>> Eventually, it produces (give or take a few bits) a copy of an MP3 file of a copyrighted work.
...which you recognize by having a copy of that work and specifically matching for it. If I were a copyright lawyer, I'd argue that your algorithm for plucking this value out of the stream of randomness was the infringement.
>> I could quite easily create every possible variation of an MP3 file of a given length.
If you're prepared to pay $35 each to register the copyright on all of those, knock yourself out. I'll enjoy not paying taxes anymore.
> ...which you recognize by having a copy of that work and specifically matching for it. If I were a copyright lawyer, I'd argue that your algorithm for plucking this value out of the stream of randomness was the infringement.
I don't have to recognize it myself. Say I put all the resulting files up for download on an FTP server, and the RIAA stumble across the collection. Within, say, a collection of every possible 30 second long MP3 file encoded at 128kbps, I'd probably be infringing on a few thousand copyrighted works.
For each infringement there'd be many, many more 'infringing' files (i.e. every slight variation on a work that a copyright lawyer would deem indistinguishable from the original work)
> If you're prepared to pay $35 each to register the copyright on all of those, knock yourself out. I'll enjoy not paying taxes anymore.
Apparently you can register copyright for music tracks in bulk. In any case, where I live you don't have to register copyrights.
>> Say I put all the resulting files up for download on an FTP server
128kilobits per second * 30 seconds = (128 * 1000 * 30) = 3,840,000 bits per file.
There are 2 to the 3,840,000 possible combinations of that many bits. Ignoring the fact that many of those won't be valid mp3s, each of those is about a 0.46 megabyte file.
I'm guessing you don't have enough hard drive space to put all those mp3s up. :)
Assuming you did, the RIAA would have a tough time crawling all that content for infringement.
It would make an interesting test for the theory that "linking isn't infringing," since the link would be the only thing distinguishing a song from random noise.
Obviously I'd set the random generator up such that it operates within the rules of the mp3 specification and only creates valid mp3 files; I don't think that detracts from the experiment.
Storage space is the only major limitation here. With current computing power I could easily have random mp3 files spat out at an alarming rate, such that it wouldn't take too long (I'm guessing a matter of months) until I managed to produce an infringing file this way.
I could probably speed the process up by teaching the 'random' mp3 generator certain patterns to pursue; fade-ins and fade-outs, repetition, etc. Again, I don't think these detract from the substance of the experiment.
It's kind of like teaching someone to play a sport; you show them the rules of the game, and a bunch of 'patterns' that players tend to adhere to. Eventually, they'll make a sequence of movements, lasting 30 seconds or so, near enough identical to that performed by a famous sports star.
According to Ross Millikan over there, the number of possible 3,840,000-bit files is a number with more than a million zeros. The number of atoms on the universe is only around 10 to the 80. So if you could use the entire universe as your hard drive, storing a bit on every atom, you'd need many, many universes to store those files.
You're going to have to use some serious algorithmic bias to get mp3s, much more bias to get non-static, much more to get anything resembling music and containing any English words, etc etc.
Bluntly, you won't get copyrighted works ever unless you're specifically targeting them, for any reasonable value of ever. It's theoretically possible only in the sense that it's possible for someone's DNA to spontaneously appear at a crime scene.
This is why the "songs are just numbers" argument is misguided. Yes, they can be represented as numbers. But you'd never discover them that way.
All you've done here is take the 'hidden in pi' argument and customized it. If you are only making a sample of random mp3s you have a statistically zero chance of infringing on anything. If you make a thorough set of X-length mp3s then the actual infringement is in the url, because a datastore that has every number is equivalent to one that has no numbers--it's really just an encryption algorithm between url data and mp3 data.
...which you recognize by having a copy of that work and specifically matching for it. If I were a copyright lawyer, I'd argue that your algorithm for plucking this value out of the stream of randomness was the infringement.
>> I could quite easily create every possible variation of an MP3 file of a given length.
If you're prepared to pay $35 each to register the copyright on all of those, knock yourself out. I'll enjoy not paying taxes anymore.