Beethoven or Bot? – Evaluating an AI’s Completion of Beethoven’s Tenth Symphony

Although most of the hubbub surrounding Beethoven’s 250th birthday has subsided, a few bits of the celebration have lingered into 2021 as concert halls worldwide open up once again. However, one recent Beethoven item has raised many eyebrows in the music world. Back in September, the news broke that a team of computer specialists and music scholars had “completed” Beethoven’s unfinished Tenth Symphony using AI technology and scant scraps that Beethoven left behind upon his death in 1827. Dubbed “Beethoven X: The AI Project,” it was also announced that a recording of the results—played by a live orchestra—would be released the following month. Like many, I was skeptical. This just seemed like another lame excuse for more Beethoven deification, one that would take the focus away from issues that are currently more pressing, like promoting diversity and equity in classical music.

Still, I was curious what the results would sound like. So, I recently met over Zoom with my friend Tanner Cassidy—a PhD candidate in music theory at UC Santa Barbara—to listen to it and share our impressions. (There may have been a smidge of alcohol involved as well… *wink*) Below are excerpts from our discussion, which have been edited for length and clarity. What did we think? Does a computer have the potential to live up to Beethoven or is this something best left in the trash bin? Let’s find out…

Kevin McBrien (KM): So about a year or two ago, this German telecommunications company [Telekom]—with AI specialists and music scholars—was like, “Hey, let’s take these incomplete sketches of Beethoven’s Tenth Symphony and feed it into an AI program and see what happens.” And this is the resulting piece that came out of it. There was this musicologist—Barry Cooper—who reconstructed the first movement in the 80s and that’s been recorded and released as kind of a hypothetical Beethoven 10. And this one, apparently, is just the third and fourth movements.

The musicologist Barry Cooper realized the first movement of Beethoven’s Tenth in the 80s, which was subsequently recorded by the London Symphony Orchestra.

Tanner Cassidy (TC): So there’s no second movement that exists?

KM: I guess not, or there’s not enough to go off of.

TC: I actually have some experience with AI-generated music. An undergrad friend of mine was writing bebop-based algorithmic composition, where he fed it Charlie Parker licks, and then he had me play what the computer spit out, which was just nonsense. AI has really struggled with rhythm, so I’m really curious to see what rhythm sounds like. [KM edit: Similarly, Google acknowledged J.S. Bach’s birthday in 2019 with a Doodle that, with the help of AI, generated a Bach-style harmonization around a two-bar melody you fed it.]

KM: Yeah, I heard a snippet of it on an NPR story, and it’s weird. So, I’m also curious to listen to the whole thing.

TC: Well, I’ll mute my audio, and then we can listen to this.

[We listen to the third movement.]

TC: It’s very boring. Not that I’m surprised.

KM: Yeah. I mean, some of it’s plausible. Stylistically, it seems like it would fit in the late period, but there are other things that just seem out of character from what he would do at the time… I mean, you have the Fifth Symphony motive throughout.

TC: To me, it sounds the most like the middle symphonies. It doesn’t sound like a tenth symphony. It sounds like a 5.5 or a 6.5. [KM: That’s true!] On the one hand, I’m surprised at how much it does utilize a motive, but it doesn’t do anything well with it. It does some things with texture, but I don’t really get a sense that there’s anything being developed throughout, and it’s very, very, very boring harmonically. I also find that there are some odd repetitions of motivic blocks. There are weird parts where it just kind of restarts. It hasn’t finished, but it has ground to a halt, so it just restarts. [laughs] That happened in the beginning, and it happened at the end.

There are some pretty wonky things about this.

KM: Yeah, one of my notes says, “Trio seems too long and completely unrelated.” It keeps going, which isn’t necessarily something he wouldn’t do, but I was like, “Wait, how is this related to what came before?” And then suddenly we’re back.

