I read this thing the other day, about how Sir George Martin's son Giles and a mathematician essentially re-synthesized The Beatles' only released live album, The Beatles Live at the Hollywood Bowl. It was cleaned up by modeling the spectrum of each of their voices and instruments to a spreadsheet inside of MatLab, and letting an Ai map out the music and isolate each part that was recorded on the original 3-track master tape. Each part was reversed engineered from the recorded mess. Fascinating. Brilliant Groundbreaking. Earthshaking.
You can read about the restoration process and here the musical results in the original article, here:
So I was in the shower just now, and I had a thought. As I am wont to do, in having that thought In reminded myself That As I am having this thought, a million other people are having this thought now too, and maybe one of those million are acting upon this thought...and hopefully that one person is Sir Paul McCartney or Giles Martin.
So, Ai develop by feeding them information, which they digest and process to improve themselves, and to - theoretically - better themselves, and come up with better ideas than human beings could ever come up with. For example, among many things that it can do, IBM Watson can take groups of ideas and things, and it then maps the trajectory of the progression of those things over time to come up with new, better, and similar things to a given specification. Another Ai takes physics into account and designs safer, more usable car chassis that are easier to produce than any that any human being would have ever thought of. You get the idea.
So, what if IBM Watson was fed the entire Beatles' song catalog, in order of writing, and asked to write new songs based upon their style and content?
Then I thought about the Hollywood Bowl cleanup, and realized they could model the actual Beatles delivering said new songs. Physical Modeling can recreate the sound of any instrument - or singer - singing or playing anything. Theoretically we could have an Ai write new Beatles songs, and then we could hear the sounds of the actual Beatles performing new Beatles songs that never existed - or even their future selves, aged realistically or not, performing those songs that they had never written or performed.
Then I realized they could do this with anybody's material, and they could have an Ai manifest songs that are creative or psychedelic, as well as songs that are calculated to be hits based upon a pile of songs that were already hits, or whatever that computer's model of The Beatles "wants" to come up with. In fact, you could map the trajectory of those hits over the time in which they were written, evolve what their previously-non-existent hits would have sounded like over the intervening time between 1970 and now, and come up with, in just this example, Beatles songs that would be new hits today - as if the Beatles had never stopped writing and recording (or Pink Floyd or Led Zeppelin or or or)....or anything that they possibly could have written at anytime in between then, when the last Beatles song was recorded, and now, or heck, why not the future? You could listen to that Beatles album from 1990, LOL, that never happened, or that Beatles album that might "have been" made in 2099!!! What would they call this body of work? Duh: they could call it "Evolver".
This is not science fiction. The technology to do this is actually, really, truly here right now. So, I believe that in ten years time or less we'll be doing this with pocket computers and listening to incredibly sophisticated algorithmic compositions that are perfectly matched to our musical tastes, as those songs are generated, modeled, and recorded. It's a whole new listening party, folks. You could have a tiny device in your brain that runs a tiny program that makes up an endless stream of new songs that are guaranteed to please, please you (so sorry).
The arguably bad side of all of this is that the natural extension of the controversial use of tools like Autotune and their Antares Voice toolkit voice-modeler is that anyone with access to those nee tools will be able to sing badly and make their tracks sound like any famous or imaginary singer that they want to sound like. Same with instrumental players - Roland's V-stomps did it with Guitars and Basses 20 years ago or more. But the creative possibilities of such technology are truly limitless.
Then there is the list of possibilities for realtime Karaoke processing...
The upside is that we hopefully soon won't have to ever listen to any more music that we think sucks.
Maybe Cleverbot is right: maybe I am a program. Or a towel.
...or maybe You're a towel....
