REMIXING THE BEATLES: “AI is not creating John’s voice. John’s voice existed on that cassette and we made the song around him”: Giles Martin explains why you’d be wrong to think ‘AI’ created Lennon’s parts for The Beatles’ ‘Now and Then.’

Billed as the final single from The Beatles, Now and Then was always going to be a big deal. The track, which is nominated for Song of the Year at this weekend’s Brit Awards, has its roots in a 1977 demo recorded by John Lennon.

Its mix features all four original members – Paul McCartney and Ringo Starr, along with the late Lennon and George Harrison – and comes with a long history that includes an aborted mid-’90s attempt at a finished release.

There’s another headline-grabbing element to the eventual release of Now and Then though. As has been widely, and on occasion a little misleadingly reported, the track’s creation has been made possible thanks to the use of neural network technology. Or, as how some have more flatly termed it – AI.

This latter fact seems to have led to misconceptions among some fans, as the track’s co-producer Giles Martin (son of late Beatles producer George Martin) demonstrates in an exclusive video for MusicRadar.

“I think there is this supposition that we used AI to recreate something, or to perhaps enhance John Lennon’s voice,” Martin tells us. “This simply wasn’t the case. All we did was clean a cassette recording he had made all those years ago.”

It’s true that some of the public reaction to Now and Then, particularly in the wake of its recent Grammy win for Best Rock Performance, has led some to raise a suspicious eyebrow. But this perception stems from the entirely false notion that the track has harnessed some form of generative AI. Which it very much hasn’t.

The demixing technology that Peter Jackson has developed seems like Clarke’s Third Law sort of stuff: “Any sufficiently advanced technology is indistinguishable from magic.” Every commercially-available demix software that I’ve either used or demoed, generates serious artifacts when separating parts, creating digital chirps and a sort of watery filtered sound that’s particularly noticeable when a track is soloed. But listening to Giles Martin playing the soloed tracks of Beatles songs in the video below, and little or no audible artifacts can be heard:

Remixing Beatles songs and digitally cleaning vocal tracks seems like pretty benign stuff. Still though, no one should forget that AI and machine learning technology can also be used to far more terrifying ends: