AI Completed Beethoven's Unfinished Tenth Symphony (thenextweb.com) 64
Ahmed Elgammal, Professor and Director of the Art & AI Lab at Rutgers University, writes via The Conversation: When Ludwig von Beethoven died in 1827, he was three years removed from the completion of his Ninth Symphony, a work heralded by many as his magnum opus. He had started work on his Tenth Symphony but, due to deteriorating health, wasn't able to make much headway: All he left behind were some musical sketches. Ever since then, Beethoven fans and musicologists have puzzled and lamented over what could have been. His notes teased at some magnificent reward, albeit one that seemed forever out of reach. Now, thanks to the work of a team of music historians, musicologists, composers and computer scientists, Beethoven's vision will come to life. I presided over the artificial intelligence side of the project, leading a group of scientists at the creative AI startup Playform AI that taught a machine both Beethoven's entire body of work and his creative process. A full recording of Beethoven's 10th Symphony is set to be released on Oct. 9, 2021, the same day as the world premiere performance scheduled to take place in Bonn, Germany -- the culmination of a two-year-plus effort.
[...]
The AI side of the project -- my side -- found itself grappling with a range of challenging tasks. First, and most fundamentally, we needed to figure out how to take a short phrase, or even just a motif, and use it to develop a longer, more complicated musical structure, just as Beethoven would have done. For example, the machine had to learn how Beethoven constructed the Fifth Symphony out of a basic four-note motif. Four notes famously serve as the basis for Beethoven's Fifth Symphony. Next, because the continuation of a phrase also needs to follow a certain musical form, whether it's a scherzo, trio or fugue, the AI needed to learn Beethoven's process for developing these forms. The to-do list grew: We had to teach the AI how to take a melodic line and harmonize it. The AI needed to learn how to bridge two sections of music together. And we realized the AI had to be able to compose a coda, which is a segment that brings a section of a piece of music to its conclusion. Finally, once we had a full composition, the AI was going to have to figure out how to orchestrate it, which involves assigning different instruments for different parts. And it had to pull off these tasks in the way Beethoven might do so.
In November 2019, the team met in person again -- this time, in Bonn, at the Beethoven House Museum, where the composer was born and raised. This meeting was the litmus test for determining whether AI could complete this project. We printed musical scores that had been developed by AI and built off the sketches from Beethoven's 10th. A pianist performed in a small concert hall in the museum before a group of journalists, music scholars and Beethoven experts. We challenged the audience to determine where Beethoven's phrases ended and where the AI extrapolation began. They couldn't. The success of these tests told us we were on the right track. But these were just a couple of minutes of music. There was still much more work to do. At every point, Beethoven's genius loomed, challenging us to do better. As the project evolved, the AI did as well. Over the ensuing 18 months, we constructed and orchestrated two entire movements of more than 20 minutes apiece.
[...]
The AI side of the project -- my side -- found itself grappling with a range of challenging tasks. First, and most fundamentally, we needed to figure out how to take a short phrase, or even just a motif, and use it to develop a longer, more complicated musical structure, just as Beethoven would have done. For example, the machine had to learn how Beethoven constructed the Fifth Symphony out of a basic four-note motif. Four notes famously serve as the basis for Beethoven's Fifth Symphony. Next, because the continuation of a phrase also needs to follow a certain musical form, whether it's a scherzo, trio or fugue, the AI needed to learn Beethoven's process for developing these forms. The to-do list grew: We had to teach the AI how to take a melodic line and harmonize it. The AI needed to learn how to bridge two sections of music together. And we realized the AI had to be able to compose a coda, which is a segment that brings a section of a piece of music to its conclusion. Finally, once we had a full composition, the AI was going to have to figure out how to orchestrate it, which involves assigning different instruments for different parts. And it had to pull off these tasks in the way Beethoven might do so.
In November 2019, the team met in person again -- this time, in Bonn, at the Beethoven House Museum, where the composer was born and raised. This meeting was the litmus test for determining whether AI could complete this project. We printed musical scores that had been developed by AI and built off the sketches from Beethoven's 10th. A pianist performed in a small concert hall in the museum before a group of journalists, music scholars and Beethoven experts. We challenged the audience to determine where Beethoven's phrases ended and where the AI extrapolation began. They couldn't. The success of these tests told us we were on the right track. But these were just a couple of minutes of music. There was still much more work to do. At every point, Beethoven's genius loomed, challenging us to do better. As the project evolved, the AI did as well. Over the ensuing 18 months, we constructed and orchestrated two entire movements of more than 20 minutes apiece.
