As most of you reading this probably already know, I fell in love with the Ninth when I was quite literally a baby. It was then that this masterwork gained a special place in my heart, above all other forms of musical expression, and there in its prominent place, it will stay forever. You may also know that, for more than five years as of when this post is going live, I’ve been working on a book manuscript about the Ninth. I listen to it daily, and whenever I have new ideas about it—quite often, actually—I make a note on my phone. When I collect enough notes, I write something coherent out of those notes. Doing this since 2016, I’ve produced almost 100,000 words, and I have far more to say about this masterpiece.
As much as I have to say about the Ninth, today, I actually want to talk about the Tenth, the unfinished symphony commissioned at the same time by the same ensemble. The Ninth and Tenth were commissioned together, like other pairs of Beethoven’s symphonies: Op. 67 (the Fifth) and Op. 68 (the Sixth), and Op. 92 (the Seventh) and Op. 93 (the Eighth), in 1817, by the London Philharmonic Society. Seven years later, premiering in Vienna rather than London—to the chagrin of the English public—Ludwig van Beethoven gave the world what would be his crown jewel, the Ninth. 34 months passed between the premiere of the Ninth and Beethoven’s untimely death at the age of 56 in the earliest days of spring, 1827, the second of the two symphonies London commissioned still unfinished.
Ludwig van Beethoven was not like Wolfgang Amadeus Mozart, in that he did not have an apprentice working under his tutelage at the time of his death, whose job it would be, according to his will and other instructions he left before his death, to finish the works he could not complete but for which he had outstanding commissions. Mozart had the benefit of the presence of a young but very talented Franz Xaver Süssmayr under his wing at the time of his death. Those works which he left unfinished—most notably the Requiem which Mozart saw presciently as being written by himself for his own funeral—he instructed the then-25-year-old Süssmayr to finish. And finish those works he did. It is precisely because of Süssmayr's work that we have a complete version of K.626, the catalog reference for the aforementioned Requiem, considered among the greatest of Mozart’s output, and among the world’s great masterworks.
Beethoven died without completing the Tenth, leaving only piecemeal sketches and with no designated “heir” (like Süssmayr was for Mozart) who should finish the unfinished works in the now-departed master’s style. The world’s interest in the possible completion of the Tenth when a musicologist named Barry Cooper claimed in the late 1980s to have put together a reconstruction of the symphony to be performed in the venue of and by the ensemble which had commissioned the original work 170 years before. There is one major problem with this completion of the Tenth – one that no one could ever blame on Barry Cooper himself: Beethoven only wrote sketches for the first movement before he died, and so that is all Cooper could faithfully complete from the sketches he had, and so no completely reconstructed Tenths exist, as far as I know.
In the course of my casual reading up on the Cooper completion and his process, I came across a piece of news immensely relevant to both my deep love for music and my career in computer science still a few years away. Modern computer science, no doubt, placed tremendous emphasis on big data, AI, neural networks, and machine learning. As evidence for that, try creating a new account anywhere. You’ll have to select every occurrence of some object (or maybe “objects” if you have multiple rounds) in a grid. If and only if you find all the objects you should, the server will let you continue creating your account. Want a new email? Do this. Want a new Twitter, Facebook, or Instagram? Do this. Want a new dating profile? Do this. It’s everywhere. Whether you know it or not, you’re teaching the machine learning algorithm “I am not a robot because I know that all of these pictures, I’ve selected are the item you asked me for. Now, given that, when we flip the script and I ask you to help me identify an image, you just grew your knowledge base a little.” That you can do this correctly is, to the computer, proof that you are human and not another computer like it.
Beethoven died without completing the Tenth, leaving only piecemeal sketches and with no designated “heir” (like Süssmayr was for Mozart) who should finish the unfinished works in the now-departed master’s style. The world’s interest in the possible completion of the Tenth when a musicologist named Barry Cooper claimed in the late 1980s to have put together a reconstruction of the symphony to be performed in the venue of and by the ensemble which had commissioned the original work 170 years before. There is one major problem with this completion of the Tenth – one that no one could ever blame on Barry Cooper himself: Beethoven only wrote sketches for the first movement before he died, and so that is all Cooper could faithfully complete from the sketches he had, and so no completely reconstructed Tenths exist, as far as I know.
