In 1990, Ricardo Chailly and the Concertgebouw Orchestra gave the first ever full performance of Luciano Berio’s piece Rendering.
Rendering was based on a number of sketches that were left behind by Franz Schubert, originally intended to eventually become his 10th symphony. Unfortunately, Schubert had died at the mere age of 31, before he could ever finish the work. Still, the symphony was there, for the most part at least. The notes were written on two staves, like a piano reduction, and there were strange gaps in places, like Schubert had known where he eventually wanted to go in the music, but didn’t quite know how to get there yet. But, if so much of the piece was there, there was nothing to stop anyone from completing it – finishing the orchestration in a way that fit the music, and closing the gaps to make the piece whole again. It would have been exciting to hear what Schubert’s final musical ideas were, just before his death, right?
The idea of finishing a piece someone else has left behind has its most famous example in Wolfgang Amadeus Mozart’s Requiem. According to the most common historical explanation, Mozart’s wife Costanze, needing the money, paid a number of Mozart’s students to finish the work in secret so that she could present it to its commissioner and get the promised sum. Composer Joseph von Eybler first started working on some sections of the piece before giving up and leaving the piece to be completed by Franz Xaver Süssmayr. While it is impossible to discern with precision where Mozart’s work ends and Süssmayr’s begins, it’s often said that Süssmayr orchestrated the work up to the Confutatis, wrote the Lacrimosa from bar 9, wrote the Sanctus, Benedictus, and Agnus Dei, and copied Mozart’s Introitus and Kyrie to write the Communio.
Süssmayr is not the only one who has completed Mozart’s requiem. Among those who also tried their hand at it are Duncan Druce, Richard Maunder, and Robert Levin. My favourite version is currently Levin’s; in my opinion, it shows great technical skill and is the one that sounds the most “Mozartean”. But judging a completion in this way also carries with it questions of whether a completion should be like a restoration, or a creative undertaking in itself. Is it important to come as close as possible to what Mozart may possibly have written himself – a task that will require rigorous historical research, leading to a piece that is functions just as much a thesis of what defines Mozart’s late style? Or is the quality of the piece the only thing that matters, whether it follows Mozart or not, and therefore we should allow the composer pure creative freedom?
I’m not going to answer these questions; you can answer them for yourself. But one thing should be kept in mind in all cases of a completion: Someone is doing them, and it is not Mozart. When we listen to the Requiem, despite what Costanze may have wanted, we are not just listening to Mozart. We are listening to Mozart and Süssmayr, or Druce, or Maunder, or Levin, or someone else. And this other, not-Mozart person, did real work, has real knowledge of music, and real skill, that they put into the work. Whether the work is in restoration, or in creation, it requires real ideas, from the person who is doing it – who is not Mozart! So, when we listen to a work that is a completion, we should recognise the human work of whoever was the completionist, and respect their knowledge, their craftsmanship and ideas. Their contribution cannot be forgotten, should not go unnoticed, because without all of their skilled, intensive labour we would not even be able to hear the work as it is performed.
When Berio wrote Rendering, he decided to orchestrate Schubert’s material in the way he thought was accurate to Schubert’s style. To most classical music listeners – save for perhaps the real Schubert experts – it would be indistinguishable in style from any other Schubert symphonies. However, whenever there was a gap in Schubert’s sketches, he decided not to fill it with any composed transition that would sound like (an imitation of) Schubert. Instead, he chose to fill them with new, unheard of material; material that was contemporary, modern, and alien to Schubert and his contemporaries. Some people say that Berio was ‘giving back’ to Schubert, adding in something of his own style, but I prefer the interpretation that Berio essentially composed emptiness, blank space, a kind of silence beyond ordinary silence. In those moments, the music loses its direction, its propelling force, and instead gives the listener time to think, to reflect. It makes us crucially aware of the fact that Schubert is dead, he is no more, and therefore the piece cannot truly be finished as it was conceived; after all, the only one who knew what the completed version was supposed to sound like was Schubert himself.
