New Music-Focused AI Can Guess Your Tastes & Habits

A new AI program created by computer scientist Bruce Ferwerda of Switzerland’s Jönköping University and marketer Mark Graus from The Netherlands’ Maastricht University is supposedly able to figure out a person’s “musical sophistication” by checking out their Spotify. The AI looks to a number of metrics concerning a person’s music listening habits to determine their musical sophistication score, most notably what tracks they listen to and how often they do so. Based on this, the AI can tell how varied a person’s tastes are and even how likely they are to engage in “musical” behaviors, such as listening to a wide variety of different music or even exhibiting musical talent themselves.

The study that helped create and train the program centered on 61 participants. Between all of their top tracks on Spotify, the AI was given a list of 21,080 pieces of music to analyze. The tracks were analyzed using various Spotify API plugins, looking for things like valence, beats per minute, danceability, and how much of the tracks were instrumental, among other metrics.The participants were then surveyed about their music tastes and habits to create a profile for each of them that detailed things like how much they spent on music, from albums to instruments, along with their emotional responses to music and how often they could be found listening. All of this data was enough to classify the participants and create a scale of “musical sophistication”. The higher somebody’s score was, the more likely they were to be deeply involved in music. The most sophisticated individuals were those with extremely wide-ranging tastes who created music as well as consumed it.

The practical use cases of such an AI are many; it could be used to vet music school students, to help cater music playlists and discovery algorithms on music services for listeners, and even to help train the presumably more human-like AI of the future in how they should “feel” about music, just to name a few possible uses. As to real-world examples, Google already has a musical AI project known as Magenta, which serves as a baseline for music-focused projects to build from. Something like this could greatly expand the knowledge base and training data set of an AI like Magenta.