Musical Intelligences: Human and Artificial
Recently I happened to read two articles about music back to back—and they generated strands of thought which may inform one another.
It’s long been realized that individuals familiar with a musical idiom have some capacity to anticipate what’s going to happen next. If it’s a surprise, it may be a pleasant one, a harsh one, a boring one, and—though rarely—one that is totally discombobulating. We speak about the amount of information in the signal and how much new information is provided by the next tone or sequence. To use examples familiar to musical aficionados, this phenomenon may explain the delight heard at the first performance of George Gershwin’s Rhapsody in Blue, the anger expressed at Igor Stravinsky’s The Rite of Spring, and the range of reactions to the final movement of Beethoven’s Ninth (Chorale) Symphony.
In the first article I read, I learned about research conducted at Aarhus University in Denmark. The puzzle: Are we more likely to anticipate a change, when a musical fragment ends; or if we are to be surprised, must we await the appearance of the unexpected? How do we know that a musical phrase has ended and something new is about to begin? Can we recognize an ending before something new begins, so that we can start anticipating something new? Or, to shift metaphors, do we have to see the first scene of the second act to know whether it’s going to be “business as usual”.
The researchers invoke the concept of entropy. High entropy tones are unexpected; low entropy tones are expected. Their study confirms that in listening to music, the mind (and, of course, the brain) is one step ahead of the musical signal—it anticipates the next signal and notes whether or not it meets expectations.
As investigator Niels Chr. Hansen describes it, “We clearly see that people have a tendency to expect high-entropy tones (ones that are relatively unexpected) as musical phrases endings.” In other words, we experience melodies that end in an unexpected way as more complete than those that end in a low entropy expected way. Participants lingered for longer on high entropy tones… as if they were expecting something new to emerge after the perceived end of phrase.
In terms of “MI theory” one could say that individuals who are especially good at such anticipation are displaying or exhibiting musical intelligence. And we could further probe that capacity by exposing subjects to new styles of music and see how rapidly they can anticipate when a “high entropy” break in the expected pattern is about to occur.
The second article is far grander, even grandiose. Over the last two years, musical experts have accomplished a feat that eluded the greatest of classical musical composers—they have completed Beethoven’s 10thSymphony—and in fact, a full recording of Beethoven #10 was released on October 9, 2021. If you thought Beethoven had only completed 9 symphonies, you would have been literally correct. Before his death in 1827, Beethoven just left some musical sketches for the commissioned 10thsymphony. But as the article reports “Now, thanks to the work of a team of music historians, musicologists, computers and computer scientists, Beethoven’s vision will come to life”.
In their words: “In June 2019, the group gathered for a two-day workshop at Harvard’s music library. In a large room with a piano, blackboard and a stack of Beethoven sketchbooks, spanning most of his works. We talked about how fragments could be turned into a complete piece of music, and how AI could help solve this puzzle, while still remaining faithful to Beethoven’s processes and vision.”
How did AI solve this puzzle? The “AI” system needed to “learn” from Beethoven’s entire body of work how he might have approached and completed this final symphony. As described by the leader of the AI team “We would need to use notes and completed compositions from Beethoven’s entire body of work—along with the available sketches from the 10thsymphony—to create something that Beethoven might have written.”
After many efforts, they did a test—appropriately, in Bonn, Beethoven’s hometown. The team printed some AI-developed scores and played them for an audience on a piano. The audience was challenged to determine where Beethoven’s phrases ended and where the AI extrapolation begins. The audience failed the test… or, as we might quip, “AI passed the test!” This process was repeated several times. Over 18 months, the team constructed two entire movements, each lasting longer than 20 minutes. Now you can listen to this human-AI teams work and judge for yourself.
Note that these articles both highlight the capacity to synthesize. In the first instance, listeners put together—synthesize—what they have heard to this point and make their best guess about what is to come. In the second instance, humans and AI programs each review the earlier Beethoven musical corpus and put together—synthesize—a possible new work in the Beethovian style.
As this pair of articles illustrates, our understanding of musical intelligence is being enhanced thanks to research by psychologists, neuroscientists, AI experts, and—of course—musicians and musicologists. Perhaps, thanks to these scholarly undertakings, the musical intelligence of all human beings can be enhanced. We then have to hope that, as with the case of all intelligences, they are mobilized for positive ends… as did happen with Beethoven’s Tenth Symphony.
REFERENCES
The Brain Is a Prediction Machine, and Music Reveals How It Works - Neuroscience News. (2021). Retrieved 13 October 2021, from https://neurosciencenews.com/brain-prediction-music-19364/
Elgammal, A. (2021). How AI helped to complete Beethoven's 10th Symphony. Retrieved 13 October 2021, from https://www.straitstimes.com/opinion/how-ai-helped-to-complete-beethovens-10th-symphony
Hansen, N., Kragness, H., Vuust, P., Trainor, L., & Pearce, M. (2021). Predictive Uncertainty Underlies Auditory Boundary Perception. Psychological Science, 32(9), 1416-1425.
Photo by benjamin lehman on Unsplash