I fed all of Beethoven into a computer and told it to compose

I fed all of Beethoven into a computer and told it to compose


norman lebrecht

September 24, 2021

Professor Ahmed Elgammal, Director of the Art & AI Lab, Rutgers University, explains how he created a version of Beethoven’s partly sketched tenth symphony using cutting-edge Artificial Intelligence.

I presided over the artificial intelligence side of the project, leading a group of scientists at the creative AI startup Playform AI that taught a machine both Beethoven’s entire body of work and his creative process.

A full recording of Beethoven’s 10th Symphony is set to be released on Oct. 9, 2021, the same day as the world premiere performance scheduled to take place in Bonn, Germany – the culmination of a two-year-plus effort….

The to-do list grew: We had to teach the AI how to take a melodic line and harmonize it. The AI needed to learn how to bridge two sections of music together. And we realized the AI had to be able to compose a coda, which is a segment that brings a section of a piece of music to its conclusion.

Finally, once we had a full composition, the AI was going to have to figure out how to orchestrate it, which involves assigning different instruments for different parts.

And it had to pull off these tasks in the way Beethoven might do so…

Read the full article here.


  • John Borstlap says:

    Yes that’s how I do it too: you don’t need to compose yourself if you can find another, easier way. I bought a very sophisticated contraption in Japan when there on holiday years ago and fed it all the scores I had myself plus anything I could find in the music department of the public library. The system worked better than any AI could do, because through a unique invention by Dr Hosoki Osokawa Kijani the machine could really invent new music, not just imitate, on the basis of what it was fed. It took all the scores as examples and learned from them, but made something new out of the material. Performers were very happy since it souded really great and was even expressive. It made an elderly lady weep in 2012. After 8 years, when the printing part got a bit rusty, I opened it at the back and found a short Japanese man in the machine, who – after some physical pressure – admitted he wrote all the stuff himself. We agreed to tell nobody and in this way we worked together for quite some time, until his wife came at the door and claimed him back. But there is enough for the next 20 years so I no logner need the system.

    I can recommend any budding composer to rely on technology to compose music to compensate for the awful times we live in which makes it so hard to write masterworks.

  • Freewheeler says:

    I hate this sort of thing. And I majored in A.I. (and Computer Science). Reminds me of that Honda robot that could “play” the violin. Some folk were impressed, but not I.

  • J Barcelo says:

    BS. There’s a real simple way to test this: give the computer only a few bits and pieces of say the Ninth and see if it comes up with anything remotely like the work as completed by Beethoven. Of course it won’t.

  • Frank Flambeau says:

    The computer is programmed by humans but not by Beethoven.

  • Deazely says:

    Pray God this doesn’t work otherwise we will get Mahler’s Symphony No 12 and then 13, 14 usw. When the confected Frank Sinatra song came out, I thought this would happen.

  • Paulo says:

    Two words: completely pointless.

  • Jim says:

    Well, one comment before I go and read the article:
    One thing that Beethoven was *very* good at was not composing in the style of Beethoven. Just about everything was thought out from scratch; that’s why you can seldom mix and match movements from different works (ever thought of tacking the pastoral finale onto the eighth?)

    • John Borstlap says:

      Very true.

      With every piece he tried to forget what he had written before.

      All the great works have a style and a mood of their own, they come from a specific and unique musical vision.

      This idiotic idea of AI is the result of thinking that only matter exist, and that therefore music consists exclusively of matter. But this is not so. The same mistake is made with thought experiments where one thinks that, if all the information of a human being, including all of his memory, is transferred to a super computer, the person will be transferred as well.

  • sam says:

    No one could ever predict Beethoven’s 9th from knowing the first 8 of his symphonies. That is the essence of genius, to create the unexpected.

    That is why no algorithm, no AI, can capture genius, because it requires a production that defies and breaks previous patterns in a wholly unexpected way.

    And plus, there is no artistic or economic value to a new Beethoven symphony (or a new Picasso), because humans also value authenticity.

    So, if artistic AI programmers were smart, they would secretly create a program to produce the next bestseller or blockbuster movie or must-binge TV series, and put their own name to it. Much more lucrative.

  • Nick says:

    Obviously Ahmed Elgammal has nothing better to do and Rutgers can afford to waste tons of money that could have been much better spent helping out needy students then feeding stupid, wasteful and irrelevant projects like Ahmed’s! SHAME on RUTGERS!