Rewritten by machine and new technology (Score:5, Funny)
Intersting test - try to create an earlier one (Score:5, Interesting)
Re: (Score:3)
Re: (Score:2)
Re: (Score:2)
I can guarantee you that test would fail, unless Beethoven's works are particularly repetitive and predicable.
Re: (Score:1)
Heh. They could create his Minus Third symphony.
Re:Intersting test - try to create an earlier one (Score:5, Insightful)
The AI never completed anything. The AI completed one possible outcome, but it is not the same as what Beethoven would have produced had he been alive longer.
If there were multiple AI's trained on these data sets, they would probably not even agree on the outcome, just like multiple humans with the same background and training as Beethoven would not produce the same symphonies, not even after having been inspired by previous works.
The notion that the AI can fill in missing gaps in information is about as ludicrous as the notion that you can compress data multiple times to get smaller and smaller results. The missing information is simply not there. All you get is something that was derived and could be completely left out and derived again as there is no actual NEW information stored in those missing parts.
Re: (Score:1)
Re: (Score:3)
Wouldn't it be more correct to say the algorithm is not filling in gaps (as you point out, that's impossible), but rather, it's extrapolating a likely result?
Interestingly, it is interpolating based on similar data in the training set. Neural networks are remarkably bad at extrapolation and often make horrible mistakes when required to do so. Their results tend to look like extrapolation to our human eyes, but that's just because we're not used to dealing with such large datasets.
Re:Intersting test - try to create an earlier one (Score:5, Insightful)
You could argue that Beethoven himself, on two different days would also not have produced the same result.
Turing Test (Score:5, Insightful)
We challenged the audience to determine where Beethoven's phrases ended and where the AI extrapolation began. They couldn't.
The AI product sounds a lot like Beethoven because it very clearly copied and pasted Beethoven ideas from previous symphonies into the new one. Anyone can make something that sounds like Beethoven in this way. What would Beethoven have done, written something like his previous works or written something new?
Here is a sample from the 'new' symphony, scroll down to the bottom of the page [thenextweb.com]. It sounds a lot like the music he was writing in the 1805 era, so I imagine they used music from that period to train the NN.
Here is a sample of music that Beethoven was writing towards the end of his life [youtube.com]. It is quite different than the reconstructed symphony they built, and anyone can hear a difference.
Finally the reconstructed symphony lacks meaning, which Beethoven surely would have imbued into his symphony had he written it.
Re:Turing Test (Score:4, Interesting)
I also want to point out that the approach is very clearly copied by the AI from Beethoven's 5th symphony, 3rd movement. Again, listen to the new '10th symphony' here (scroll down to the bottom of the page) [thenextweb.com], and then compare it to this [youtu.be].
You might argue "well, that is just how Beethoven wrote scherzos" but that is not true at all, compare the music to this scherzo which is quite different [youtu.be]. Beethoven didn't write much that was a copy of previous work, he was filled with new ideas.
Re: (Score:3)
The AI product sounds a lot like Beethoven because it very clearly copied and pasted Beethoven ideas from previous symphonies into the new one
I felt that it didn't have the excitement that Beethoven normally imbues into his work. (Although as you say, there are very clear similarities and its clearly a derivation)
But its very hard to judge when there isn't a learnt tradition of how to play or to listen to the works.
Re:Turing Test (Score:5, Interesting)
I felt that it didn't have the excitement that Beethoven normally imbues into his work.
Leonard Bernstein went through the notebooks of Beethoven, and showed that Beethoven rewrote sections if they weren't exciting enough. He had a feel for when it was exciting enough, and didn't stop until he got it. Some sections he rewrote as many as twenty times.
Listen to it here [youtube.com]. You might say the AI composition sounds like the sections that Beethoven discarded.
Re: (Score:2)
Leonard Bernstein went through the notebooks of Beethoven, and showed that Beethoven rewrote sections if they weren't exciting enough.
