On human messiness / when the algorithm doesn’t understand me
I’ve had one of those weeks where Spotify just can’t get it right. Everything is off. My AI DJ sucks. My song radios are weak. My finger is tired from hitting the skip button and I’ve been thinking that either I’m in my luteal phase or the algorithm is dumber than I’ve been led to believe. Or both. I thought our digital overlords were supposed to be observing our every move and adapting in order to keep us plugged in?
For example, I have a playlist called BTS no skips which I’ve been listening to almost exclusively for 2 months, and yet my Spotify DJ hardly ever serves me BTS. I rarely skip a Bangtan song when one does come up — I’m a very committed and loyal ARMY member by now. Does the AI DJ discriminate against Asian artists? Or does it keep reverting to serving me white people because I’m white?
I asked ChatGPT to give me a succinct list of Spotify’s algorithmic inputs. It generated this using research papers and Spotify’s own information, not internet opinions:
Behavioural Inputs (What I do)
Listening Behaviour
— Full plays
— Early skips
— Late skips
— Replays
— Pauses / scrubbing
— Volume adjustments (soft signal)
Engagement Signals
— Liking a song
— Saving to library
— Adding to playlists
— Removing from playlists
— Following an artist
— Searching for songs/artists
— Sharing tracks with others
Contextual Behaviour
— Time of day you listen
— Day of week
— Your device (phone vs laptop vs speaker)
— Activity context (gym, commute, study — inferred from patterns)
— Location patterns (not precise tracking, but general contexts like “at home”)
Session Patterns
— How long your listening sessions are
— What you typically play first
— How you move between genres within a session
— How often you skip in certain moods/contexts
Cultural Inputs (What the world is doing)
Community & Collective Listening Data
— What people with similar taste profiles listen to
— Trends among users who like the same artists you like
— Genre clusters and taste “neighbourhoods”
Social + Internet Signals
— Music popularity on TikTok, Instagram, YouTube, Reddit, etc.
— Blog reviews, music journalism, and online sentiment
— Memes and viral moments tied to specific tracks or artists
Audio + Metadata Analysis
— Tempo, mood, danceability
— Vocal/instrument ratios
— Genre tagging from global databases
— How critics and listeners describe the artist or sound
Release Trends
— New music that’s breaking out in your region
— Tracks rising quickly on playlists you follow
— Global vs local cultural shifts in genre popularity
It’s always a bit terrifying when you’re reminded how closely you’re being watched by the machine, but considering all the data it must be retrieving from my extensive listening habits; I’m fascinated that it’s still getting it so wrong.
I asked if Spotify’s algorithm has a cultural bias and Chat thought for about 8 seconds before it said yes:
Spotify’s algorithm shows cultural bias because it’s built around Western listening patterns, English-language metadata, and genre systems that don’t map well to non-Western music. As a result, artists like BTS often get misclassified, sit outside Spotify’s main “taste clusters,” and are recommended less often even when you listen to them heavily.
Spotify itself has pretty much admitted its product is biased toward western artists. That’s a massive shame because my BTS journey has really opened a window to global music for me, and I’m suddenly exploring artists who sing predominantly in other languages (and loving them). How many more am I missing because I can’t find them? I miss music stores.
Chat even suggested that the Spotify algorithm is probably viewing my BTS bias as a ‘phase’ that I’ll get over soon. It’s just biding it’s time, waiting for me to revert to Taylor Swift. More than anything else, that proves the algorithm might be ‘smart’ but it doesn’t understand anything. I’ve been aggressively skipping every Taylor Swift song since her album came out because I feel emotionally betrayed by her. Even when a fave TS song comes up I’m actively choosing not to listen because I’m performing my parasocial outrage. I guess there’s no digital input that could convey that information to Spotify (except maybe this blog 😅).
It's also outrageous and amusing to me that the algorithm is treating my fixations like many people in my life have treated them over the years – as an embarrassing side effect of my ‘obsessive personality’. Ugh.
So I’m taking the power back. I’ve made a new playlist called rnrn filled with songs I know I wouldn’t skip right now if they were served to me. I’ve made a rule that if I ever feel the urge to skip a song on the list I have to remove it.
My taste is changing all the time because of all the parts of my brain the algorithm can’t understand. Messy motivations like: if the book I’m reading is depressing, or if I’m having a joy-spike because I’m dancing in the kitchen with my kids, or if I’ve spent too long on LinkedIn and feel unsuccessful and inadequate.
Maybe the Singularity will be reached when there are no undecipherable parts of humanity left. It feels rebellious to take the power back while we still can.
That's why I’ve decided to make this blog WOM-only. I’ve been calling it a secret blog for a while but that’s not actually the case — I bring it up in conversation all the time because I’m insufferable.
My only hope for this place was that it would get me writing again. Finding any kind of audience is a bonus. A few of my friends are reading it and that’s led to some fun conversations so it would be disingenuous of me to say that I don’t want or need an audience.
Instead, I’ve settled on the idea that I’ll indulge my urge to self-promote through the mediums of conversation, email or text message only, and continue to resist posting on digital platforms. Since it’s too tiny for Google or AI search to care about, it’s very unlikely it’ll be found through digital means and I like that about it: my irrational, pointless and very human little rebellion against the machine.