Music’s Golden Age

Any song, any time, anywhere, right now. But for how long?

In the nineties, I discovered EDM. I stumbled across Underworld, Orbital, Chemical Brothers, The Future Sound of London, and others. They weren’t on the radio; instead, they were part of soundtracks to movies like Hackers and Songs From The Cool World, and games like Wipeout 2097.

I was hooked. This kind of music has become the soundtrack to my life. I couldn’t find music like this on mainstream radio—the alt/nineties/EDM/nerd demographic doesn’t really justify its own radio station. I hate broadcast radio anyway: The ads kill the vibe, the DJ is way too cheerful, and the song quality is, well, meh.

In the early days, I had mixtapes. I’d hover over play and record until a song came on. Then I’d burn CDs (remember Apple encouraging everyone to Rip, Mix, and Burn, before they got DRM religion?) I started mixing MP3s, recording live sets. I dragged those nineties bands into the oughties, and now I have hundreds of gigabytes of music copied (5,097 folders last time I checked) backed up on several hard drives.

None of this is true for my daughter, who is entering a completely different music ecosystem:

• Playlists are her radio.

• Music is a constant companion, often listened to alone.

• Searching for a song is trivial and instantaneous.

• Sharing a track with anyone in the world takes three taps.

• Every kind of music is equally easy to find.

She also has better quality music than I had. Headphones, AirPods, and portable speakers are the norm, and the digital recordings are usually well restored or digitally mastered. Thanks to her devices, her personal soundtrack moves from car to bus to work to bed as she goes through her day.

A young woman juggling music styles, rendered with DiffusionBee

Liberated from the songs a radio station chooses to play, she jumps across genres constantly: Musicals. Satire. Dance. Cartoon title songs. Video game themes. Some kid with a ukulele and a loop pedal. Their Discord friend’s weird GarageBand track. A new genre is just a hashtag away.

She also time-travels. She’s as likely to listen to Queen or The Cure as she is to put on Cavetown or Billie Eilish. It’s been a gradual process, from torrents to digital downloads to streaming audio, but the way we consume music is radically different from what it once was. The entire span of music is laid before her (and all of us), across all of history.

She doesn’t have to choose between “popular” and “oldie” stations. She can choose the best songs from all time. Consider, for example, that Queen released Bohemian Rhapsody 35 years before my daughter was born. She loves Queen. By contrast, 35 years before I was born, Fred Astaire’s Cheek to Cheek was the top track.

I do not have Fred Astaire on a playlist.

(Sidenote: All three 1934 runner-up bands had “and his Orchestra” in their name.)

Her musical exploration is helped by algorithms and social networks. On streaming services like Spotify or distribution sites like Soundcloud, once you find your niche, a combination of crowdsourcing and recommendation models points you at new things.

No listener ever had it so good. It truly is the Golden Age of Music.

What will the next age be like?

Music is about to change again, dramatically. ChatGPT is upending how we brainstorm, learn, and produce written content. Stable diffusion is transforming art and design. But we don’t spend as much time talking about what it’s doing to how we produce and consume music.

Here are a few speculations: Personal soundtracks, democratization, generation, and flooding.

Music as vitamins

Post-COVID, we’ve all insulated ourselves from the world. Our headphones are armour, giving us a modicum of privacy. We have playlists to get us pumped while exercising, to focus us during work, or to relax us before bed. If we share biofeedback, or keystroke rate, or other context with an algorithm, maybe it will generate a personal soundtrack that helps our brains change state, or perform better.

Is a personalized, generative soundtrack a health supplement? If you think I’m being too futuristic, Biobeats—an app that generates custom music based on your heartbeat—launched in 2013.

Making everyone a creator

Music is math, and there are plenty of algorithms already integral to how modern music is produced. There are plugins for Ableton, a leading digital production tool (that I love!) which will generate a pattern of chords from a note. You simply tweak parameters like “happiness” or “funk.” There’s another where you sing and the plug-in turns that into notes. You can beatbox your beats into a drum synth.

How quickly will AI help pretty much anyone make a song? Software can generate lyrics, suggest notes and phrases, simulate instruments, automatically master a recording, correct voices, and more. Want Snoop to sing your track? There’s an app for that.

There are also many efforts to generate music algorithmically. Google has Project Magenta, which actually turned into several Ableton plugins including a synthesizer that moves “between” different instrument types.

From streaming to generation

But most of these just help humans who aren’t musical prodigies make songs more easily. What about AI that makes its own music?

Back in 2017, Spotify hired François Pachet, a researcher at Sony and one of the leading experts on music and AI. Which begs the question: Will Spotify shift from music streaming to music generation?

Spotify isn’t just a streaming service—it’s a petri dish. It is, quite simply, the best platform for split-testing AI tracks and improving algorithms based on what gets added or skipped. When you like a song, you add it to a playlist, telling the system the song is good, and what else it sounds like. When you skip a song, you tell the system you don’t like it.

Pretty sure that for some of what I listen to (techno, ambient, drone) I wouldn’t know an AI-generated track from a human-made one. Spotify could just inject songs into my playlists. In fact, many of its recent changes the company has made suggest it’s preparing for just this:

• You now add a track to a “liked songs” playlist instead of just clicking a heart icon to show you like it.

• The “Enhance” feature adds tracks Spotify thinks you’ll like to existing playlists.

• A new “AI DJ” creates custom playlists based on your tastes.

When Spotify plays a song it generated, it doesn’t have to pay royalties to artists, either. That’s good margins.

You May Like

If generative music gets really good—not just the bleeps and clicks of electronica, but actual music that sounds like a human band—it’ll be a floodgate. Will we be inundated with recommendations to things that sound just like Pink Floyd, or Rhianna, or Adventuretime? Every time we listen to generated, rather than human-created, music, the platforms make more profit, so they’re incentivized to do so.

The royalty disputes alone will be bonkers: We’re about to have a huge debate about the line between inspiration and exploitation (I wrote a bit about this earlier in Play Your Part, which you might want to read.) On the one hand, Picasso inspired most artists, but Picasso’s estate doesn’t get royalties on every painting. On the other, AI is trained on the work of others, and monetizing that work may be theft.

Getty Images is suing the creators of Stable Diffusion over unauthorized use of its images. They have a pretty convincing case. But the terms of service on many music platforms is less clear: What rights do you retain when you upload a photo to Facebook, or a track to Mixcloud? Let’s just say that if I were investing money I’d be looking at law firms.

The end game of this is that each of us has a set of songs custom-generated for us. What will it mean for the social part of music when the music you like best is so individualized that others don’t like it as much? What are the consequences of ultimate personalization?

The Golden Age

Anyway, here’s my spiky point of view: We’re currently living in the Golden Age of music: Any song, any time, anywhere, right now. Soon, that will be: Every possible song, all the time, everywhere, always. I’m not sure if that’s better or worse, but it will definitely be different.

I’m just starting to think about this phase of AI, and I’d love to hear some of your more outlandish predictions too.

Postscript

• What’s that, Internet Stranger? Not all music is streamable? Lots of artists aren’t online? You’re right! My point is that there is a vast amount, not a complete amount. I’m sorry it isn’t perfect. But we also live in a world where anyone can compose, record, and publish music from their phone, which is a far cry from the recording studios of old.

• I wrote a short book for O’Reilly on Music Science, as some of the early AI work was emerging and streaming services emerged. It’s a fascinating industry, and I learned a ton.