Wednesday, September 29, 2021

Beethoven's "Unfinished 10th Symphony"

The upcoming premiere of Beethoven's 10th symphony has been making headlines this week, I first learned about it through this feature at The Conversation.  

The article confuses more than it reveals.  It reads like a 1500 word pat on the back for doing something that is poorly defined and poorly described.  The main purpose of the article is to build anticipation for whatever this symphony turns out to be.  

In short, a team of AI researchers and musical consultants took some rough sketches of unfinished Beethoven compositions and attempted to shape them into a completed symphony.  Getting a computer to perform this task requires data, the more the better.  But the sketches contain barely any useful information in this regard, by definition they are incomplete and hardly representative of a completed Beethoven work.  Obviously the team had to rely on Beethoven's completed works to get a true sense of the composer's style.  In that case, what is being "completed" here?  This isn't the first time that computer scientists have trained computers to create music in the style of a famous composer.  If this project had been presented as a reinvention/reincarnation of Beethoven via computer, it would be fine.  But claiming to have completed an unfinished symphony has more sizzle.  

The phrase "Beethoven's creative process" or something to that affect appears six times in the piece, but I never had any inkling of what it is supposed to mean.  Did they teach the computer to throw temper tantums and yell at its programmers?  Is there really a linear, programmable way of extrapolating a piecemeal unfinished product into a polished, performable work?  The descriptions provided in the article are vague.  Usually AI uses the finished products as the inputs for the algorithm.    

Dave Hurwitz made a number of good points in his recent video rant on this article. First, he notes that the sketches they used weren't necessarily written for a symphony.  They were just that, sketches that could have developed into anything.  Assuming they were the basis for a new symphony is an unprovable assumption that is essential to the viability of the entire AI project.  Second, the AI team had to attempt to reproduce Beethoven's orchestration -- how exactly can you try to orchestrate like a deaf person?  Late in his life, were Beethoven's orchestration choices a byproduct of his genius, or inexcusable mistakes on account of his deafness?  There's no way to know.  Trying to sort this out via algorithm is not much more than a shot in the dark.  

The three minute advance clip of the symphony certainly sounds like Beethoven, a bit too much like Beethoven in fact.  It comes across like a variation on the 5th symphony with a dash of the 8th symphony's lighter moments.  There isn't one iota of the fury and pugnacity that appeared in the first movement of the 9th symphony, for instance.  I find it difficult to believe, based on Beethoven's progressions in his later symphonies, that he would have attempted to go retro for his much anticipated 10th.