'The 100 Days of Colour' Paper
This is a condensed version of the full reflection on ‘The 100 Days of Colour’ which will one day be published as a book. This paper details how I did the experiment, what the results were, and key things I learned from this experience. For more details on what this entails, see ‘Additional Info’.
Price in Canadian $.
This is a condensed version of the full reflection on ‘The 100 Days of Colour’ which will one day be published as a book. This paper details how I did the experiment, what the results were, and key things I learned from this experience. For more details on what this entails, see ‘Additional Info’.
Price in Canadian $.
This is a condensed version of the full reflection on ‘The 100 Days of Colour’ which will one day be published as a book. This paper details how I did the experiment, what the results were, and key things I learned from this experience. For more details on what this entails, see ‘Additional Info’.
Price in Canadian $.
Paper Abstract
After discovering that over 95% of the music I listened to and was recommended by the Spotify program was made by white artists, I sought to identify whether this bias was unique to me, or was there something in the programming itself that contributed. If I was being recommended more white artists than black through unconscious habit, could I get the trend to flip to recommend more black artists than white? If I could facilitate the flip, then this would show that it is up to users to facilitate equitable representation online. If I could not, then there is something in the programming itself preventing the change.
For 100 days, I stopped following white artists, listening to them only if they were featured in a song by someone else. Every day, I listened to the discographies of new artists to increase the amount of “colourful” input data. To monitor the program’s reaction to my change in input, I recorded the racial demographic of eight core Spotify playlists designed to be unique to the user based on their history: the six Daily Mixes, Release Radar, and Discover Weekly. After 100 days of controlled input, I recorded the data again on Day 200, to see how the program reflected my new habits.
Combining my observations with my understanding of Spotify’s systems, music, and psychology, I was able to make a prediction as to the demographic of the Spotify team, which was correct when compared to the employee ethnicity data for the United States provided on the Spotify website. This is evidence that our technology is not exempt from human bias. In regards to my initial intention of the flip, I was able to facilitate the shift, but it took a long time and there were certain playlists that were less responsive. These findings show that all humans involved in a technology - the programmers and the users - have a responsibility to ensure equitable representation online.
Because of this shared responsibility, we need to integrate policies and tools on both sides. From the programmer/producer side, I suggest third party audits where a neutral body reflecting various perspectives can analyse the system, check for biases, and help come up with targeted solutions. General economic and educational elevation of those groups whose voices are not represented is also crucial to ensure their input is included, without being susceptible to accusations of tokenism. Users should be empowered to give feedback to the algorithms, including having tools to steer recommendation algorithms, and the education and reminders to use those tools.
My third recommendation and goal is to create an add-on or program that uses the patterns and teachings from ‘The 100 Days of Colour’ to analyse representation in a data set or online content to highlight where bias arises. This could be used on both sides, from the programmers and the users. Users could use it to analyse the content in their feed/history and see who is represented. Programmers could use it to see where the coding generates bias output.