I should do this for words.
You really got to do it for letters, as you can't know what new words might develop.
Programmers generate every possible melody in MIDI to prevent lawsuits
11–20 of 340 posts
Re: Programmers generate every possible melody in MIDI to prevent lawsuits
#12Now is the time to train convolutional NN to find "catchy" melodies, and then unleash it over this dataset, and become famous songwriter.
You wouldn't use a CNN for this because there's no spatial correlation to take advantage of. Maybe an RNN somehow but the basic problem is scoring (ie how melodic a song is) not classification so honestly I don't know you'd do this using neural networks.
Re: Programmers generate every possible melody in MIDI to prevent lawsuits
#13But this will of course not work, because enumerating the bitspace does not produce bits of the right colour: https://ansuz.sooke.bc.ca/entry/23
Re: Programmers generate every possible melody in MIDI to prevent lawsuits
#14It's interesting to think about this from a more general perspective: what if we're not limited to midi, and we're talking about a more general kind of _design_ problem, where the design problem is to select a good solution to a problem within some (large) but finite and theoretically enumerable search space.
We can think of selecting a good melody to fit into a song as one example of a design problem -- searching through some finite enumerable space of melodies and selecting a melody that's a good fit. A lot of the effort in this process is testing and evaluating the melody to see if it is any good to listen to / any good in context. This effort hasn't been done if you simply brute force enumerate and list the search space without testing anything.
There are many kinds of design problems where the effort of checking if a proposed solution is any good vastly outweighs the effort of suggesting a solution.
We could do the same thing for other kinds of design problems.
In principle, what if we enumerate some (vast) finite space of digital circuits -- if we put a few limits on the amount of stuff we can put into a single circuit, and discretise any continuous parameters to produce a finite space. Does that mean we can copyright all these possible circuits, even if we put no effort into seeing if any of them are fit for any purpose?
Can we enumerate some large finite space of possible arrangements of atoms into molecules and then copyright them all, without putting any effort into analysis or experimental testing to see if any of the proposed molecules are fit for any purpose?
Snisarenko made a similar criticism when this was posted 15 days ago:
> As other commenters have pointed out, this is gimmicky, shallow and clickbaity. All that they did was count from 1 to 68 billion. Any piece of digital data can be converted to a number, and if we apply their argument then you can be "creative" by just counting numbers. But we all know that's not true. Once the search space becomes that big, you can't actually "enjoy" any of these melodies. Because you can't listen to all of them in your lifetime, the ones that you WILL hear are going to be awful 99.999 percent of the time. Hence, the "creative" process is navigating this search space, and figuring out which melodies are catchy. Better yet, trying to figure out how or why our brain decides to like or not like a melody.
-- https://news.ycombinator.com/item?id=22306857
From my perspective of framing this situation: this is dumb, as no design of any melodies has been performed. The design is the uniform prior distribution over some space of all possible melodies, most of which will be uninteresting / not fit for any particular purpose, which is a pretty uninspiring notion of design.
Re: Programmers generate every possible melody in MIDI to prevent lawsuits
#15Now is the time to train convolutional NN to find "catchy" melodies, and then unleash it over this dataset, and become famous songwriter.
You wouldn't use a CNN for this because there's no spatial correlation to take advantage of. Maybe an RNN somehow but the basic problem is scoring (ie how melodic a song is) not classification so honestly I don't know you'd do this using neural networks.
On the contrary, with a more generous reading of the previous comment, it holds some merit.
1. CNN's are used fairly commonly for sequence tasks nowadays. Convolutions can be 1D after all. 2. It's also possible the previous comment was referring to using 2D convolutions on the spectrogram of the audio, which is a common approach. 3. Neural networks are capable of more than classification. Scoring is a regression task which is common application of neural networks.
Re: Programmers generate every possible melody in MIDI to prevent lawsuits
#16But this will of course not work, because enumerating the bitspace does not produce bits of the right colour: https://ansuz.sooke.bc.ca/entry/23
A tl;dr would be nice it you post such a long article that doesn't directly seem to have anything to do with the current topic.
Re: Programmers generate every possible melody in MIDI to prevent lawsuits
#17But this will of course not work, because enumerating the bitspace does not produce bits of the right colour: https://ansuz.sooke.bc.ca/entry/23
A tl;dr would be nice it you post such a long article that doesn't directly seem to have anything to do with the current topic.
It's actually very relevant, the TL;DR (from memory) is that lawyers (and judges) aren't programmers, and the source of data (referred to as the "colour of the bits") is as important as the actual bits of the data. Thus "I have this bit pattern as part of a bit enumeration" is different, to the legal system, than "I have this bit pattern because I specifically created it".
Re: Programmers generate every possible melody in MIDI to prevent lawsuits
#18Adam Neely (professional musician and YouTuber mostly about music theory) did a video about this with the guys behind it: https://youtu.be/sfXn_ecH5Rw
Re: Programmers generate every possible melody in MIDI to prevent lawsuits
#19The idea might be to prevent lawsuits, but presumably the authors have opened themselves up to being sued by anyone who already owns a melody. If melody owners can't sue the authors then their work can't protect anyone either.
Re: Programmers generate every possible melody in MIDI to prevent lawsuits
#20But this will of course not work, because enumerating the bitspace does not produce bits of the right colour: https://ansuz.sooke.bc.ca/entry/23
A tl;dr would be nice it you post such a long article that doesn't directly seem to have anything to do with the current topic.
It's one of the more insightful things on the Internet; well worth the time to read in full. Still, a TL;DR as requested:
"Colour" is a term that encompasses provenance, chains of cause-effect, intent and related concepts. Bits are obviously colorless from compsci point of view, but are colorful from the legal POV. For instance, bits encoding a copyrighted video are just numbers, but for the judge it'll matter whether I got them from a lucky PRNG run or from a torrent site.
Not understanding that colour is real in legal systems is the core source of confusion people have wrt. intellectual property and illegal information (e.g. child porn).