Page 5 of 8 FirstFirst 12345678 LastLast
Results 61 to 75 of 117

Thread: AI has finished Beethoven's 10th Symphony...

  1. #61
    Senior Member Woodduck's Avatar
    Join Date
    Mar 2014
    Location
    Ashland, OR
    Posts
    18,277
    Post Thanks / Like

    Default

    Quote Originally Posted by hammeredklavier View Post
    You mean something like this actually has a price tag on it?:
    My favorite part is "the work may...last any length of time." Obviously, that's why it's titled 4'33."

  2. Likes hammeredklavier, Haydn70, BachIsBest liked this post
  3. #62
    Senior Member SanAntone's Avatar
    Join Date
    May 2020
    Location
    Tennessee
    Posts
    7,929
    Post Thanks / Like

    Default

    Quote Originally Posted by Woodduck View Post
    My favorite part is "the work may...last any length of time." Obviously, that's why it's titled 4'33."
    That score is from the revised version from 1961 in which Cage made the change in duration and number of instruments.

  4. #63
    Senior Member
    Join Date
    Sep 2018
    Posts
    3,673
    Post Thanks / Like

    Default

    Quote Originally Posted by Couchie View Post
    I think they should train the AI to compose Vivaldi concertos, and see if people can pick out which is the AI and which is an authentic Vivaldi concerto.
    Quote Originally Posted by Woodduck View Post
    Besides, who hasn't lain on his deathbed regretting that there weren't more Vivaldi concertos for soprano recorder to fill out the long days and years of retirement?
    Pardon, but I think that you are underestimating Vivaldi. Perhaps his ordinary concerti from early in his career are not so hard to emulate, but there's a kind of élan in his best ones that, I believe, would be difficult for an AI to recreate. Besides, rhythmically Vivaldi is one of the most interesting composers of the Baroque era in my opinion, together with Lully and Rameau, and some of his concertos play some unique "rhythmic tricks" that make me want to move every time I listen to them (I'm thinking for example in some moments in the first movement of the piece in the video below). I believe that these too would be a challenge for the AI to do.

    Last edited by Xisten267; Sep-27-2021 at 04:35.

  5. Likes ORigel, science liked this post
  6. #64
    Senior Member hammeredklavier's Avatar
    Join Date
    Feb 2018
    Posts
    4,949
    Post Thanks / Like
    Blog Entries
    2

    Default

    Quote Originally Posted by SanAntone View Post
    That score is from the revised version from 1961 in which Cage made the change in duration and number of instruments.
    He could have put in more effort in the revision, like
    Quote Originally Posted by hammeredklavier View Post
    The work pales (in terms of "extra-musical" complexities) in comparison with Schulhoff's In Futurum, which predates Cage's 4'33" by 33 years https://www.youtube.com/watch?v=FBfOG0D39eo

    specific instructions for hand postures to the performer:


    daringly complex use of notation:

  7. #65
    Senior Member hammeredklavier's Avatar
    Join Date
    Feb 2018
    Posts
    4,949
    Post Thanks / Like
    Blog Entries
    2

    Default

    Quote Originally Posted by Xisten267 View Post
    I believe that these too would be a challenge for the AI to do.
    I don't think AI can write this - Largo e spiccato 3:42

  8. #66
    Senior Member
    Join Date
    Apr 2021
    Posts
    1,091
    Post Thanks / Like
    Blog Entries
    1

    Default

    Quote Originally Posted by Xisten267 View Post
    Pardon, but I think that you are underestimating Vivaldi. Perhaps his ordinary concerti from early in his career are not so hard to emulate, but there's a kind of élan in his best ones that, I believe, would be difficult for an AI to recreate.
    I'd say that the achievement of Vivaldi was not concerti 101-500 but the first few dozen that set the frame/style for extrovert virtuoso solo concerti. An impressive feature would be to feed an AI with Italian music of the 1650-1700, such as Stradella, Corelli, Torelli, earliest Albinoni etc. and then to get out how Vivaldi wrote in 1711. Not generating exemplar #600 of an established style with hundreds of samples to learn from.
    Last edited by Kreisler jr; Sep-27-2021 at 08:55.