TC: I find that it’s just a collection of late classical orchestrational textures, and there are just these fuzzy seams to connect them. And then the motive comes back because it has to come back because it’s a scherzo and trio, but it doesn’t feel like a resolution. It’s very odd.

KM: There’s also some syncopation stuff that seemed a little wonky, especially at the end of the main motive. The first half makes sense, but the second half just seems like a programming glitch or something.

TC: Well, and also there’s a couple of times, particularly in the trio, where there’s this, what would almost—dare I say it—be like a topic. There’s a little fanfare thing or whatever. And it just kind of happens and then nothing is made of it. It’s weird because it has what could in a different context seem like a musical reference to something, but the computer doesn’t understand why you would do that or where, it just knows that you can do that.

KM: Right. I forgot to mention this, but I think when they were feeding the program the sketches, they also—because there’s so little to go off of—ended up feeding it Beethoven’s entire output as well, so it would learn how to solve musical problems and how he would go about writing things. I was reading something that said there were different versions that came out, and this is the one that made the most sense in the end. So, on one hand, it makes sense that there are clear callbacks to things that the computer knows that Beethoven wrote and things that we know Beethoven wrote.

The AI program, probably.

TC: That would make sense. It’s probably a matter of how much they need to feed a program for it to learn. But the problem is that you can’t just give a computer all of Beethoven’s oeuvre, because I would imagine that it would probably treat everything with equal importance. His First Symphony sounds nothing like his Ninth Symphony, but you can’t just give a computer those two things and tell them that they’re the same, because then you get something really weird like this. What I’m hearing here is so old school by the 1820s. It’s just so crazy to me that—this is going to come across as really harsh—the people who are so desperate to fetishize Beethoven some more just completely ignored all the mythologizing around his three periods of composition. Because that defeats the whole purpose of why we even like Beethoven.

KM: Right. We can talk about this more after we listen to the fourth movement, but one problem I have with this whole thing is the legitimacy it has been given. Like, A) it was created by this esteemed team of artificial intelligence specialists and musicologists and B) was recorded by the Beethoven Orchestra Bonn [conducted by Dirk Kaftan]. It gives it this veneer of a “product.”

TC: It’s the fact that they’re trying to fill in the gaps with a computer. They’re doing all the wrong things. You can do computer-generated music that’s still very much original, even if it’s building on reference. Say you did algorithmic composition by feeding it certain parameters, but all are references to classical pieces. That’s still the composer. They’re still a composer creating a work. On the flip side, it’s like the people who wrote those supplementary movements for The Planets. [KM: Yeah!] There’s a person writing something original and they’re piggybacking it onto something “great,” but they’re still a human. But this doesn’t benefit anyone. It’s the worst kind of exercise. It’s what I tell my composition students not to do.

An example of computer-generated music that’s actually good and interesting.

KM: And with Barry Cooper’s realization of the first movement, at least that was a scholar who sat down with the sketches and had this knowledge of Beethoven’s style at that point and what he was doing. But there’s still a person behind that. [TC: Yeah.] But then I think of something like Berio’s Rendering, which is a completion of Schubert’s unfinished tenth symphony [in D major]. You have Schubert clearly, but then you have Berio’s “fingerprints” throughout. And in a way, that’s more interesting because you have the dual authorship rather than just, “We gave this to a computer and this is maybe what Beethoven would have wanted.”

Behind-the-scenes footage of the Beethoven X team.

TC: There are all kinds of ways you can ruminate on the past, even within older idioms. Polystylism is like that. There are so many ways to do something retrospective. This sounds like you’ve got a second-rate DJ with a MIDI controller just willy-nilly picking between different gestures and then forming those into a loose three-part structure. It’s crazy. [laughs] Honestly, the fact that we’re even hearing it means that it’s not offensive to the ear. I was expecting it to be way worse just sonically, even though it’s exactly what I thought it would be artistically.

KM: Well, I’d be curious to hear what some of the other versions… [TC: The failed versions!] I’m sure there were some that were just really rough, but that’d be more entertaining, I think. Well, let’s subject ourselves to the last movement.