According to an article I’ve stumbled upon as a result of my research into the Cooper process, while we’ve been busy making accounts on digital platforms everywhere and playing the game of “identify every picture with a chicken in it,” a team in Germany has essentially been doing the complementary process (“here, humans have identified all of these as containing a chicken; now when I ask for a chicken picture, only show me these”), but with the music of Beethoven. Essentially, researchers in Germany have been working for months, if not longer, to teach a very powerful computer system all the patterns that underly Beethoven’s music, and to identify Beethoven from non-Beethoven.
From the recent reading I have done, this team claim they have been able to complete the Tenth. I am highly suspicious of this, for a number of reasons. First, if I understand what these researchers have claimed they have done (and what they will perform with a fully human orchestra in the coming weeks) is a full reconstruction of the Tenth stretching far beyond Cooper’s bound. Cooper only extrapolated from what we had in completing the first movement. This team, as far as I know, has claimed to have finished at least two movements, without any sketches whatsoever for the second, thanks to an AI which wrote the music based on the patterns the researchers fed it. Therein lies my first objection: we have no idea what the second movement would have sounded like, so there is nothing truly Beethovenian at all about the second movement—there is no original material from his hand at all in this version, only music made by the algorithm.
Second, I was able to listen to a few minutes’ snippets of the part of the Tenth written by the program. From this, I will certainly grant to the researchers that they have done a marvelous job teaching the AI the Beethovenian manner of composition. However, unless the Tenth was deliberately anachronistic—and I don’t believe it was intended to be—this AI falls significantly short of realizing a post-Ninth (that is, post-1824) Beethovenian symphonic undertaking.
There is certainly precedent for an “anachronistic” symphony among Beethoven’s works. The Seventh, written in 1811-12, feels very much in line with where Beethoven had by this time taken the symphony, given the leaps he had already made, particularly in the Third and the Fifth. The orchestra is much larger than it was when Beethoven wrote the First, and the Seventh feels much more Romantic. The Seventh, then, falls perfectly in line with the trajectory Beethoven had been steadily tracing since the First in 1801. But then, the Eighth, written at exactly the same time and being given the Opus number that immediately follows the Seventh’s (92 for the Seventh, 93 for the Eighth), feels completely out of place, perhaps as if it appeared 11 or 12 years too late. The style of the Eighth is so Haydenesque or Mozartian that it almost feels like it should be Opus 20 or Opus 22—the First is given number 21, and so 20 and 22 are the immediate predecessor and successor. For this reason—because it’s so different from the Seventh and the Ninth, and because it stylistically resembles the First—it’s been labelled the “ugly duckling” or the “odd one out.” But, as the fairytale tells us, it is unwise to ignore the ugly duckling, since the duckling can transform into a beautiful swan. And what we get out of the Eighth is not the ugly duckling, but the beautiful swan. Sure, the Eighth is so retrospective that it might be mistaken for the “Zeroth,” but, nevertheless, situated where it is, the levity of the Eighth provides a welcome contrast between the intense rhythmicity of the Seventh and the utter jubilation of the Ninth, a… palate cleanser between the two titans, if you will. Make no mistake, the Eighth is wonderful; but because of its situation between the Seventh and the Ninth, it is all too often unfairly ignored. Clearly, then, as I’ve just explained with the Eighth, there does in fact exist a Beethoven precedent for committing musical anachronism and hearkening back to a bygone era thereby bucking a trend of musical progress. This is precisely what this German AI-written 2-movement part of an unwritten symphony seems to do, from the recording I listened to.
