The algorithm failed music

"I used to have this ritual. Every Tuesday, I would get off the train at 8th Street on my way home from work. I would pop into Other Music, buy a new CD (or three...) and then walk the rest of the way to the Staten Island Ferry listening to my new CD. Even if there wasn't a new record out that week that I was looking forward to, I would buy something."

"That started to change in the 2000s with the advent of the first algorithmic recommendation engines. Pandora was the big pioneer there with its Music Genome project. The aim was to break songs down to easily quantifiable traits such as "gender of lead vocalist, level of distortion on the electric guitar, type of background vocals," and the like. Then it would look for other songs that shared a certain number of traits in common and play that track."

A personal ritual involved buying CDs at Other Music on Tuesdays and listening to the new purchases on the Staten Island Ferry. Until the 2010s, most people discovered new music by browsing record stores, recommendations from friends or older siblings, and curated mix CDs. The advent of algorithmic recommendation engines in the 2000s began to change discovery patterns. Pandora's Music Genome project analyzed songs by quantifiable traits like gender of lead vocalist, distortion level, and types of background vocals. The algorithm matched songs sharing sets of traits to generate playlists. Early algorithmic recommendations gained popularity but also tended to repeat the same small set of songs frequently.

Read at The Verge