AI Playlists

Spotify is evolving all the time and has now jumped on the AI bandwagon. Actually, they’ve been using it for a while and already have a “DJ” function which tries to give you music it thinks you might like accompanied by interludes of cheesy banter in an American accent. If you’re not too discerning it can work quite well and even help you discover new artists you like. But every now and then it’ll throw something like Rick Astley at you and you’ll be like “seriously?”. But now there is a more specific use of AI for you to experiment with in the form of AI Playlists. At the moment it is still a beta function and available on phone apps only, to Premium subscribers.

To generate an AI playlist you touch the ‘Your Library’ icon at bottom right of your phone screen and then the ‘+’ button top right. A pop up menu will then give you the options to create a new playlist, one of which is ‘AI Playlist’. When you select this you will be asked for a prompt. This is where you can get as creative and specific as you want. The more detail you provide, the more the AI has to work with and the better your results should be. You could just type ‘give me Katie Melua songs’ but that would be a bit too general since you might as well just search Katie Melua and play her artist page shuffled. I typed ‘give me Katie Melua songs to make me feel happy when I’m feeling a bit sad’. AI generated a playlist called “Katie Melua: Feel Good”, which you can check out below.


The AI always seems to generate playlists of 30 songs, with about two hours of playing time. That’s fair enough. As for its choices, well, it’s complicated. It isn’t merely picking a bunch of Katie’s songs at random but in theory is basing its selection on my particular listening history. That means that if you ask it to generate a playlist with the exact same prompt I used you would almost certainly get a completely different playlist. As to how well it matches my prompt, you have to accept that it was a rather vague instruction for it to work with. I don’t really see how it can be expected to know which Katie songs will make me feel good when I’m down (for the record, pretty much all of them!) so it is hard to judge the results. Certainly, a lot of my favourites are in there but so too are a few I probably wouldn’t have picked if I’d been creating this playlist manually. If I ask myself whether this playlist would fulfil the remit of making me feel good if I was sad then I would conclude absolutely it would. The caveat is I didn’t give it a difficult goal to achieve and I can’t really see any way in which it might have failed dismally.

In the interests of giving the AI a more challenging task, I asked it to create a playlist of quiet and relaxing songs for a calm Sunday morning. Now this is where it gets more interesting. Not specifying a particular artist gives the AI infinitely more options. Again, it succeeded in giving me exactly what I wanted. There were no tracks in the playlist that jarred with the remit and it is definitely a playlist I would use. There are songs and artists I’d never heard of in there but most interesting was the larger than expected amount of artists I had heard of. Some of these are obscure enough for me to realise that this list was genuinely constructed just for me based on artists I have listened to in the past as well others that it correctly anticipated I would probably like despite them being new to me. This is the real genius of AI. It can get to know you better than you know yourself. If a friend recommends music to you it is usually biased towards what they like rather than what they know you to like. AI knows every song you’ve played on Spotify and how often, and it can compare your choices with those of every other Spotify user to analyse what people who mostly listen to similar music to you listen to that you don’t and hence form recommendations that stand a good chance of hitting the mark. It won’t always get it right but the more music you listen to the more complete a picture the AI can get of your tastes and therefore the more likely it will be to succeed. Here is my “Sunday Morning Serenity” playlist, generated by AI. Note track 2, which beautifully fits my request but, as a piece by a Georgian composer performed by a Georgian concert pianist, is highly specific to my own listening history and I doubt would make it on to any generic “Sunday Morning Serenity” playlist. For me, it was proof positive that the AI was tailoring this list specifically to me.


I decided to give the AI one more challenge. I asked it to create a playlist of quiet instrumental music I could use as inspiring background music for writing. Once again, I was perfectly happy and surprised with the “Writing Atmosphere” playlist it gave me. This one is definitely based on my personal choices. I have an existing playlist of Writing Music with over 1300 tracks on it and I have no doubt the AI has noticed that, along with my general predilection for ambient music. This playlist has a much higher proportion of artists already known to me but the most interesting thing here is that AI could easily have fished all 30 tracks from my Writing Music playlist. That would certainly have been an easy choice and a safe bet for success. It didn’t do that. It has again successfully integrated a few new artists into the list and not one of its choices has caused me to hit the three dots and delete the track from the playlist. So, that’s three attempts at AI playlists and all three are keepers. We are still barely scratching the surface of what AI can do for us but one thing is for certain: it is here to stay and it is going to become ever more ubiquitous and powerful. Whether you find that scary or exhilarating you’d best buckle up for the ride. The genie is out of the bottle…