It seems to be totally anachronistic, taking listeners all the way back from the Opus 130-somethings (likely where within in the sequential order the Tenth would have been cataloged if Beethoven had published it) to the late 60s, referring to the position in the catalog, not the decade of composition. The Fifth gets number 67 in the catalog, and the Sixth is number 68, so perhaps I might, for the sake of my argument, I’ll name what the deep-learning algorithm managed to compose 67a and then proceed to defend this choice. It seems the algorithm is good, but not great. Good, in that there are no obvious faux pas of tonality (nothing microtonal, a la Jacob Collier), meter (nothing in a ridiculously obscure time signature like 7/4), or melody. But I’m calling this composition 67a because it seems that that’s precisely what the algorithm produced: nothing more than an “extended director’s cut” or “blooper-reels-included” or “outtakes included” version of the Fifth. The Fifth and the Ninth are so distant, one so much later than the other, that I find it almost impossible that the real Ludwig van Beethoven would ever have opted to turn back the clock after the Ninth in favor of a style more reminiscent of his Fifth, for his Tenth. With the already-discussed exception of the Eighth, each time Beethoven gave the world a new symphony, something drastic changed: new instruments were added, new forms experimented with, length boundaries pushed, voices included for the first time. The Ninth was light-years ahead of anything else he or anyone else had ever written, and the trendline clearly displays that, on average, one symphony is more progressive than the one that comes before it.
After thinking about this problem for quite some time, I believe I've found the best possible rational explanation for why this machine-learning-generated reconstruction of the 10th seems to be so similar to the 5th. Machine learning relies on patterns. The researchers involved in this project certainly fed the computer the fifth, together with all the other symphonies, and a lot more of Beethoven's work so that the computer would have a sense of his style.
Particularly in the 5th, Beethoven was a proponent of writing with leitmotifs-- that is, of picking one or two short ideas and using those simple ideas ground the majority of a composition. The reason I have a problem with the AI-generated 10th in particular for ripping off the Fifth is precisely that the "short short short long" leitmotif in the 5th appears there so often-- and the computer incorrectly concludes that since the figure is so dominant in the Fifth, it should be equally dominant everywhere else. The result is a 10th that sounds like an amalgamation of rejected sketches from the Fifth, rather than an original composition of its own right.
There are 2 possible solutions to this problem. I know next to nothing about working with AI, so I don't know which would be more practical or easier to implement. One possibility is to show the AI all of the leitmotifs in all of the symphonies so that it can use all of that material to create a sort of "mashup leitmotif" drawing on the characteristics of all nine, or--and this is the solution I prefer-- the relevant knowledge libraries could and should be adapted so as to give the artificial intelligence its own tools to write unique leitmotifs so as to truly, completely differentiate the 10th from the others.
Of course, this is only because this music is not written by a human but by a machine, but there is another notable objection I must raise. In his mid-40s, Beethoven wrote the Eighth in the style he preferred in his late 20s, but with the maturity of a composer in his mid-40s. So, in a sense, though the Eighth hearkens back to an era long gone, it feels mature, round, and developed, like Beethoven otherwise was when he was writing other things in the mid-1810s. It in no way feels like this is the writing work of Beethoven from 1792 and rediscovered in 1812. No, it’s clearly written in 1812—though it exhibits the earlier style, within that, it certainly possesses the later period’s maturity.
