With the slow creep of artificial intelligence into every aspect of life, there is growing worry that musicians will fall behind robotic contemporaries.
We are a long way from that shitstorm of a situation, but Google’s MusicLM tool is the butterfly flapping its wings.
The free tool turns text prompts into two separate twenty-second musical clips. Prompts can include specific genres, instruments, and even the #vibe you want the clip to be.
If you can't think of something to type, the tool has a few ideas of its own for you to try - ideas that are surprisingly abstract.
Google's AI Music Generator, MusicLM
Understanding subjective information - one suggested prompt is "sounds for a long walk on the beach" - is an interesting development for AI. The tool has never been to the beach so how does it even know where to start on 'beach music'?
Before it was released, humans fed over five million audio clips into MusicLR. Musicians described the “genre, mood, and instrument” present in some of the clips. Supposedly, this helped the tool understand what kind of musical sounds are suitable for different situations, such as walking on the beach.
That's a lot of musical knowledge inside this audio weapon, and they do say knowledge is power.
If it is going to take over the world, and by extension our airwaves, it needs to show the potential to entertain us at least as well as the crop of human musicians we're currently dealing with. Therefore, it’s only right that we find this AI's creative limits.
I typed: “Drum and bass music playing at a 4 AM festival slot when I'm completely lost in the campgrounds and want to go to sleep but I'm absolutely wired for some reason and I'm pretty sure I don't have my own shoes on”.
And it did not produce any kind of music you would hear in such a dilemma, but it did produce digital lo-fi jazz kind of songs.
I assumed the AI picked up on the actual musical terms (“drum and bass music”) and other descriptive keywords, such as “sleep”, “fun”, and “weird”.
That assumption came shattering down when I prompted: “Fun and weird drum and bass music that makes me sleep” and it spat out clips that could only be lullabies for demons. Weirdly, the clips are far more suitable to the previous prompt.
To be fair to the robot, I don’t even know what the first prompt should sound like - I’d just know it when I hear it. What I do know though, is what music gets the fellas and fellettes grooving.
And so does MusicLM, nailing the prompt: “Electronic music a bunch of hooligans play before partying that gets them in the mood to have a good time”.
It’s a long way from becoming a curator of the perfect tune for the moment but it knows how to dissect well-known vibes/moods. Where it is truly impressive - to a scary level - is its functionality as a genuine musical tool.
We found that using technical terminology, while not perfect, produced the most precise results. The prompt: “Dark sounding half-time drum and bass with robotic synth sounds and horns and a dirty bassline”, did produce a synthy, darker clip that is most definitely drum and bass. It definitely was not half time though.
This instrumental accuracy is exactly what it was made for.
Neil Zighodour, a research scientist on the tool, said: “The most important thing for us is that it’s used by artists in a way that is meaningful to them and can allow them to explore new ways of creating sound."
Neil's not lying. If an aspiring guitarist wants a backing drumming pattern during practice, this tool provides just that. They can even customise the tempo, the genre of music they want the drum to replicate, and the addition of other aspects of the drum kit (hi-hats, for example).
The 'music' the AI produced won't be making any artists with functioning ears jealous, but it's not even a year old yet. Its ability to provide a basic instrumental loop, as well as catch a vibe someone lays down, is genuinely useful.
'Electronic Music A Bunch Of Hooligans Play Before Going Out' has already received one like and one share on Soundcloud, barely an hour after it went public.
Give it a year or two and a completely AI-made song will be in at least one of your playlists. Give it five and the #1 song in the world will be AI-made. Within ten years I don't think it's crazy to say the headline act at Rhythm and Vines will be a robot.
That is not exactly a desirable future for DJs and fans alike. Nothing will ever beat a DJ dropping a never-before-heard BANGER as they lift and drop the arm. So enjoy those moments while they're still here.
