How AI brought John Lennon back to life for the last Beatles song

The Beatles will release what is said to be their last ever song this week, pieced together from recordings spanning more than four decades – and it would not have been possible without AI.

Now and Then has been edited together from a recording of the late John Lennon playing piano and singing at his home in New York in 1979. Now, artificial intelligence has been used to extract usable sections from that noisy tape.

These have been combined with guitar tracks from the late George Harrison, recorded in 1995 when efforts were made to finish the song. The endeavour was reportedly called off due to poor sound quality, which AI has now been able to solve.

Finally, new recordings made earlier this year from Paul McCartney and Ringo Starr were added. The single will be released at 2pm GMT on 2 November and will bookend the band’s career by also including Love Me Do, the first ever single by The Beatles, as a B-side.

McCartney has teased the existence of the song in interviews for months and said in a statement: “There it was, John’s voice, crystal clear. It’s quite emotional. And we all play on it, it’s a genuine Beatles recording. In 2023 to still be working on Beatles music, and about to release a new song the public haven’t heard, I think it’s an exciting thing.”

Film director Peter Jackson’s production company WingNut Films is credited on the single for “source separation”, but the company was not available for interview.

The company, which worked on the Lord of the Rings franchise, also produced the Get Back documentary, featuring footage of The Beatles recording Let it Be.

When creating Now and Then, engineers at WingNut worked on 60 hours of recording captured by a single microphone that picked up the musicians’ instruments in a noisy jumble rather than a carefully crafted mix. The microphone also captured background noise and chatter, which made much of the recording unusable.

The team decided to use AI to separate the dialogue from other noises to help editors create a workable documentary. Ultimately, the team was able to develop bespoke AI powerful enough to remove all background noise and isolate not only speech but even the sound of each instrument played in a band.

Jess Aslan at Goldsmiths, University of London, says The Beatles’ track is an interesting experiment because it was done transparently with the blessing of the living members of the band, but adds that AI is a double-edged sword that also presents risks to artists.

“One significant issue is that generative AI is squeezing the already extremely narrow creative job market,” she says. “Another is that of ownership, in that these large-scale models are in effect bypassing copyright laws and reconfiguring artists’ data without consent.