Deep thought 💭 on Apple Music's algorithmic steerings

My house is filled with music floating through many of the rooms for hours each day. Making dinner, entertaining or relaxing without the TV are all accompanied by a soundtrack. The music starts by yelling into the air, “Hey Siri, play 80s music.” The first songs are emblematic of the requested music, but soon Apple Music will wander. The wandering songs are still in the requested parameter (80’s music, in my example), they tend to get less popular songs within the genre, something like B-sides or obscure bands with sonically adjacent songs. Soon enough, the musical path takes one of two paths: obscure songs that stay in the genre or extended versions of popular songs.

This weekend, the HomePod played a version of “Purple Rain” that felt like it was 20 minutes long. Long enough to have noticed when the song started, do kitchen tasks for several minutes and realize that the same song was playing, followed by several more minutes of work and a further realization “damn, is this still Purple Rain?”

Here is my hypothesis—Apple knows that HomePod owners use the device to have music playing in the background for hours. Song royalties add up after hours of play, particularly based on which catalogue or library they come from. Apple Music engineers steer the algorithm to less costly songs, or when playing popular songs, the algorithm selects the longest version possible. Longer songs for a single royalty payment cost less in totality than a several shorter popular songs.

I find it interesting that Apple Music users are being steered to a compromise position, in which we get mostly the music we want while also trying to keep costs down. It is seamless, almost undetectable, and insidious.

Published: Apr 11, 2023 @jeredb →