  9. #67
    Senior Member EdwardBast's Avatar
    Join Date
    Nov 2013
    Posts
    5,467
    Post Thanks / Like
    Blog Entries
    8

    Default

    Quote Originally Posted by MarkW View Post
    A hallmark of Beethoven was that he was always experimenting - hence there was little natural-seeming "progression" between one piece (or group of pieces) and the next (except over a very long time frame). I doubt if you could program an AI to do that at this time.
    This is a good point. Arguably, what makes Beethoven Beethoven is the degree to which he himself didn't sound like Beethoven from one work to the next, meaning that successive works in a genre often break any mold established by their predecessors, that none could have been predicted or anticipated on the basis of past work. Line up all of the piano sonatas from the Op. 27s to the last (excluding the "sonatinas"), for example, and each is a significant departure from all of the others. When defying expectations is a central stylistic trait, AIs which necessarily rely on simplistic extrapolation from existing works are at a disadvantage.

    The other obvious problem for AIs is that a central feature of Beethoven's mature style is its organization into large paragraphs and dramatic units whose unity is beyond the scope of mechanistic algorithms.

    Your frogs make me shudder with intolerable loathing and I shall be miserable for the rest of my life remembering them.
    — Mikhail Bulgakov, The Fatal Eggs

    When a true genius appears on the earth, you may know him by this sign, that all of the dunces are in confederacy against him.
    — Jonathan Swift

  10. Likes Xisten267, Woodduck liked this post
  11. #68
    Senior Member Woodduck's Avatar
    Join Date
    Mar 2014
    Location
    Ashland, OR
    Posts
    18,277
    Post Thanks / Like

    Default

    Quote Originally Posted by EdwardBast View Post
    The other obvious problem for AIs is that a central feature of Beethoven's mature style is its organization into large paragraphs and dramatic units whose unity is beyond the scope of mechanistic algorithms.
    This seems to me the essential difficulty. Understanding the "logic" of musical structures - what makes a composition coherent and meaningful - is often difficult or impossible even for human beings, who bring the full range of perceptual, rational and emotional faculties to bear on the task. This is true for the composer as well as the listener; the fundamental creative process is intuitive, much of what emerges is not, and need never be, fully analyzed on a conscious level, and attempts to explain why a piece of music "works" are helpful only to a point. What goes on in a composer's subconscious during the act of creation is extremely complex, vanishingly subtle, and fugitive, and the more original a work is the more impossible it is to explain the how and why of it. The very idea that AI could duplicate this process or produce significant new music comparable to what human beings can create is certainly beyond my imagination. But maybe that's my failing.

  12. Likes EdwardBast liked this post
  13. #69
    Senior Member EdwardBast's Avatar
    Join Date
    Nov 2013
    Posts
    5,467
    Post Thanks / Like
    Blog Entries
    8

    Default

    Quote Originally Posted by Woodduck View Post
    This seems to me the essential difficulty. Understanding the "logic" of musical structures - what makes a composition coherent and meaningful - is often difficult or impossible even for human beings, who bring the full range of perceptual, rational and emotional faculties to bear on the task. This is true for the composer as well as the listener; the fundamental creative process is intuitive, much of what emerges is not, and need never be, fully analyzed on a conscious level, and attempts to explain why a piece of music "works" are helpful only to a point. What goes on in a composer's subconscious during the act of creation is extremely complex, vanishingly subtle, and fugitive, and the more original a work is the more impossible it is to explain the how and why of it. The very idea that AI could duplicate this process or produce significant new music comparable to what human beings can create is certainly beyond my imagination. But maybe that's my failing.
    I agree. That is the big problem. Scholars and critics have been spilling ink on this for two centuries and there is little consensus on what makes Beethoven's mature works work. That is a toxic environment for algorithms.