  • christopher storey says:

    It would be very interesting to give the AI machine the material that Anthony Payne had for the Elgar 3rd Symphony, and see how what it comes up with compares with Payne’s wonderful realisation

  • HugoPreuss says:

    There have been several efforts to “recreate” the lost St. Mark’s Passion, using the models of the St. Matthew and the St. John, to do it almost exactly like Bach certainly or almost certainly or perhaps would have done it.

    And I am grateful for these attempts bc they demonstrate painfully that “almost” won’t do, and that no one can recreate Bach. Something is missing. The stroke of genius, perhaps?

    Still, good luck to the professor and to his (inane) scheme…

  • IP says:

    Incidentally, have you heard Schumann’s song cycle “The Body-with-Vagina’s Love and Life”?

    More here: https://www.telegraph.co.uk/news/2021/09/25/lancet-receives-complaints-scientists-quit-sexist-cover-calling/

  • BrianB says:

    The obvious acid test would have been to feed the computer all of Beethoven minus, say, the 5th Symphony; then give it no more than a comparable amount of sketches for that work, and see what it came up with. I’d be amazed if the end result would prove to be anywhere close to the symphony we know.

    • Jim says:

      Yes, an interesting idea to try. But, just imagine if they did that and the result turned out better than Beethoven’s attempt.
      Inconceivable? The inconceivable is always so until it is conceived.
      (and, yes, I used the word ‘attempt’ provocatively)

  • Byrwec Ellison says:

    Prof. Elgammal’s claim that “Beethoven’s vision will come to life” is probably overstating it more than a bit (that’s a computer joke), but the value of his exercise is monumental for AI itself. Breaking down human thinking process into a series of manageable bites (oops, did it again) lets his machine mimic individual steps to set a path along thousands or hundreds of thousands of decision points. From those, the edifice is built brick by brick. It may not exhibit that inexplicable creative “spark,” and the result or “vision” won’t exactly be Beethoven, but it’ll mimic the various ways he elaborated on musical phrases.

    The embedded audio clip of the Tenth Symphony excerpt does sound like Beethoven spinning musical threads together (there’s more than a hint of Fifth Symphony in it, which is probably a function of the musical fragments he left behind). It would have been a more interesting test to take his sketches from a known work — like, say, his “Eroica” Symphony — and let the machine create a whole new “Eroica” so we could compare the two side by side. On the other hand, that’s a challenge best left for another day when AI technology is more evolved.

  • We privatize your value says:

    The result is a very backward-looking Beethoven, who sounds more like Mozart and Haydn than he did in his 8th symphony! At least from the snippet we can hear. I wonder if the finale of Bruckner’s 9th, as soon reconstructed by AI, will sound more conservative than the human reconstructions, too.

  • Philip says:

    The biggest issue is that the AI process apparently did not take into account Beethoven’s artistic development from the first to the ninth. Simply feeding in all of his output ignores this important creative issue which is that the Beethoven who composed the first was different than the Beethoven who composed the ninth.

  • rob_h says:

    Does this mean they didn’t teach the computer how to develop musical ideas?

  • Lance B Brady. says:

    Make a machine which is able to create people.The real way is not making a quality product anymore.

  • Composer David Cope spent close to 40 years at UC Santa Cruz developing algorithms to imitate musical styles. Between 1991 and 2012 he wrote nine books on the topic.

    There is value in this sort of work for music theorists in terms of better understanding the compositional techniques and stylistic idiosyncrasies of composers–a form of statistical analysis if nothing else. It’s similar to the use of computer models in understanding many types of natural phenomenon such as traffic flow, population growth, earthquakes, avalanches, plant diversity, and countless other things. Even variations in bird songs. ,mn

    Probably the closest correlation to this sort of work is the neural machine translation algorithms of Google Translate, though I’m guessing the program used for the Beethoven is more related to earlier Google programs that used the less complex algorithms like statistical machine translation.

    There’s something particularly Western about this approach to music, since from the days of Pythagoras we have correlated many aspects of our art with mathematics.

    (I suppose one could even create algorithms that could effectively imitate the reactionary, anti-intellectual bluster common among many SD respondents. On the other hand, it’s so predictable we hardly need any further models to understand it…)

    • John Borstlap says:

      That’s true, I can smell reactionary blustering from a thousand feet away here. I hate this constant protesting against something pure & rational, there’s nothing so calming as that pure sphere of rational thought! As we hear in, for instance, some of the plis in pli selon pli. (By the way I recently read that this beautiful piece was originally meant for a ballet, hence the name, must have been a very quiet & reassuring one)


      • Sally is actually a computer generating posts to set the world straight about modern music. She lacks slightly in tolerance for other views and thinks her way is the only way. Quite forgivable. She’s only an algorithm.