Like the movies, popular music is both an influencer and an indicator of public opinion. Artists can use music to apply pressure to a political regime, and to explicitly communicate a feeling to a large group of people. A good example is Jimi Hendrix’ performance of The Star Spangled Banner at Woodstock in 1969, which expressed a complex combination patriotism and protest simultaneously. No words alone could communicate the same message.

It sounds like popular music is an opportunity to influence public opinion, but there is a problem. Playback on contemporary hit radio stations is the largest factor in determining which songs become hits. Record labels have a huge amount of power determining which songs get played on the radio. Songs that pop labels produce are not written by artists. They are written by songwriting teams, and engineered for mass appeal. It is difficult for an independent artists to create music that competes with the record label song writing machines. Childish Gambino is the exception, not the rule. This is where the geniuses in Silicon Valley will propose:

We’ll use machine learning identify the the musical qualities of the hits that Max Martin wrote for the Backstreet Boys, Britney Spears, Pink, Avril Lavigne, Usher, Jessie J, Katy Perry, Taylor Swift, Ariana Grande, and many others. Then we can “disrupt” the business of writing songs, and “democratize” the process of making hits!

This is not a technology that I’ve worked on directly, but it is almost certainly a project that making its way into production at this moment. The problem: How can we insert greater diversity sounds and perspectives into popular music? While pop artist Ke$ha was signed to a label, her producers had commercial success selling her as a sexualized party girl with songs like Tik Tok and Right Round. But Ke$ha didn’t want to that to be her image. She wanted to remake pop music with a message and with her own voice, and this presents challenge. In the words of songwriter and lyricist Bonnie McKee:

“People like hearing songs that sound like something they’ve heard before, that’s reminiscent of their childhood, and of what their parents listened to. I mean, every once in a while something new will happen, like dubstep, where it’s like, ‘This is robot future music!,’ but most people still just want to hear about love and partying.” (Seabrook, J.. The Song Machine, 2017, Chapter 21)

Is it possible to write a song that is entertaining as it is informative — or as popular as it is progressive? In Ke$ha’s newer songs you can hear a tones of the 60’s rock and roll, and they espouse positive and progressive messaging, writing about abuse, depression, and heartbreak, and marriage equality. Despite the backing of Sony owned RCA Records, singles on her new album did not get achieve the commercial success that her earlier songs did.

Her story illustrates a larger picture. Female artists are sexualized and exploited, and the labels and executives profit handsomely. No one said it more simply than artist Sinéad O’Connor in a 2013 open letter to Miley Cyrus after Cyrus’ performance at the American Musical Awards ceremony. Popular music is already produced be a mechanical and formulaic process. As machine learning optimizes pop to capture our collective attention, how artists stay competitive?

Who is positioned to address the the problem? Is it the artists? The songwriters? The producers? Record Labels? A consortium of all of them? No one can do it alone. Stories of sexism in the entertainment industry, and the lack of women is leadership roles,  suggest that women may be best equipped to lead the industry toward solutions. However, we cannot but the burden of making progress on women alone. The fight against the industrial song machine will not be one easily… especially if they are the ones with the resources to tap machine learning for the purposes of crafting the perfect musical hook. The obvious problem with using machine learning to write songs, is the same as the using machine learning to identify people: Inheriting bias, and perpetuating inequality. These are our problems, and machine cannot learn the solutions for us. Norbert Wiener probably said it best in his 1950 book The Human Use of Human Beings:

“Any machine constructed for the purpose of making decisions, if it does not possess the power of learning, will be completely literal-minded. Woe to us if we let it decide our conduct, unless we have previously examined the laws of its action, and know fully that its conduct will be carried out on principles acceptable to us!”