The Spotify algorithm has always served me well, recommending music to me that makes sense to recommend. I’ve discovered artists and music I don’t see how I would’ve ever found without the help of Spotify.

However, mycket vill ha mer as the the Swedish idiom goes, which lazily translates to “much wants more”.

At times, Spotify recommendations feel artificial and clinical. This point bothered my previous podcast co-host Johan so much that he left Spotify for Apple Music, with the argument that he’d take recommendations primarily based on human curation thank algorithms any day of the week even if the hit rate was lower.

The upside is, of course, the element of surprise.

You can categorize music with endless attributes to make the algorithm calculate similar music to recommend to it’s user. That covers the broadcasting end of the music listening experience pretty well. The problem lies in the other end, the listener. You can inform the algorithm of what you think of recommended music based on behaviour such as repeat listenings, saves to playlists, songs skipped. That covers the receiving end, but not very well. Because it’s harder to quantify taste. It’s more likely than not that you’ll like a song that’s similar to the attributes of music you’ve saved than a song that’s not similar. But the mind doesn’t always respect the boundaries of it’s own taste, and it’s hard to artificially extrapolate into the unknown when taste is so fickle. 

I digress, what I’m trying to say is that it’s hard to guess what else you like when you don’t even know yourself what else you like… until you hear it.

And that’s where humans come in handy. They can throw curve ball recommendations at you without caring for any of the actual metrics that matter, only based on their own hunches, vibes, mood, feelings or any other hard-to-quantify quality. There’s a high risk a lot of music recommended to you that doesn’t match your preferences will not be of interest to you. But it might! High risk, high reward. 

So I have a feature request for Spotify. It’s not likely that they will leave the artificial space of music recommendation. It’s truly a fantastic tool (side note, I’m so looking forward to the AI prompt playlist feature), but like with all products, there’s room for improvement, so here’s my proposal:

Looking at my saved music, my playlists, and my stats. My music taste likely doesn’t make sense. It’s all over the place, which is a good definition of how I’d define it myself. And I think there are many others like me, and many of those might have overlapping data with me. If there was another user our there that shared similar traits in their saved music, playlists and stats, let’s say we have a 50 % overlap in some arbitrary metric, I would be HUGELY interested in listening to what that other 50 % was.

So I would love access to digital twins in Spotify. A way for me to browse other Spotify users, anonymised of course, to learn more about what they like. Perhaps even ways to sort and filter based on metrics, like show me best matches in a specific genre. Or find my a digital twin for a specific playlist. Or time of day. 

This would allow Spotify to offer a human element. Crowdsourced automatically by it’s users, to recommend music, to extend upon or beyond it’s own algorithmic recommendations, to provide that much needed element of surprise.