TC: Here we go…

[We listen to the fourth movement.]

TC: Why so much triangle?

KM: I don’t know! [laughs]

The AI really loves triangle, apparently.

TC: There was just so much triangle! That stuck out more to me than the organ did. There’s one part where it’s just the triangle hitting every… What’s going on? [laughs]

KM: Beethoven’s lost Concerto for Organ and Triangle. [laughs]

TC: Also, the organ’s not used like I think he would have used it at all. It’s super Baroque.

KM: That’s what I wrote! “This sounds like Handel or Bach.”

TC: Yeah, because there’s a couple of moments where the organ is just a part of the texture, which is interesting. [KM edit: Speaking of, the organ part in this recording was played by Cameron Carpenter.] But most of the time, it’s doing these really weird… There’s, like, Baroque ornamentation! Bruckner and Mendelssohn would have loved it, that’s for sure. Geez…

KM: Yeah, there are points where it sounded like the technician mixed in some stuff from his iPod. There are hints of Mendelssohn’s Reformation Symphony. There’s Russian-ish stuff, like Tchaikovsky…

TC: Well, the opening is just so freaking weird!

KM: Yes! I wrote down, “Stravinsky’s Firebird.” [laughs]

TC: Yeah! The beginning is just bizarre. And then, boy, oh, boy… It turns out they did import piano music into the algorithm—we know that for a fact—because the f***ing “Pathétique” is in it! [laughs] I heard that and I was like, “What?”

KM: And it’s a cyclic symphony, I guess. The Beethoven 5 motive is back! [laughs] The form was harder to pick out, too.

TC: I wrote in my notes that the form works for me here, but that’s not really the right way to put it. The form works for me here as much as it’s a longer movement, but there’s more stuff happening, whereas in the other movement, like you said, the trio just feels overly long. Here, it moved from one thing to another in a way that to me, felt much more organic. But no, you’re right, it’s still all the same problems. The AI really struggles with contrast, because whenever it does anything orchestrational or dynamic, it goes way too far, and it doesn’t do them in places that are earned.

KM: It’s either really homogeneous, or it sticks out like a sore thumb.

TC: And I still found this uninspired harmonically, too. I found it a more palatable thing to listen to than the third movement, but most of my problems from the third movement are still here. It was just so weird. It was the weirdness that made it interesting to me. [laughs]

KM: Yeah. I mean, the third movement sounds more like something Beethoven could have done at some point. [TC: Sure.] This one was a little less so.

Perhaps the AI had something like this in mind…

TC: I also wonder if organ wasn’t in the thing. I wonder if the people that were making the AI or if the AI just decided, “Oh, well, the fourth movement of the Ninth is when it ‘bursts into song.'” [KM: Yeah.] And I wonder if that was the thought here, is that, “Well, in this fourth movement, we should have just this banger of an organ solo.” [laughs]

KM: “What we really need is an organ!” [laughs]

TC: The metaphor of “music could not contain the boundless passion words could…”—that’s the Wagner thing, right? [KM: Right!] That’s why the Ninth Symphony should have been the last symphony for Wagner. It’s like, Beethoven did all he could musically and then it burst into song, and the Meister said, “Let there be voices and therefore opera.” That’s the whole thing! So you can’t just copy that rhetorical strategy for organ, right? It just doesn’t work! [Both: laugh] It’s just it was so Baroque and so liturgical. I was so taken aback by that.

KM: Yeah, the melody there is very much like Mussorgsky’s “Great Gate of Kiev” [from Pictures at an Exhibition] or Tchaikovsky’s 1812 Overture or something.

TC: Yes! That’s what made me think a little bit of Bruckner with this very Austrian sense of form and content, but with a heavy piety and liturgical counterpoint. But also, surely the algorithm is just freaking out over “Eroica,” right? [KM: And the Fifth!] Because I don’t remember anything else in Beethoven’s oeuvre that is just like… the clear, punctuated, tutti chords. But it just doesn’t understand how that works; they’re clipped too much.  They’re so pointillistic. It’s weird.