    Your frogs make me shudder with intolerable loathing and I shall be miserable for the rest of my life remembering them.
    — Mikhail Bulgakov, The Fatal Eggs

    When a true genius appears on the earth, you may know him by this sign, that all of the dunces are in confederacy against him.
    — Jonathan Swift

  14. Likes Woodduck liked this post
  15. #70
    Senior Member Livly_Station's Avatar
    Join Date
    Jan 2014
    Posts
    296
    Post Thanks / Like

    Default

    Quote Originally Posted by Woodduck View Post
    This seems to me the essential difficulty. Understanding the "logic" of musical structures - what makes a composition coherent and meaningful - is often difficult or impossible even for human beings, who bring the full range of perceptual, rational and emotional faculties to bear on the task. This is true for the composer as well as the listener; the fundamental creative process is intuitive, much of what emerges is not, and need never be, fully analyzed on a conscious level, and attempts to explain why a piece of music "works" are helpful only to a point. What goes on in a composer's subconscious during the act of creation is extremely complex, vanishingly subtle, and fugitive, and the more original a work is the more impossible it is to explain the how and why of it. The very idea that AI could duplicate this process or produce significant new music comparable to what human beings can create is certainly beyond my imagination. But maybe that's my failing.
    Just to give some food for thought, all you're stating here is agreeable but somewhat arbitrary, building from the premise that since we personally can't fully grasp the inner mechanisms of a particular style of music, then it can't be done. That's fine, but maybe it's just our failing, as you said. At the end of the day, it's not impossible that a super intelligent AI with a powerful hardware may understand the building blocks of that style and replicate its formulas, ideas and structures in a way that you wouldn't expect, with finesse and wisdom, and even sounding passionate despite, supposedly, the creative process being "cerebral" instead of "emotional" (more of that at the end of my post).

    Truth is that a lot of the human creative process is actually just learning/studying patterns and tropes, mastering those, and then replicating them with a slightly new organization in order to be "new" (not plagiarism) -- and we still call that Art. Some of those techniques are broader and simpler, but some are actually very subtle and crafty, but still patterns possible to learn. In many ways, the 'human way' is not so different from what an AI does it, and, believe me, they can be quite creative and have their own personality even if this sounds like science fiction.

    Sometimes, what makes a great artist is the ability to fusion different influences and styles through a minute and deliberate process in order to create his own artistic voice. Well, that's not beyond an AI, in theory and practice. Instead of making the AI learn only Bach or only Beethoven, make it learn a bunch of different artists -- including contemporary classical and pop music --, and then let it freely compose music, choosing from whatever techniques it wants from all these influences. What will come out of it? Maybe a monstrosity (but still art), or maybe something even interesting to human ears and totally original.

    Another aspect of this discussion is that we romanticize the creative process because a great artist may try to express personal feelings through pure music, brilliantly finding notes and harmonies that translate into non-musical ideas, all of that following a dramatic narrative (programatic or not) to shape the structure. There's truth to this notion, which is something that may be beyond an AI (or not) considering they're not organic life, and they don't experience reality like humans do -- that said, they're learning from us, which is relevant.

    However, the "hermeneutics of art" is a little bit more complicated, since (a) the listener is free to take away from the music his own abstract feelings that maybe were not intentional by the artist, (b) the meaning is always internal to the art regardless of external factors, so we can always try to make sense of it as long as our interpretation is loosely coherent with the object. And since there's this independence, any music can convey emotion and ideas even when they're not intended -- and most likely any musical combination will necessarily generate an emotional reaction on the listener, posite or negative, strong or soft.

    Besides, the more we let an AI self-develop and experience the world (especially through self-teaching), the more it'll probably create its own thoughts and ideas about all things, and all of these ideas can be food for its creative process, making metaphorical art just like humans do. We just can't predict if it'll be relatable.
    Last edited by Livly_Station; Sep-28-2021 at 23:33.

  16. #71
    Senior Member BachIsBest's Avatar
    Join Date
    Feb 2018
    Posts
    911
    Post Thanks / Like

    Default

    Quote Originally Posted by SanAntone View Post
    That score is from the revised version from 1961 in which Cage made the change in duration and number of instruments.
    Those edits really changed the entire complexion of the work. It sounded totally different in the unrevised version! Maybe it will eventually be like Bruckner symphonies, where performers pick and choose bits and pieces of each edition to conceive their own distinct artistic vision of how the piece should unfold.