Beethoven was notorious for hacking his scores. That would mean literal cut-and-paste, where you chop a bit out with a knife, and paste in a new bit.
There is a great deal of an artist's work that you don't see, because they threw it away. You only see the finished perfection. You don't see the working out. It is like being presented with a great mathematical theorem, but no indication of why it is true, or what lead to it.
Re: (Score:1)
There is a great deal of an artist's work that you don't see, because they threw it away. You only see the finished perfection. You don't see the working out. It is like being presented with a great mathematical theorem, but no indication of why it is true, or what lead to it.
Then there was Mozart who had virtually no corrections or changes made once he put the music to paper. He "threw away" what he didn't like in his head before it even made it to paper.
Re: (Score:2)
Then there was Mozart who had virtually no corrections or changes made once he put the music to paper.
That is a myth spread by his widow. Modern scholarship suggests he worked out his compositions through multiple written drafts [wikipedia.org].
Re: (Score:1)
Perhaps. But surviving letters and accounts are where some this possible "myth" stems from. I suspect that it's somewhere in between. I have a hard time believing theories from people that were born a couple hundred years later. Especially since there are no recordings or other evidence either way.
With all of our modern science and technology we still haven't figured out how to make Wootz/Damascus steel, or how exactly to reproduce a Stradivari family violin. We actually have many of of each to examine.
Re: (Score:2)
Especially since there are no recordings or other evidence either way.
There is literally evidence. Turn your brain on. We have his drafts.
Re: (Score:2)
I believe J.S. Bach was the same. Different artists work in different ways.
The working drawings of artists can be interesting. They are not thrown away as such, but were never intended to be published. If I recall, Pablo Picasso had a fine classical drawing technique, but you did not see that in his paintings. The drawings of Leonardo da Vinci might actually be more important than his paintings, in terms of understanding art.
Re: (Score:2)
Perhaps his 10th would have sounded more like his previous symphonies simply because symphonies were basically made for the public consumption and therefore a composer usually didn't go completely wild with it. On the other hand string quartets or pieces made for a few instruments were made for a much more intimate crowd and a composer often used them to do strange things or experiment with weird ideas. There's doubt that he would be trying to compose a grosse fugue like piece with the 10th.
Here's the ske
Re: (Score:2)
Perhaps his 10th would have sounded more like his previous symphonies simply because symphonies were basically made for the public consumption and therefore a composer usually didn't go completely wild with it. On the other hand string quartets or pieces made for a few instruments were made for a much more intimate crowd and a composer often used them to do strange things or experiment with weird ideas.
Have you seen all the weird things he did with his 9th symphony?
There's doubt that he would be trying to compose a grosse fugue like piece with the 10th.
The grosse fugue wasn't a weird experiment, it was the most appropriate music for the ideas and feelings he was trying to express.
Re: (Score:2)
Are you trying to imply that the 9th was anywhere as innovative as the Grosse Fuge? The 9th is a continuation of progression from his other symphonies.
The Grosse Fuge was pretty much revolutionary and definitely not been appreciated by the masses at that time. E.g. Wiki states "it was universally condemned by contemporary music critics. A reviewer writing for the Allgemeine musikalische Zeitung in 1826 described the fugue as "incomprehensible, like Chinese" and "a confusion of Babel"."
It wasn't until the
Re: (Score:2)
Are you trying to imply that the 9th was anywhere as innovative as the Grosse Fuge?
Yes
Re: (Score:2)
Then show where it's nearly as revolutionary. Surely not due to its tonality or perhaps technicality or that it feels incomplete. Moving around movements, and adding a choral section is interesting (at that time, and perhaps shocked people's expectations) and perhaps had not been done before but not revolutionary. Or are you talking about it in a more political sense. In that case perhaps it was more revolutionary.
One piece was instantly loved. The other took more than a hundred years to appreciate by
Re: (Score:3)
"We find Beethoven's Ninth Symphony to be precisely one hour and five minutes long; a fearful period indeed, which puts the muscles and lungs of the band and the patience of the audience to a severe trial..." -- The Harmonicon, London, 1825
"His great qualities are frequently alloyed by a morbid desire for novelty; by extravagance and by a disdain of rule...The effect which the writings of Beethoven have had on the art must, I fear, be considered injurious." -- Letter to the Editor, Quarterly Musical Magazine and Review, London, 1827
"The fourth movement is, in my opinion, so monstrous and tasteless and, in its grasp of Schiller's 'Ode,' so trivial that I cannot understand how a genius like Beethoven could have written it. I find in it another proof of what I had already noted in Vienna, that Beethoven was wanting in aesthetic feeling and in a sense of the beautiful." -- composer Louis Spohr, a contemporary of Beethoven
Don't make things up. Research before posting.