KM: It was surprising, but not a good way.

[We then start to wrap up our conversation.]

KM: Well… final impressions?

TC: Are we done? Can we not do this anymore? Not you and I, of course! [KM: laughs] Just like… is there anything else? Let’s just keep going. Let’s just do the Eleventh. Like, why would anyone do this? Think about how many composers will never have a major orchestra perform their work despite being career, lifelong composers. It’s just so exhausting that this even exists. It’s such an affront to what art should be good for. [laughs]

KM: It’s a stunt. There are so many other, interesting things that could have been done, should be done.

TC: It’s the audacity of it having, like you mentioned earlier, this stamp of legitimacy. I mean, if somebody had done this as a master’s or a doctoral thesis and had done it with Logic, it would be—

KM: Yeah, just like an experiment!

“Mary Shelley has told us the evils of this kind of thing.”

TC: Yes! It’s giving it this the stamp of legitimacy as if this were on the same pedestal as anything else the Meister has bestowed upon us “mere mortals.” I would rather listen to f***ing Wellington’s Victory than this. [KM: laughs] At least that was a piece that a human being wrote! And it’s not that I’m opposed to computer-assisted or computer-generated music, but this isn’t… I mean, Mary Shelley has told us the evils of this kind of thing. [Both: laugh] Although the problem is that Beethoven can’t speak for us. He has no consent in this.

KM: That’s the thing, too, that scares me about this is the ethical implications. If you have a contemporary artist who passed away and you find—I don’t know—scraps that are reconstructed and then it’s suddenly passed off as their final opus. You could have things like that, and people will find ways to capitalize on that without the person having a say, unless they explicitly have it on record or something in their estate that prevents that. But, yeah, this could just lead to a lot of ethical gray areas. Like, who is the creator?

TC: You’d think that with the sum total of all the wonderful things that scientists and engineers and scholars can do with technology, we’re not bettering mankind; we’re just resurrecting corpses. It’s horrible. This doesn’t benefit anyone. It only serves to further dilute whatever untouchable legacy these people have.

KM: It’s canon obsession. It’s the pining of, “Oh, if only Mozart had lived longer! All these works that could have been!” Maybe that’s just part of the magic—the “what if,” the supposition.

TC: Yeah, there’s a certain point where you could wonder hypothetically, “what if?” But it’s when you actually try to take that line of question further than just a rhetorical one that you’re removing real estate from people that are alive. There’s just something so immoral in that to me.

What would Beethoven think of this?

KM: And even with the dead folks that never got their chance. [TC: Yeah!] It’s great to see renewed interest in people like Florence Price and Clara Schumann and William Grant Still, but it’s also like, is it going to have any real effect until we let go of this obsession…?

TC: It’s like an addict looking for another fix. There’s no more Beethoven here. That’s it. We’ve got all that we will ever have. He had his day. Like you said, whether it’s expanding to just more dead people, or—God forbid—focusing on living ones, that’s where we should be putting our time and attention. Because, again, I’m not saying that Beethoven needs to go away because Beethoven won’t go away. As long as people like his music, he won’t go away. But we need to stop.

KM: It’s just reeks of desperation.

TC: It’s like when Spotify tells me that Mozart’s dropped a hot new track. It’s like, no… [Both: laugh] Well, thanks for subjecting me to this. [KM: You’re very welcome!] I hope you get a lot of clicks. [Both: laugh]

KM: We’ve all found ways to use classical music for personal gain, and this is my way of doing so. So, thank you! [laughs]

TC: Wow…

Our listening faces pretty much sum it up.

2 thoughts on “Beethoven or Bot? – Evaluating an AI’s Completion of Beethoven’s Tenth Symphony

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s