  17. #72
    Senior Member BachIsBest's Avatar
    Join Date
    Feb 2018
    Posts
    911
    Post Thanks / Like

    Default

    Quote Originally Posted by Woodduck View Post
    This seems to me the essential difficulty. Understanding the "logic" of musical structures - what makes a composition coherent and meaningful - is often difficult or impossible even for human beings, who bring the full range of perceptual, rational and emotional faculties to bear on the task. This is true for the composer as well as the listener; the fundamental creative process is intuitive, much of what emerges is not, and need never be, fully analyzed on a conscious level, and attempts to explain why a piece of music "works" are helpful only to a point. What goes on in a composer's subconscious during the act of creation is extremely complex, vanishingly subtle, and fugitive, and the more original a work is the more impossible it is to explain the how and why of it. The very idea that AI could duplicate this process or produce significant new music comparable to what human beings can create is certainly beyond my imagination. But maybe that's my failing.
    Yes, this is a huge challenge in AI, but perhaps not quite in the way you think. With machine learning algorithms we have essentially learned to make computers think intuitively rather than precisely which is a huge step forward. Rather than figure out exactly what's going on, we can tell computers to find a really good approximation to what's going on, which is what humans do in virtually all tasks. Think, for example, of catching a baseball; you don't literally solve the physics of the ball flying through the air and put your hand exactly where the ball will be, but, rather, use your previous experience of balls flying through the air to approximate with a high degree of accuracy where the ball will fly. This is what we can now tell computers to do.

    When we want to get an AI to compose music in the style of Beethoven, it is taught based on the previous works of Beethoven and potentially other composers to get a larger sample size. In a basic setting, the computer might look at the previous bar of music and determine something that Beethoven might have written as the next bar. Obviously, this would be incoherent as the AI loses track of what happened a mere bar ago. If you want the AI to take into account more bars of music, the amount of memory this takes up on your computer will increase exponentially to the point where taking into account a large number of bars quickly becomes unfeasible.

    Computing power is increasing, but it is not increasing nearly fast enough to except any sort of resolution of this issue, so people who write AI music algorithms will have to come up with other ways to circumvent this problem. However, this main issue is why, for example, most AI produced recipes, books, or music, make a lot of sense for a couple seconds and then quickly start to feel like something is off.

    Ultimately though, whether or not AI can replicate human creative endeavors probably depends on two main philosophical questions:
    -Is there a cap on the possible computing power available (given finite amount of resources) and where this cap is?
    -Is human thought computable (reproducible by an algorithm which terminates in a finite amount of time) or can be approximated by computable algorithms?

  18. Likes MarkW, Livly_Station liked this post
  19. #73
    Senior Member hammeredklavier's Avatar
    Join Date
    Feb 2018
    Posts
    4,949
    Post Thanks / Like
    Blog Entries
    2

    Default

    Quote Originally Posted by Xisten267 View Post
    some of his concertos play some unique "rhythmic tricks" that make me want to move every time I listen to them
    Actually, come to think of it, you have a good point. This is an area even Emanuel Bach isn't so exceptional at. Particularly his slow movements that are 8 minutes long and go on like https://www.youtube.com/watch?v=OvRBLMJeJN0
    https://www.youtube.com/watch?v=qbc39LF3Q8E

  20. #74
    Senior Member
    Join Date
    May 2020
    Posts
    969
    Post Thanks / Like

    Default

    Quote Originally Posted by Couchie View Post
    I think they should train the AI to compose Vivaldi concertos, and see if people can pick out which is the AI and which is an authentic Vivaldi concerto.
    An AI, Emmy, did compose a Vivaldi type piece.

    https://m.youtube.com/watch?v=2kuY3BrmTfQ

  21. #75
    Senior Member Livly_Station's Avatar
    Join Date
    Jan 2014
    Posts
    296
    Post Thanks / Like

    Default

    Wasn't it supposed to be release on October 9? Where is it?

Page 5 of 8 FirstFirst 12345678 LastLast

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •