About a year ago, Spotify introduced AI Playlist curation to its Premium subscribers, harnessing the tech to create any kind of playlist to suit your mood, the weather, any activity, your existential crisis... you name it.
Enter your prompts, ChatGPT-style, and voila – a playlist of songs adhering to your parameters will be created for you. My colleague Harry McKerrell had a grand old time testing it out, with his prompts getting more and more creative and outlandish to push the boundaries of what this tech could do, ending up with playlists titled "Startled Chick's Dubstep Journey”, the “Abyss Stares Back” (there’s that existential crisis) and one that’s just songs about cheese.
As the sun finally broke through the dreary, rainy weather we've had for months in the UK, I started off with the prompt that Spotify showed in the demo: “Uplifting tracks for the moment it stops raining in London.”
It takes a few seconds (sometimes a minute or so for more complex prompts) to start generating the playlist, and I get a 12-strong list of songs such as Here Comes The Sun, Walking On Sunshine, Levitating, Uptown Funk, and so on. It’s a good playlist, all on theme. As I basked in the early March sunshine with the playlist running, it certainly gave me a spring in my step.
If you like your resulting playlist, you can even set it to update daily or weekly on certain days, like your own curated Discover Weekly or Daily Mix. You can also add songs, edit the list, change the name of the playlist and edit the prompt itself as much as you wish.
I pivot to asking it for a moody, dark academia-inspired playlist to help me write (something I often search for in existing playlists), and it comes up with tunes from Lana Del Rey, Noah Kahan, Of Mice And Men – they're a little too wispy, gentle and indie-folk compared with the deeper tones that I was after, so I edit the prompt to include strings, violins, piano and "with few lyrics".
This changes the whole playlist to include Arvo Part, Max Richter, Ludovico Einaudi, Nils Frahm, Clint Mansell – which was much closer to what I had in mind. However, the tone of the tunes were still perhaps on the light side, and I realised what I really wanted was the Ramid Djawadi-composed Game Of Thrones soundtrack, full of more dramatic cellos and more dynamic upheavals, and found myself tweaking the playlist to tunes I knew would hit the spot better.
I set it a harder task next: “10 songs that sound exactly like an Elliott Smith song. But don’t add any Elliott Smith songs.” This proved far trickier and showed the flaws of how an Al algorithm "thinks" compared with how a human links connections together. The Prompted Playlist of "Elliott Smith Soundalikes" included Leonard Cohen, Yo La Tengo, Jeff Mangum, and Spotify even tried to be cheeky by including tracks that had Smith on guitars, but not singing. I get it; it's hard to emulate the genius of his technical guitar playing and softly-sung confessional lyrics.
But the choices didn't seem close to what I consider to be Smith's sound. Personally, I'd have expected to see a few tracks from, say, Death Cab For Cutie or some early Jeffrey Lewis, while my husband (a big Elliott Smith fan) pointed out that tracks from Fionn Regan’s The End Of History would have been a better shout than most of what Spotify’s AI turned up here.
Where the Prompted Playlist comes to life, however, is when dealing with more data-driven prompts. I ask for the first 10 songs I ever played on Spotify, and then the top tracks from every year I first started using Spotify (since 2011), in ascending track length order. These parameters proved far more successful, bringing back musical memories in the same way as when you get picture memory snapshots from 10 years ago from your social media channels.
There are so many songs I’d completely forgotten I had played so much – especially songs that I had used repeatedly as test tracks when I joined What Hi-Fi? in 2012. I have absolutely no idea why I streamed Van Morrion’s Sweet Thing 36 times that year, while I had forgotten that I was obsessed with pianist James Rhodes’ albums – they were great for background music when I was writing my early reviews.
Every entry also comes with a little snippet, which I like. From stats on how many plays, track length and year, to even little blurbs on what the song contains and how it relates to your prompt.
Getting more specific on factual parameters rather than try to emulate my mood continued to prove more successful: I asked it to compile a playlist of songs used in the TV shows The Detectorists and Small Prophets – and this is the kind of data-driven resourceful activity that AI is best used for, saving me time to painstakingly look up each song and add them manually to a playlist (which I have done before with other TV shows).
While that time-saving nature is definitely appealing, the process of building a playlist where it needs to suit a mood or specific purpose is such a human, interactive activity. Every song you choose, the connection between them, and the order in which you place them, tells a story.
Sometimes the choices and connections you make are quirky, chaotic and make no sense to anyone but you – and it's these connections that not just make playlist curation so rewarding, but are also completely alien to an AI algorithm.
I spent a week creating Prompted Playlist as and when the mood struck, and found the results quite fleeting. You can be as precise as you want with your prompts, but it's not quite as tangible or as lasting as picking out the music yourself with intention and feeling behind it. After a week, it was really only The Detectorists/Small Prophets playlist that I kept coming back to.
Another of my colleagues, Robyn Quick, made the point that Spotify's existing recommended playlists already do a great job in offering playlists you'll like, but notes that "I trust my own tastes more than AI", and I can't help but agree.
Spotify's Prompted Playlist feature is undoubtedly clever and the scope of possibilities is vast. While it can't quite replace emotional connectivity that a human-curated playlist can provide, if you're a keen playlist builder, especially for a specific activity or mood on the fly, it's worth a whirl.
MORE:
The ultimate music tracks to test your hi-fi system
Qobuz rolls out a tool to protect us all from AI-generated music
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
