Artificial playlists, little intelligence

Can Spotify's AI playlists handle punk rock?

The UK recently became one of the first countries to get access to Spotify's new AI playlist function, which creates song lists for its premium subscribers in response to any mix of their questions, requests or non-sequiturs.

It feels like one of the bazillion generative AI, or Gen AI if you prefer, applications that have burst from the tech scene since ChatGPT went public in November 2022, upending creative expectations with breathless enthusiasm. 

There’s plenty to say about the ethics and implications for musicians about AI, and John Strohm articulated much of this on his Substack earlier this month, but applying it to playlist creation is at least somewhat interesting (and it can’t be worse than the experience of using Spotify’s clown-like AI DJ).

So, with Spotify already set to face competition from Amazon Music’s AI playlist generator - I thought I'd see if it might actually be useful for music fans.

And what could be more useful than punk rock?

The TL;DR, it's... OK. 

Spotify’s AI playlist generator is certainly better than the exceptionally average, sometimes misleading, copy ChatGPT spits out and it’s a reasonable way to pass the time.

Accessed via Android and iOS devices, for now only in the UK and Australia, the AI Playlist claims to “turn your ideas into playlists using AI” by creating a personalised playlist based on the tracks, artists, and genres the feature ‘thinks’ you’ll like.

It’s a process that starts off with a jaunty “What do you want to hear today?” (Though it seems immune to snarky prompts.)

Trying something more straight-forward (Prompt: 10 punk rock classics) yields songs by the Ramones, Sex Pistols, The Clash and more. 

So far so good. 

A song by X is on the list, so I ask Spotify to swap that for something by The Damned to bring it nearer to ‘punk rock classics’ and get ‘Love Song’.

But now music fans’ arguments with the tech will start. Obviously, music is very subjective, particularly when you talk about the best, greatest or most influential, but asking for classics feels like it should be fairly straightforward - particularly with all the data a company like Spotify has access to.

So, I’m a little confused to have the Ramones represented by ‘Bonzo Goes to Bitburg’ from their ninth album, 1986’s Animal Boy, the Black Dots version of Bad Brains’ ‘Pay to Cum’ and a 2003 live cut of Buzzcocks’ ‘Ever Fallen in Love…’.

It’s not as if there’s a bad time to hear those songs, but they’re not quite what I asked for.

I’ve yet to find a band that hits me as hard as a certain Twin Cities trio, not for want of trying, so I try to challenge it with a prompt of “Songs like Hüsker Dü”.

Again, the playlist generated is fine, even if it can require a bit of thought to connect some of the selections. These range from the thematically perfect (Bob Mould’s 90s band Sugar and Hüsker Dü contemporaries like The Replacements, Descendants and Bad Brains) to connections of various strengths (Mudhoney, Fugazi, Rocket From The Crypt and Archers Of Loaf).

The likely lesson here is that “Songs like [Band X]” is almost as subjective as any list of best, greatest or most influential songs. But at least the Spotify-generated playlist titles are improving, graduating from “Punk Rock Classics Plus The Damned” to “Echoes of Hüsker Dü”.

I love dipping into new genres and sub-genres of music and listen to quite a bit of Deafheaven and recently discovered Alcest, so would like to dig into similar ‘blackgaze’ bands, but I’m losing patience with literal commands so try something different.

Prompt: Exploring the connection between space and time with blackgaze

I don’t know what that means or even exactly what I want, but in Chat GPT style it feels all very believable in its list of atmospheric tracks, albeit a bit heavy on lesser known side projects. At this point a conspiracy theorist might speculate that Spotify favours songs with lower numbers as part of its determination not to pay all the songs that are streamed on its platform.

Beyond that, the playlists are somewhat hit and miss.

Prompt: “10 song playlist of US hardcore punk from the 1980s with no band appearing more than once.”

Answer: “I’ve tried a bunch of different ways to make that playlist, but it’s not coming together. Perhaps you could simplify your request.”

It also can’t help with something useful like: “punk rock songs released in the last week”. 

Giving the application a wider time zone isn’t always much help either.

“I’ve successfully found 10 punk rock songs from 2009 for you. Enjoy the throwback!” it celebrates one playlist creation.

Sadly, not one of the songs on the list was released after 2006, and it included Green Day’s ‘Basket Case’, which has just celebrated its 30th anniversary.

It’s almost as if Spotify has a data problem. As if a company that routinely mis-catalogues music… might not really care about, or truly understand, it.

There’s also a nagging suspicion that AI-playlists remove a little of the meaning, the feeling inherent in the music, replacing it with a utilitarian view of functionality.

It’s a suspicion reinforced by Spotify’s suggestions of prompts for aural wallpaper - “Give me some classical music for studying” or “Help me get pumped with a 1 hour workout playlist”.

The thing is, Spotify already has plenty of its own “Soundtracks for Studying” and “Power Workout” type playlists.

AI’s going to overwhelm them, which is fine. But the coming waves of playlists are only likely to be ok, which is not fine.

At this point I should probably swiftly insert a plug for my definitely not AI generated or otherwise assisted Fighting with the Undertow monthly playlists of new punk rock, alt rock and indie tracks, but all this messing around with tech of marginal value has drained me of enthusiasm for that, even as I can’t help but try one more combination.

Prompt: “The 5 most popular punk songs.”

Answer: “Here’s a playlist with 5 of the most popular punk songs, focusing on the genre’s top hits.”

And to its credit, ‘American Idiot’ probably is at least one of the more popular punk rock songs, but it’s not the place you’d want to find three Daft Punk tunes.

Now, one more time Spotify and this time… get lucky.

Main image by Rock'n Roll Monkey on Unsplash