If We'll Be Listening to AI Music Anyway, It Might As Well Be Good

If you’ve heard AI music you know the tells - meaningless lyrics, repeating keywords, unexpected instrument transitions. It’s immediately obvious to audiophiles, and twice as jarring to professional musicians. But a few years from now it may be impossible even for experts to tell.

🙋‍♂️ How is this technically possible?

The current technology is actually already good enough to fool normies - it’s just that many people using these tools don’t put enough effort into their creations, and out comes something easily identifiable as slop.

If we wanted to make AI music that is indistinguishable from human-made, instrumental music is the easiest target - electronic/MIDI music in particular. If models, instead of synthesizing raw audio, use the same virtual instruments, samples, and effect chains as human producers do, the sound will be exactly the same.

Realistic instruments, human-like composition, lyrics, and vocals are a more difficult technical problem, but also eventually solvable with better models. High-fidelity voice cloning, natural breathing, vibrato, and phrasing have been possible for years.

🤨 Wouldn’t artists refuse to make AI-generated music?

Artists have been singing and playing other people’s music for decades - buying it from ghostwriters, heavily borrowing from their idols, sometimes outright stealing. Surely today some human artists do not balk at the idea of using music they didn’t write, and are already using AI to generate lyrics or experiment with different composition ideas. If you really want to sell it as your work, once you’re happy with the product you can rewrite it a bit, sing the vocals yourself, or play on your own instruments.

But soon there will be end-to-end systems that will be able to take your voice, your instruments, and your existing discography, and generate more with very little guidance - indistinguishable from doing it yourself, even to experts. Years ago people were already making new 2Pac albums and Michael Jackson covers of The Weeknd.

In the end some big artist will license their style and start raking in millions from other people’s generations, and the floodgates will open.

🤔 Wouldn’t consumers refuse to listen to AI-generated music?

They might, as long as they can still identify it. But artists can hide behind pseudonyms, and conceal the fact they’re using AI.

Yes, our genuine connection with the artist can be an important part of the experience of music - going to a concert, buying merch, getting an autograph, building a parasocial relationship by following their every move on Instagram. Yet there are famous artists like Sia, Deadmau5, Buckethead, Ghost, even Banksy if you will - whose personalities we don’t know that much about, whom we mainly connect to through the art they make.

And there are a lot of cases where music is a utility - not the main activity but rather the background: games, films, ads, even physical businesses like malls, shops, or restaurants. Suno has 2 million paid subscribers, and some of them are already selling the slop. Here’s an anecdote: a few months ago we went to a tavern in Greece - the music there vaguely mentioned “beach” every other line, had no hits on Shazam, and was almost certainly AI-generated.

I was the only one in our group who noticed. Normies already don’t care.

🤷‍♂️ So what then?

I love real, human music, but many don’t pay much attention, especially if it’s elevator music running in the background of their busy lives. People can certainly vote with their wallets and refuse to buy AI music, or avoid businesses that play it. But if we are to judge by the explosion of mobile apps targeting normies - fashion apps, Ghiblify apps, what-will-our-baby-look-like apps - the professed AI hate is a luxury belief that more and more will abandon as the technology improves.

Even though it’s AI, it will eventually end up being better than 99% of other stuff - there’s just a lot of crappy music out there. I actually hope these tools improve sooner rather than later - slop music is bad for the soul.

So if we’ll be listening to AI music anyway, it might as well be good.