This recording, on the other hand, seems overly repetitive. It is not repetitive to establish a mood like the Fifth—that repetition has a purpose. It repeats the Fifth’s signature rhythmic motif as its own and, several times, nearly verbatim, quotes material from the Fifth while trying to pass it off as original material. Anything that was truly Beethoven’s music, however Romantic it may have become toward the end of his life, always possessed a kind of Classical logic to it, even Classic logic about Romantic mood swings. Of course, there were mood swings, but at some level, they made intuitive sense even at Beethoven’s most Romantic. This demo certainly has the Romantic mood swings, but neither how the character changes nor (usually) either the beginning or the ending passages moving through these mood swings make sense being placed where they have been. At one point, a tune orchestrated as a chorale that bears what I would say is an astonishingly striking resemblance to the 16th Century German hymn melody “Gelobt sei Gott im höchsten Thron” (“Give Praise to God, in the Highest Throne”). The resemblance, for me, is suspiciously strong—as if perhaps the researchers taught the computers several common hymn melodies as single ideas that could not be broken down further, and the computers were taught how to harmonize those motives according to Baroque rules of part-writing, and either the computers or the researchers (or both) had a check-list of forms they were running through which they wanted to include in the movement. This kind of formulaic writing is, I must say rather unfortunately, the antithesis of the Beethovenian compositional process. In the Ninth, clearly, he takes the Ode to Joy through dozens of iterations beginning two minutes into the symphony and lasting until the very final chord. But never does this exercise of melodic development and variation ever feel formulaic, like someone is going through the motions or just checking boxes. Authentic Beethovenian music never feels that way; nothing—and especially not repetition— in Beethoven’s music ever exists there gratuitously.
Obviously, as a computer science student, I am amazed at how far out technology has come, such that we can now teach a machine how to write music like Beethoven did by feeding it hours and hours of his music and teaching it how to learn to discern the patterns contained therein. And as a musician who is so devoted to the study, performance, and pedagogy of Beethoven’s music, I am deeply intrigued at the possibility that anyone—let alone an artificial intelligence—might even attempt to finish for the world the unfinished symphony of Beethoven. But, on the other hand, as a student of the Ninth for my whole life and as someone also deeply familiar with the music that the Ninth directly influenced, I believe we don’t need to go looking for the Tenth anywhere. No human needs to try to write it, and no algorithm can learn to write it, because the work has already been written. Those of you who know Beethoven well or who have knowledge of what is in my book that I am writing about the Ninth can, by now, tell I am going to present what may seem like a rather cliché alternative solution to this enigma that completing or locating the Tenth has been, all the way back to the first performance of this other work about half a century after Beethoven died.
The key to the mystery of the unfinished Tenth was first revealed to me—though, of course, at the time, I did not know this—just a few weeks after my fourth birthday, in early summer, 2005. At that time, the BBC’s Radio 3 ran a 6-day series, beginning on a Sunday and continuing through and including Friday. Part of the commentary that accompanied that performance mentioned Brahms’s First was often thought of as Beethoven’s Tenth. At the time, so young, I didn’t make the connection. Years later, though, I would make the connection between the Ninth and Brahms’s First.
That connection has made its way into my book, and I’ll present it here as the ultimate solution. Listen back-to-back to the Ode to Joy from the Ninth and the principal theme of (only) the finale of Brahms’s First and you will notice a rhythmic-melodic figure that appears in the third phrase of the Ode which is presented nearly verbatim early on in the Brahms theme. Neither theme is presented right away, and the periods in the movements before the presentations of the two themes share similar characteristics and tonal qualities. Anyone who listens to both themes as I have suggested will very likely come away with the impression that the themes are very similar; transposing one into the key of the other (that is, both in C or both in D) yields a particularly heightened “melodic déjà vu.”
Brahms certainly knew of this connection—it was, in his words, absolutely deliberate—and so we know he fully intended to invoke this “déjà vu” to pay homage to the music of the great master Beethoven. In Brahms’s First, he continues the Beethovenian trajectory of turning the symphony into an ever-expanding medium. Beethoven’s final symphony paved the way for Brahms, who in turn opened the gates by his own symphonic oeuvre for Dvořák, and, after him, Mahler. This pseudo-relationship between Brahms’ First and the Ninth is the closest Beethoven came to a Mozart-Süssmayr in terms of continuing his creative arc beyond the Ninth. And it is precisely in Brahms’s First, a work that took 21 years to write from start to finish—for a sense of perspective, I am 20—that clearly have already found the musical, intellectual, and emotional successor to the Ninth. In it, we have something as close to a true “Tenth” as humanity will ever find. Let us always remember this connection each time we hear these two magnificent works.
Posts
No comments:
Post a Comment