Did you hear? They finally finished Beethoven’s tenth symphony using artificial intelligence! Pretty cool, right?
Well, no. That spectacular news story misrepresents Beethoven’s tenth and what AI can do for music.
Let’s start with the easy part.
Beethoven’s tenth is not ‘unfinished’, it simply doesn’t exist
While music history’s most famous unfinished symphony – Schubert’s eighth – is pretty much complete, Beethoven’s tenth was never begun. Even though the promo text of the AI project subtly tries to convince you otherwise:
“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.”
Really? Click here to listen to those sketches. Then let me know how they make you feel. Lamenting about unfulfilled promises? Or just ‘Meh’?
No, Beethoven fans and musicologists have not been madly pulling out their hair about what could have been. Most of them couldn’t care less. One of them already claimed to ‘finish’ the tenth more than thirty years ago. Which is to say: he wrote a symphony in Beethoven’s style using that handful of melodies. Because that’s all you can reasonably do. Unless, apparently, you can “harness the power of AI”.
The contribution of AI to the ‘finishing’ of the ninth is probably minimal
That aforementioned promo text is extremely vague about how they used AI to ‘complete’ Beethoven’s tenth symphony. So all I can do is make an educated guess – and I’m not an educated data scientist. Fortunately, this guy is, and he does a nice job of explaining how it works for Bach chorales, so I’ll start from there.
Up until now, those AI-generated Bach chorales are the most famous examples of computer compositions. They’re impressive, but the project is also a bit of a scam. The name ‘Bach’ inflates the implied accomplishment, while the key word is actually ‘chorale’.
A chorale is a four-voice setting of a Lutheran hymn. Here’s a classic example:
Beautiful. Yet also, in many ways, simple:
- The main melody (usually in the top voice) is a given so doesn’t need to be composed.
- The instrumentation – four voices and basso continuo – is fixed.
- All voices have a more or less equal number of notes to sing and move together in the same rhythm.
- It lasts no longer than a few minutes.
When you look at it from the standpoint of a computer, those are many fewer variables that it needs to worry about then when it’s asked to compose a, well, let’s say, … Beethoven symphony.
Nevertheless, the AI-composed chorales are extraordinary. How does it work? Not by music-savvy programmers – or IT-savvy musicologists – who write all the rules, that would take decades. It’s achieved by a process called deep learning where the computer kind of writes its own code. It works like this:
- The computer is given an input, such as a few notes of a melody of a chorale.
- The computer is asked to guess certain parameters, such as what the next note will be or what the underlying voices are.
- If the computer is ‘right’ – makes the same choice as Bach – the algorithm is slightly adjusted accordingly.
- After many, many trials and errors, the algorithm becomes so refined that it always guesses right or at least almost right.
- You can now use it to write new stuff in the style of Bach.
Number 4 is important here: you need a lot of input to train a deep learning system. In this case, there are about 350 Bach chorales, which our data scientist source calls “an extremely small dataset”.
Compare that to a measly nine Beethoven symphonies and you’ll probably agree that something’s not right here. The promo text of the project mentions that they used “completed compositions from Beethoven’s entire body of work”, but that’s not very impressive when you realize that a lot of that isn’t even orchestral, and that Beethoven significantly changed important aspects of his style during his lifetime. Is the AI offering us the tenth symphony as it would have been composed by the 1827 Beethoven or by the ‘average’ Beethoven?
For all those reasons, I find it hard to believe that this tenth symphony was completed by artificial intelligence. I suspect that a lot of work was done by the composers and musicologists involved. So much that they could have done it faster and cheaper on their own. But then of course, they wouldn’t have made the news.
Why is that idea so easy to sell? Why do we instinctively believe that artificial intelligence can do a better job of imitating Beethoven than a 21st-century composer? This quote from the CEO of Playform AI, the company that did the AI part of this project, speaks volumes:
“At every point, Beethoven’s genius loomed, challenging us to do better.”
For a man who probably uniwheels to work and says ‘engaging in ideation’ when he means ‘thinking’, that’s a statement with surprising 19th century overtones. Didn’t we put behind us this idea of ‘great men’ who lived in a ‘golden age’ and now hover like demigods over us mere mortals? Apparently not. A lot of us still believe that present-day composers (F/M) are no match for Beethoven. And that only our new deity can come to the rescue: the Almighty Algorithm.
AI and music: servant rather than master
All this doesn’t mean there isn’t a case for using AI in music making. In the end, artificial intelligence is no different than a harpsichord, a synthesizer or a laptop – a tool that can also inspire.
Instead of using AI to come up with music that we can just as well imagine ourselves, why not take advantage of its ability to make connections that we would never come up with, to think of completely weird, but sometimes oddly beautiful sounds? Please decide for yourself whether this piece of music falls under that definition:
“I find something hopeful about the roughness of this piece of music. Amidst a lot of misleading AI hype, it communicates something honest about the state of this technology; it is still a baby.”
Amen to that.
Want to keep up with my classical musings? Enter your email address and click subscribe.