For The Record: Artificial Intelligence in Music?

For as long as anyone can remember people have been criticizing popular songs for being formulaic. Is there actually a mathematical formula that can determine hit songs? Can a computer generate a “perfect hit” on its own? If a computer can create a hit song, what are artists meant to do?

A website called UPlaya, produced by Music Intelligence Solutions, has developed a scoring guide in which a computer judges a song based on its harmony, chord progression, length, lyrics, tempo, etc . . . and gives it a rating out of ten. The higher the rating, the more likely that song is to be a hit. “I Gotta Feeling” by the Black Eyed Peas, for example, got a score of 8.9. Researchers at Bristol University have also developed a song scoring system based on danceability, tempo, lyrics, and another important factor: current trends. The Bristol team analyzed the Top 40 Singles charts for the last fifty years and found that the algorithm for a hit has to change over time. In the eighties for example, danceability was a huge factor. During the nineties, however, songs became slower, more ballads than disco. The Bristol University team released a website: scoreahit.com, which explains in more detail the math behind their system. Another website: musicxray.com, allows artists to score their own songs. Researchers from Emory University have taken a different approach by using MRI scans of people’s brains while they are listening to a song to predict hit potential, but this method is too expensive to be widely used. 

While these websites have got the process down as far as typical hit songs go they will never be able to predict hits with complete accuracy. Many songs become hits because of unique, catchy qualities that a computer system would classify as outside the anatomy of a hit. People appreciate daring in an artist, and often listen because they are excited to hear something that they did not expect. The danger of determining song releases solely by means of software would be that it teaches people exactly what to expect every time, and music would lose all of its energy. For this reason, hit predicting software will never completely rule music. While it may become commonplace in a corporate music setting, there will always be artists pushing the boundaries of normality and safety in music.

With the ability to mathematically predict the popularity of songs, it only makes sense that somebody wanted to see if a computer could do the work for them. This idea really became widespread in 2021 with the release of an AI-generated Nirvana song titled “Drowned in the Sun.” The song was made using a Google AI program called Magenta, which analyzed the structure and lyrics of Nirvana songs, found recurring components, and stitched pieces of these together for a “new” track. This effort does have a good idea behind it: to pay tribute to Kurt Cobain and to try and produce the music that Kurt would be making were he still alive. The result, however, is far from the goal. The song does a good job of recreating Kurt’s voice as well as the instrumental styles of all the members, but if you are someone who has listened to a lot of Nirvana, you will undoubtedly notice that all the AI has done is slightly alter a riff from a popular Nirvana song, taken a riff up a few keys, played a section backwards, etc. As far as lyrics go, the song is a mess of different words commonly used in actual Nirvana songs. The lyrics are ridiculous. “I’ve got my hands right now, in every wound,” for example, is one of the songs most touching, heart wrenching, meaningful lines. 

Since “Drowned in the Sun” companies have been trying AI software with other bands with the same results. AI Beatles, Metallica, Nickelback, Red Hot Chili Peppers, etc . . . songs are in no way a substitute for the actual band. What they can be, though, is really funny. A group named Botnik, for example, has generated a Morrisey song titled “Bored With This Desire To Get Ripped” by feeding a computer a mixture of Morrisey lyrics and workout dvd reviews. For any Smiths fans, you might notice that the opening riff sounds exactly like “This Charming Man” with a couple notes changed. It also features a perfectly terrible music video.

All this said, the position of actual artists in the music industry is not in jeopardy for the foreseeable future. Unless the AI bots get exponentially better or learn how to use English verb tenses consistently, computer generated songs will not be filling the charts anytime soon. As far as computer hit predictors go, they will not be significantly changing the music industry. As can be seen in the past ten years of popular music, humans have essentially figured out the formula for a perfect song without asking a computer. The biggest effect hit predictors will have is to help artists choose what songs off of an album will be the most successful as singles.