You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
This is a project I can get behind. Superficially it's another in a long line of "here's how we'll be obsoleted by computers". But the project is more interesting than that. Based on deep learning,Magenta aims to pattern human-created music to learn how to create new music. It also has applications to user-specific music recommendation. But researchers have a more interesting and realistic goal than the purported march to human obsolescence. They want to explore the boundaries of machine generated creativity - what can it do, what can it not do and why not? Everything has boundaries - finding those boundaries is what drives science and art. If it doesn't have boundaries, it's religion.
Interesting project.Both with regards to technical realm(very interesting) , but it may also break the wrong , but common belief that computers aren't creative(at the very least in the tech domains, but probably also in art there was some work), since it would come with a lot of PR.
Ippisl - agreed computers can create. The tricky part is whether the audience (us) will like what they create. Tastes in music and art seem quite difficult to predict. Could AI do better? I really don't know. You could imagine learning patterns from very popular music that might be reproducible, but would that just provide more of the same, which might become boring. Off topic of course - the question is whether computers can create interesting music, not whether it will remain interesting. But I'm curious whether it would always be derivative or whether it could ever be truly new and at the same time popular?
>> But I'm curious whether it would always be derivative or whether it could ever be truly new and at the same time popular?
First let's ask the question: How much of human created music is truly new ? some will say it's extremely small - pink floyd inventing psychedelic music, the inventor of rap, etc - and than it's all variation on a theme and using basic qualities humans enjoy. It's probably a bit extreme but very true.
As for creating something totally new, a new style ? i think part of that comes down to new tools - the electric guitar(rock), musical effects(psychedelic/rock) and synthesizers etc (electronic music) all created new styles. So in some abstract sense they aren't that new - for example they use same chord progressions, and other similar motifs and mix them(and mixing in a way to create new stuff, sometimes).
But still you have to have "the feel" for good music - a model of what a human being will like and what he won't like. And computers are getting pretty good at building models and in perception. So getting "the feel" sounds like an achievable challenge, but surely an interesting one.
But IDK, i'm an AI optimist/traitor, so i might be wrong.
Very true - after all there are lots of human creators who fail to excite us. I'm more of an agnostic I think. I don't believe human ability is unsurpassable but I tend to feel the percentage of misfires with machine-created music would be at least as bad as for human created music unless it is very derivative. And who would listen to the misfires to find the few truly great new pieces? As you say a very interesting challenge
>> And who would listen to the misfires to find the few truly great new pieces?
There are plenty of people who do that, who like to discover new stuff. In other fields you call them "early adopters" .
Also, Google can add short segment to the millions using their music service and test the response. Since this service is relatively new, maybe this music creativity project is a result of it ?
