Team uses AI to complete Beethoven's unfinished masterpiece

A group of musicians and scientists have used artificial intelligence to complete one of the great what-ifs in classical music: Ludwig van Beethoven's unfinished 10th symphony.

Transcript

[POST-BROADCAST CORRECTION: In this story we incorrectly describe Ludwig van Beethoven as a 17th century composer. In fact, he composed in the late 18th and early 19th centuries.]

MICHEL MARTIN, HOST:

And finally, today, Ludwig van Beethoven.

(SOUNDBITE OF LUDWIG VAN BEETHOVEN'S "ODE TO JOY")

MARTIN: There is no overstating the German composer's influence on Western music, with enduring classics ranging from his epic "Ode To Joy"...

(SOUNDBITE OF LUDWIG VAN BEETHOVEN'S "ODE TO JOY")

MARTIN: ...To his heroic "Third Symphony"...

(SOUNDBITE OF LUDWIG VAN BEETHOVEN'S "SYMPHONY NO. 3")

MARTIN: ...And the fantastical "Moonlight Sonata."

(SOUNDBITE OF LUDWIG VAN BEETHOVEN'S "MOONLIGHT SONATA")

MARTIN: But for all the sonatas, concertos and symphonies Beethoven wrote, there remained one piece he never completed. It's one of the great what-ifs in classical music - his unfinished "Symphony No. 10."

WALTER WERZOWA: He wanted to dedicate his life to his 10th Symphony, but he passed away before he really got knee deep into it.

MARTIN: Walter Werzowa is a composer in Vienna who studied Beethoven for years. In early 2019, he and a group of Beethoven fanatics undertook a challenge - take the rough sketches of Beethoven's 10th Symphony and complete it.

WERZOWA: Maybe we're a little bit complacent in the beginning to think that we can do that in half a year what Beethoven would do in many years.

MARTIN: But Werzowa and his team had a bit of help.

AHMED ELGAMMAL: My name is Ahmed Elgammal. I'm a professor at Rutgers University of computer science and director of the Art and Artificial Intelligence Laboratory.

MARTIN: The plan - use artificial intelligence to finish a masterpiece. Here's a taste of what the team came up with.

(SOUNDBITE OF AI-GENERATED MUSIC)

ELGAMMAL: The way AI generates music or learn about music is very similar to the way right now - you're writing an email, for example. The email try to guess what is the next word for you. So in music, it trying to teach the AI how to predict what's the next note. And by looking at lots and lots of music, the AI fundamentally can learn how to do that.

MARTIN: Elgammal says they start by teaching this AI student on a broad range of classical music - Bach, Mozart, Haydn anything that would have been available to a 17th century Ludwig van Beethoven.

(SOUNDBITE OF MUSIC)

MARTIN: And once they've done that...

ELGAMMAL: Now, we can take that student and within the next level, which is now give the student access to all Beethoven works and now teaching the students how Beethoven would compose. And because music is very structured and very mathematical, AI can really excel at that. AI can really capture how Beethoven would continue or develop on a theme. Like, when you hear the "Fifth Symphony," you hear these four notes in the beginning - (vocalizing).

(SOUNDBITE OF LUDWIG VAN BEETHOVEN'S "SYMPHONY NO. 5")

ELGAMMAL: And you see how Beethoven take this and develop it to a whole movement. And now the job of AI is really to learn how composers, in particular Beethoven, would take a theme and develop it and try to do that in a way that you can measure how good is this prediction.

MARTIN: In November 2019, the team convened in Germany with a live audience to present an early test.

ELGAMMAL: We have a full string quartet playing one of these pieces on the stage with many people in the room and simply try to guess where the sketch ends and where the extrapolation by the AI happens.

(SOUNDBITE OF FULL STRING QUARTET)

ELGAMMAL: Most people couldn't tell where the sketches stopped and when the AI start and that proved our point that AI can do something here.

(APPLAUSE)

MARTIN: It was classical music, but not everyone in the audience said it sounded like Beethoven. So they fed it more Beethoven, tinkering with this movement, developing that theme, working hand-in-hand with the maestro inside a machine to complete the unfinished work. Composer Walter Werzowa.

WERZOWA: And so what we distill and analyze out of his work is really the essence of Beethoven.

(SOUNDBITE OF BEETHOVEN ORCHESTRA BONN'S "BEETHOVEN X: THE AI PROJECT: III SCHERZO. ALLEGRO - TRIO")

WERZOWA: There were a couple of naysayers saying, AI was never lovesick and AI never was drunk and AI didn't turn deaf.

ELGAMMAL: I always expect that skepticism and pushback from many people who think of art as sacred, and AI has no room in human creativity. I believe and art is the human. And we have to always remember also that AI itself is a human - outcome of the human creativity. So AI code itself is a piece of art.

MARTIN: Next Saturday, more than two years after beginning this project, the team will finally release Beethoven's completed "10th Symphony" to the world.

WERZOWA: This was such an intimate process. It felt many times he was sitting next to me and that we talk to each other and enjoyed each other. I hope he did enjoy me as much as I enjoyed him.

(SOUNDBITE OF BEETHOVEN ORCHESTRA BONN'S "BEETHOVEN X: THE AI PROJECT: III SCHERZO. ALLEGRO - TRIO")

MARTIN: That was Walter Werzowa and Ahmed Elgammal who both helped complete Beethoven's "10th Symphony" using artificial intelligence.

(SOUNDBITE OF BEETHOVEN ORCHESTRA BONN'S "BEETHOVEN X: THE AI PROJECT: III SCHERZO. ALLEGRO - TRIO")

Copyright © 2021 NPR. All rights reserved. Visit our website terms of use and permissions pages at www.npr.org for further information.

NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.