Re: (Score:2)
Considering how well received it was in its premier, having a full standing ovation that took an extraordinarily long time to end, I'm pretty sure that it was received pretty well by MOST people. Of course there will be some critics who say the opposite.
Here's 80 pages of then contemporary reviews. https://www.bu.edu/beethovence... [bu.edu] Most are favorable, and aren't nearly as harsh when they think there's issues.
As to the Grosse Fuge. I doubt you will find even one contemporary good review of that work. Th
Re: (Score:2)
Can't really say I like the classical period which is my least favorite period, and often change radio stations (internet radio) when they pop up. Even the classical-romantic transition until around the mid 19th, I find somewhat boring and unlistenable.
You should publish some negative reviews of Mozart.
Re: (Score:2)
Moreover, if you were to play e.g. a bunch of Beethoven's music to someone who never heard Beethoven before, and then play the 9th, I think most people would think of the 9th a continuation. Play the grosse fuge, and people may think wtf is that, is that by someone else? I heard just earlier today a piece by Beethoven that I had never heard before. I didn't know it what it was, but knew it was probably some piece by Beethoven, and it turned out to be his op.8, one of his earlier works. I don't even like
Re: (Score:2)
You recognized Beethoven's Serenade as written by Beethoven on your first guess? The only way you could have done that is if you've listened to a lot of Beethoven, or if you don't know anyone but Beethoven.
The grosse fugue is quintessential Beethoven, the main difference is it's a bit dissonant. Ignore the dissonance and listen to the energy and form. When you listen to Piano Sonata 32 you will hear a lot of similarities.
No, it did not (Score:5, Insightful)
Re: No, it did not (Score:3)
I think that this is exactly the interesting thing in this development. If an algorithm can finish a work from Beethoven like this in such a way that listeners cannot distinguish between the two, it may just mean that we taught the algorithm part of the genius of Beethoven (frozen in time, as it were). Although not the evolvement of his genius.
TFA does not mention whether they tried to have the algorithm create an existing piece of work by Beethoven, based on the basic motif. Would be interesting to see wha
Re: (Score:3)
If an algorithm can finish a work from Beethoven like this in such a way that listeners cannot distinguish between the two,
It seems like the musically competent folks here can tell the difference. They can even tell us why what was produce doesn't seem right.
it may just mean that we taught the algorithm part of the genius of Beethoven
Not really. You can very easily generate music that sounds like someone else's music. We can do it with Markov chains, for goodness sake! It's just patterns and probability, after all. Playform says they use GANs.
Would be interesting to see what would happen then.
The result is entirely predictable. You'll get something that sounds similar to whatever you used in training. I don't know what else you could expect. The
Re: (Score:2)
AI is nothing more than a derivation function.
If you give a list like this to an AI:
A...B...C...D...E...
Then one AI could say ...F...G...H... and another ...A...B...C...
The actual information is contained in those first 5 items. The rest is derivation and therefore adds nothing new, and since there are an infinite number of derivations possible picking out one and saying "This is what Beethoven would have done" is ludicrous.
Re: (Score:3)
For sure. This doesn't seem to be any different than what you can get with Markov chains.
I wonder if articles like this would get even a fraction of the attention the get now if we had stuck a word other than 'neural' in front of 'network'.
Re: (Score:2)
Computers are good at things that are hard for people to do, and people are good at things computers really struggle at accomplishing at a most basic level.
A good portion of music composition is rather repetitive and mathematical. A piece that we enjoy has a pattern that we can lock onto as familiar, new elements stop us from being board, however while surprising they are often in the realm of safety.
Beethoven talent was coming up with the pattern, and then stating what the variations and new elements to t
It doesn't matter (Score:5, Funny)
Everyone knows it's the odd-numbered symphonies that are the really good ones. The AI would have to finish the never-started 11th symphony to create something really impressive.