In a sense, Berio’s way is the postmodern way, the self-aware way, to do a completion. Rendering keeps the original sketches intact, brings them out with skill, but is also critical of what it itself is doing. It’s a completion that recognises that a true completion is impossible, a completion that says this is all the Schubert we have, and all we will ever get. There will not be any new pieces by Schubert, ever. Schubert is dead, and time goes on. Not only is this a really cool idea, in my opinion, but it is the idea that, to me, pushes the piece over the edge into being a work by Berio, rather than a work by Schubert that was finished by Berio. By being self-aware, by being thoughtful, Berio has made the piece his own, and we can recognise the creative genius of not just Schubert, but just as much of Berio.
So, why write this now, when all of this happened in the 1990’s and has been written about extensively?
This year, the Beethoven Orchester of Bonn performed “Beethoven’s 10th symphony”, completed by a team of musicologists and, most importantly for the media covering the event, using Artificial Intelligence that had been fed all of Beethoven’s oeuvre. And, by reviewer consensus, it was terrible. In the words of Richard Morrison in The Times: “[It] is recognisably Beethovenian, in the sense that it sounds as if someone has sliced up his nine symphonies, sloshed them around in a mixer, and stuck together the new bits in a random new order”.
Peter van der Lint wrote in the Dutch newspaper Trouw: “The music was hanging together like loose bits of sand. The bits the computer chewed up and spit out did not go together at all. The imitation of Beethoven’s sublimely built climaxes here deflated like a balloon time after time. It was bloodless and boring, with sometimes laughably weak connections between the motives”.
Well. Instead of explaining again why the music is bad (you can listen to it for yourself) let’s look at why completing Beethoven’s 10th using AI may not have been a good idea in the first place. Firstly, there are the historical facts. As youtube music critic David Hurwitz pointed out, Beethoven had hardly even begun writing the 10th in his life; he had written down some ideas, but never even started a first draft before he died. You can’t complete what has not been started; any attempt at doing so will just be a piece by someone else made to sound Beethoven-ish. Secondly, there is the undeniable distinction between humans and machines. As Jan Swafford wrote, even in a field in which computers could outperform humans by far, it is simply way more exciting to see a human accomplish something than to see a machine do it. A machine can hit a baseball further than any human, but we still want to see a human hit the ball, not a machine. Leaving the completion of Beethoven’s work to a computer robs us of the thrill of recognising a human’s craft, skill and dedication that is, as I mentioned before, essential to the completion of a work.
Thirdly, there seems to be a misunderstanding of what music is on behalf of the team that created the AI. Some may think that since AI can outperform humans at chess, and nowadays even at Go, it will eventually outperform humans at music too. But chess has not only clear rules, but clear win states: If you play better than your opponent, you win the game, and there is little the opponent can do to argue otherwise. Music is not like that. There will always be arguments about what makes good music, there will always be differing tastes, and we will never find out the ultimate best way to make music, because there isn’t one. The fact that that is true is a gift from the gods, something that we should embrace, even celebrate, not a problem that needs solving.
The fourth and final issue I have with this project is relatively minor in comparison, but to living composers and musicians, it may be (it should be) the most offensive: It presupposes that Beethoven was so good, so untouchable, so genius, that the only way we could have something like it again is by invoking supercomputers who can do more calculations in a second than a human can do in a year. That no other human could ever be as good as Beethoven, that it would be pointless to even try. Beethoven was a great composer, but there were many things he could not do that composers after him could, and still can, do. Humans have learned from Beethoven, and many of the things many of us know about composition we know because of him. But we have also continued to innovate, to refine and sharpen compositional technique even further, and think of ideas beyond what he, great as he was, could think of. Even if we can’t complete the 10th symphony, simply because no one can (because it wasn’t started in the first place), we can make pieces that are like it, pieces that are as good, and pieces that are even better. It takes knowledge, it takes skill and hard work, but composers today are showing that they are capable of these things time and time again. And that is what we should celebrate, right here, right now.
Pepijn Streng 