Why Music Should be a Happy Accident, Not an AI App

I like finding new music and bands by happenstance. Since I’m not a complete Luddite, I even use social media like Instagram to find new music, although I must admit the images I mostly see are of bebop jazz albums, not suggested playlists.

Two recent experiences, two happy accidents, really, led me to bands and music I’d never heard before. And now I thoroughly enjoy the sounds of each band; one still creates music and the other long gone.

No alt text provided for this image

While you may or may not enjoy finding new music–whether it’s jazz, rock, pop, hip-hop, classical or rap–in this old-fashioned way, it may not matter in the long run. Because researchers are looking at ways artificial intelligence can “create” the perfect song. At some point, you may need to do nothing more than enter a few “likes” into an app, which will spit out your new favorite song in a matter of seconds.

Back to the Music

I recently discovered two bands, new favorites, not via an app, but through day-to-day life experiences.

That’s a long-winded, professorial, English-paper way of saying: We figured out what your brain wants to hear, and we can recreate it with AI.

I play indoor soccer and while we’re running across the artificial turf, music blasts from the large speakers in the warehouse in which we have games. For a long while I was intrigued by a punk sound laced with country and western, a slide guitar, four-part harmonies and the distinctive, plaintive voice of the group’s lead singer. If I hadn’t played indoor soccer, I wouldn’t have discovered Bad Religion. (Admittedly, I’m late to the Bad Religion bandwagon. They played their first songs in 1980 and continue today.) Now, I can’t get enough of them. Their sound, passion and lyrics are on-target for the world in which we live. They just have a great sound.

No alt text provided for this image

My second encounter is with an album first pressed in 2000 of a group of songs recorded in 1995 by Crestfallen, the Denver-based punk/hardcore band active in the mid- to late-1990s.

Although living in the Denver-area since the late 1990s, I hadn’t heard of, nor listened to Crestfallen. I still don’t know much about them, but I am now a fan.

I am a thrift store habitue. And it was on one of those jaunts, pawing through piles of records, that I came across the eponymous LP. For $1.99 I decided to take a chance; the record looked unplayed and especially because it was pressed on baby blue colored vinyl. So, I washed the record, cleaned it a second time and placed it on my Sony turntable. Crestfallen had a great sound punctuated by a lead singer with a voice instilled with longing. It’s a great record.

Researchers have long acknowledged music can affect emotions. “Music is universally enjoyed perhaps because of the power it has to move us. Listening to music can boost our mood, give us chills, and even make us cry,” write the authors of “A Multimodal View into Music’s Effect on Human Neural, Physiological, and Emotional Experience,” who are looking into the ways these emotions could be harnessed and exploited by AI.

Which brings me to AI’s foray into music. Could AI “make” this music for me and you. Yes, probably, sometime in the future. Will discovering that music be as fun. Most definitely not. (AI has already made “art.” One piece sold for more than $400,000.)

Will AI Write Your Next Favorite Song?

It might. If the programmers who write AI algorithms can harness the perfect intersection of a mood-boosting, chill-giving, cry-inducing song, why not. I mean why not get emotional about every song you hear?

No alt text provided for this image

Researchers used a variety of instrumental-only songs to better understand “happy” and “sad” songs. In their graph to the right, a sad song is plotted by its “skin conductance response” or SCR. The researchers wanted to find out if the participants’ skin conducted electricity better when, in this case, he or she heard a sad song.

You can see in the graph there were several spikes indicating better electrical conductivity, which occurred when the subjects heard a violin, a cello and began to experience the song’s crescendo.

Now plotting how your skin responds to happy and sad songs might take all the fun out finding a new band or singer, but at least the new process will be fast and perfect! I suppose it is more efficacious to plug yourself into into a computer a la The Matrix, listen to few songs and have them deconstructed by AI to get the perfect song. But I enjoy the search.

No alt text provided for this image

The researchers, on the other hand, are pretty excited by their new work.

“By computationally analyzing how music influences these three modes (neural, psychological, emotional), we present a more complete picture of music’s role in human experience.

“We apply a set of multivariate time series (MTS) prediction models to neural, physiological, and subjective responses, using auditory features as predictors. We compare these models and comment on what auditory features are important for these predictions.

“We hypothesized that attention models, which non linearly synthesize aural information from previous points in a song, would be best at predicting these responses,” the authors theorize.

That’s a long-winded, professorial, English-paper way of saying: We figured out what your brain wants to hear, and we can probably recreate it with AI.

No alt text provided for this image

Or as the author at the MIT Technology Review wrote of the study: “The researchers then fed the data, along with 74 features for each song (such as its pitch, rhythm, harmony, dynamics, and timbre), into several machine-learning algorithms and examined which features were the strongest predictors of (emotional and physiological) responses.”

I could hook myself up, listen to a handful of songs to figure out what my “perfect” song is. I could monitor my galvanic response to find out what makes my skin tingle. I could input that information into an algorithm, which would spit out a brand new, never-before-heard song at the other end. A song perfect for me and me alone.

But that removes the entire experience of discovering a new band by accident or being introduced to a new song by a friend.

It’s just too perfect.

And not as fun.