The Beatles’ ‘Now and Then’ Was Made With AI (and That’s Okay)


I happened to catch some of the Grammy Awards last night, and while pop, rap, and country definitely took center stage, it was the Best Rock Performance category that most caught my attention. The award went to The Beatles for their song “Now and Then,” which is a confusing sentence in 2025. You might be aware The Beatles operated largely in the 1960s, not the 2020s, and seeing as only half the group is still alive, winning a Grammy for a new song might sound a bit strange.

Nevertheless, the song is new—at least, as new as a Beatles song can be. The track dropped towards the end of 2023, and is built from a demo John Lennon recorded shortly before his death. In the ’90s, the living three Beatles members (Paul McCartney, Ringo Starr, and George Harrison) attempted to finish the song, but never completed it, as the original audio quality was too poor. Modern technology, however, made it possible to create the product you hear today.

I love The Beatles, but the song itself isn’t really what I paid attention to. (I didn’t really give it much thought until Sunday.) Instead, an announcer made it clear the song was produced with AI, a statement that, in 2025, elicits anything from an eye roll to a heavy sigh from this tech editor. AI-generated music is very much a reality today. While there are some convincing results from these AI-generated tools, there are plenty of tells to look out for. The idea that The Beatles would put out a song with some level of AI generation didn’t sit well with me, and I bet many others out there.

However, it’s important to note the difference between AI-generated, and AI-produced. They sound the same, but they’re not. I want to be clear: “Now and Then” is not an AI-generated song: No one is tapping AI to recreate the voice of John Lennon to make another award-winning song.

Not that kind of AI

AI doesn’t just mean the artificially generated content we’re all accustomed to. While it feels like AI took over our lives with the introduction of ChatGPT in late 2022, companies have embedded the tech in our products and services for a lot longer than that.

AI is, perhaps, a bit of a misnomer. In this context, machine learning is a bit more accurate. Machine learning is, very simply, when a program is able to adapt and grow based on the data it experiences—similar to how our minds work. You feed the program training data, and it adjusts its assumptions and outputs accordingly. While the actual process is much more complicated than this, machine learning empowers programs to do some great things.

One of those things is audio track separation: Part of the reason the “Now and Then” project was shelved was because they couldn’t properly mix the song, since Lennon’s original recording was so rough. But using an audio editing tool powered with machine learning, producers were able to separate Lennon’s vocals from the piano. Neither the piano nor the vocals were generated with AI—rather, the tool was able to break these tracks apart, so producers, along with the two living Beatles, could build upon them to record, mix, and ship a completed song. McCartney posted as much on X the summer before releasing the song:

For many reasons, I’m happy this “AI Beatles song” didn’t artificially bring back John Lennon. It’s genuinely sweet to have a song Lennon started decades ago, properly finished by his former bandmates, with Lennon’s son accepting the award on the band’s behalf.

The music video, on the other hand, definitely pushes things a bit further, juxtaposing archival footage of deceased Beatles members John Lennon and George Harrison alongside current footage of McCartney and Starr. (It was produced by Peter Jackson, who both produced Now and Then, as well as the 2021 Beatles documentary Get Back.) Still, it’s more weird than anything else, and certainly isn’t a product purporting to represent reality—as opposed to much of the AI-generated content you encounter in the wild.




Source link

Exit mobile version