Re: (Score:3)
Well, actually according to Australians, it's the ones that aren't powers of two that are the really good ones. May be we will like this thing down here.
Source:
https://en.wikipedia.org/wiki/Classic_100_Symphony/ [wikipedia.org]
Re: (Score:1)
Wow, New World Symphony is number 1 in Australia. Wonder how many people know it's a programmatic piece that follows the poem The Song of Hiawatha by Longfellow.
Re: (Score:1)
You're both wrong: his Second is one of his best.
Well, one of his six best. Out of nine.
Hell! Why not do more? (Score:4, Insightful)
Re: (Score:2)
It's always only gonna be multi-layered permutations of the existing works, done by an algorithm that literally known nothing else whatsoever, no life as a human, nothing of the world in which Beethoven lived, not even that our planet exists.
AKA the media industry's wet dream to automatically generate ALL the remakes, prequels, sequels, "on ice", action figurines, and all the other shit to squeeze every last drop out of any innovative creative work until it is a useless husk, hated and despised by everyone
Re: (Score:2)
It's always only gonna be multi-layered permutations of the existing works
That's a good description of it.
Should try it on an abbreviated Asimov's C-chute (Score:2)
Re: (Score:3)
"Finally". As in "whatever the programmers... yes, *programmers* ...decided to be acceptable when *programming* the weights of the tensor of that ANN".
Hint: They do not know either. Otherwise they'd written a proper algorithm instead of using this universal black box function hack.
Re: (Score:2)
It's been a long time since I read C-chute. Where's the scene break?
Re: (Score:2)
The scene on the spaceship's hull would have been cut with a flashback into the interior, if Horace Gold had his way. He did not.
From the "Just because you can..." file: (Score:3)
What sort of mind can conceive of this and more importantly how did they think it was OK?
Touch up on the Mona Lisa anyone?
Re: (Score:2)
Touch up on the Mona Lisa anyone?
Been done [pics.me.me].
Re: (Score:1)
Someone created an AI that produced pretty good fake Bartok. He deleted the program because he didn't see any value in an infinite amount of Bartok.
Re: (Score:3)
What sort of mind can conceive of this and more importantly how did they think it was OK?
Why wouldn't it be OK?
If it succeeds, we gain enjoyable new music.
If it fails, nothing of value is lost (except the researcher's time, but even then they've at least learned what doesn't work)
In either case, the original music still exists as it always has, and nobody is going to confuse the two.
Re: (Score:2)
Cool (Score:2)
So now that the AI can express emotion - since that's exactly what artists do - do we need to consider it a sentient being and leave it alone, give it all the rights it needs? Can't recognise a white truck or a pedestrian yet it's perfectly capable of churning out symphonies?
I'm simply amazed that the number of imbecile knowitall programmers hasn't come down in the past decades one bit. If anything, it has grown steadilly, churning out new "programming languages", new "concepts" daily. Retiring from this sh
Re: (Score:1)
So now that the AI can express emotion...
I think a problem here is that the nerdy types doing this neural network stuff wouldn't recognise an emotion, unless it actually hit them in the face. Which it probably does from time to time when they meet ordinary folks, so they play computers, because it is safer.
Sorry. That was I admit a most frightful stereotype.
Re: (Score:3)
So now that the AI can express emotion - since that's exactly what artists do
Have you listened to music recently? I think the AI did pretty much what pop music does: Copy old stuff, edit just enough to avoid a plagiarism suit and shit it out.
Re: (Score:2)
Re: (Score:2)
So now that the AI can express emotion
It didn't express emotion.
Can it do it better than a natural NN? (Score:2)
Do some blind testing with audiences and see which they like more, a piece finished by a natural NN composer or by this cobbled together half expert system half ANN.
Re: (Score:2)
I think you should first of all define "better" in the realm of art.
Once you're done with that, we'll talk about pretty much anything else. I'm not holding my breath.
And... (Score:2)
And strangely, it sounds a lot like twinkle twinkle little star.
Cannabis for sale online (Score:1)
Did we really need AI to complete the symphony? (Score:2)
Any musician worth their salt can end a piece with a "shave-and-a-haircut...twoooo-biiiits" cadence.
Now they can get started on... (Score:2)
...P.D.Q Bach's "Unbegun" symphony.