How Beethoven Would Rage at the AI That Dares Finish Him Off



“How dare you?” he would thunder, stalking onstage, knocking down a few music stands along the way. The calm, self-congratulatory performance of his Symphony No. 10, finished this fall with the help of AI, would halt in jangled discord. Resurrected Beethoven would then settle his wild locks, raise the baton, and conduct an entirely different work. And Ahmed Elgammal, director of the Art & AI Lab at Rutgers University, would hide backstage.

Two years ago, Elgammal promised the Karajan Institute he would help fill in the blanks, so they could complete No. 10 in celebration of Beethoven’s 250th birthday. All the musicians had were some musical notes and a handful of impulsively jotted ideas. But soon Elgammal’s team had fed Beethoven’s entire body of work—and the best possible approximation of his creative process—into AI.

Beethoven should take this as a compliment. For 194 years, we longed for the ending his death stole. On the other hand, this is the man who dedicated his third symphony to Napoleon, then, upon learning that the man had declared himself emperor, scratched out the dedication in a rage. Beethoven threw objects at servants and spat without heeding the trajectory, and his sudden deafness only worsened his temper. His “Rondo a Capriccio” is nicknamed “Rage Over a Lost Penny,” because its energy channeled his fury with a maid he suspected of stealing a gold penny.

In his Opus 31, the quartet Beethoven considered his most perfect single work, he made our brains beg for a particular chord he withheld until the very end. The closure comes as relief and blessing, and it must have been a delight to ink in that final stack of notes. But would he feel the same about a completion he did not orchestrate?

It was at least done in good faith. One of the music experts on the team said the AI reminded him of an eager music student who practices every day and becomes better and better. When the completed Tenth was played, only those with intimate knowledge of Beethoven’s preliminary sketches could mark the transition to AI extrapolation. Granted, no one could say whether Beethoven would have written the symphony this way. . . .

That does not stop us; the urge for completion, as he well knew, is stronger than caution. The team that finished his tenth included Harvard musicologist Robert Levin, who had previously finished—but without the help of AI—a Mozart requiem and several movements to Bach cantatas. Schubert’s Unfinished Symphony has driven people crazy for more than a century—he had not died! Why not finish? Did he associate the music with his first outbreak of syphilis? Was he distracted by his Wanderer Fantasy? He wrote the first two movements and the first two pages of a scherzo, and the rest of the scherzo was found in partial form after his death.

On the centennial of Schubert’s death, Columbia Records held a worldwide competition to write the fourth movement, but the results were a disappointing hodgepodge. Then, just before the Beethoven project began, the Chinese tech company Huawei taught its Mate 20 Pro smartphone to analyze the Schubert and suggest finales. Composer Lucas Cantor then orchestrated his favorite. Working with AI was, he enthused, “like having a collaborator who never gets tired, never runs out of ideas.”

The critics were less impressed. Goetz Richter called the fourth movement trivial, attaining only “a loose and inauthentic family resemblance to Schubert.” Maybe AI needs to try harder. With Bruckner’s unfinished Symphony No. 9, perhaps? “I know of only one composer who measures up to Beethoven,” Richard Wagner said, “and that is Bruckner.” For years, it was thought that he had made it through the adagio, which was so eloquently sad, he pronounced it a farewell to life—then dropped dead. Then the truth emerged: he had finished sketches for the finale as well, and his clueless mourners had taken bits of the score home with them as memento mori. We were left with Schubert’s meticulous record of how many Hail Marys and Our Fathers he said each evening as he composed (could they not have taken that instead, as their memento?) and only three movements of a four-movement masterwork.

Mahler’s Tenth was finished by Ali Nikrang and an AI algorithm and performed at the 2019 Ars Electronica Festival. The German conductor Markus Poschner said of the result, “The technical level is astounding of course. I wouldn’t have thought such a thing possible.” That said, “we must not forget that notes are always merely codes or projection surfaces that we musicians have to fill with meaning.” Confronted with AI’s work and asked to respond to it, Poschner said, “we immediately feel a great uncertainty: are we allowed to feel anything? And if so, then what? Can the work of art tell us something, communicate something?”

We finish one another’s projects all the time. Kareem Abdul Jabbar does a brilliant job with Sherlock Holmes. John F. Callahan finished Ralph Ellison’s Juneteenth. John LeCarre’s son, like Hemingway’s, finished a manuscript for his dad. Bela Lugosi’s dentist cloaked himself in a cape to finish his role in Plan 9 from Outer Space. We are still waiting for someone to finish Gaudi’s Sagrada Familia cathedral. If we want a computer to analyze or enhance artistic work, the way Google’s AI Doodle let us turn rock ’n ’roll into contrapuntal Bach harmony, we can learn from the effort.

AI can teach us about patterns and processes, theory and extrapolation, math and aesthetics. But what it communicates will not be sensient, self-aware, rooted in profound emotional experience, or conscious of its audience. Listening to Beethoven’s Symphony No. 10 “finished” with AI is a different experience than hearing music wrought by one man’s genius, arrogance, bipolar moods, cursed deafness, secret love affair, deep joy, caustic irritability, and hard-earned serenity. Would I, who love his work without much musical understanding of its intricacies, even recognize the difference? Probably not. AI music could soon be the deepest fake of all.


Read more by Jeannette Cooperman here.