proved to be only more so, when it came to analysing audio data.
There are 44100 data snippets in one second of audio (when encoded at “standard” 44,1Khz sample rate) which still require more mathematical operations (Fourier transformation) to make this second of data sound close to what human ear perceives. Now, ideally the training data set would consist of several gigabytes of data so that computers got enough material to learn from and can create a rigid representation later on. That means, many many seconds of audio 🙂
As most of us were either good at making music or at writing code, we had to, at times, negotiate between two different worlds, trying to make sense of each other. There was one remarkable person though, who goes by as Monokeer, who was pretty good at both and ended up devising his very own AI synthesis system that sounded nothing like what I heard before:
That being said, perhaps it is not the capacity of the computer or its smartness that prevents it from spiting out awesome sounds. Rather, it is the way the human problems are presented to computer “mind” that makes the human-AI interaction quite cumbersome.
Checkout this post on how Gamma Lab AI turned into AI stage at Gamma festival later on to see how far we got collaborating. Spoiler two; we got far enough to make all the performances stand out in their own right 🙂
Other than that, here are some snippets of how it felt when we were not trying to AI.