Many of you may have heard that The Beatles released their final song this week entitled “Now and Then”. It is based on a John Lennon demo which George Harrison, Paul McCartney, and Ringo Starr added new parts to it similar to the two songs (“Free as a Bird” and “Real Love”) that were part of the Anthology project in 1995.
An aside – who thought we’d be talking about a new Beatles single and a new Rolling Stones album in 2023? Anyway …
“Now and Then” was a tune they worked on for Anthology, but ultimately shelved. The reason why may be slightly different depending on who you ask, but the story now is that they could not really extract Lennon’s vocal from the piano enough.
When the Get Back documentary was released a few years ago, director Peter Jackson’s WingNut films came up with technology to de-mix audio to get better separation. What does this mean? If you have a mono or stereo source anything would be together. They are not multi-tracks where everything is isolated. To be able to better mix audio for that project, the technology known as MAL (machine aided learning, but probably also a nice tribute to longtime Beatles roadie Mal Evans) was used. The results were phenomenal.
It was after that Macca got the idea to revisit “Now and Then”. This 12 minute documentary released ahead of the song talks about that and more. The MAL part kicks in around 5:53. It’s worth your time to watch and listen to Sean Ono Lennon, Macca, Ringo, and Jackson talk about it. As someone who has spent a decent amount of time in the studio, hearing John Lennon’s isolated voice around 7:02 is amazing. You would not hear it was from a mono demo cassette. I hear no artifacts as a lot of the de-mixing technology has suffered from over the years.
The Beatles and Technology
The Beatles embraced cutting edge techniques back then and in using MAL, they are carrying that tradition forward. It is also interesting to see that credited directly on the song is Emile de la Rey (Head of Machine Learning).
Second, there seems to be confusion around artificial intelligence (AI) and machine learning (ML). A lot of the reviews and chatter around “Now and Then” use AI. Many of the criticisms prior to its release thought that AI was going to do weird things like replace Lennon’s vocals or something along those lines.
What was employed – ML – is what enabled all of this to happen. AI is an overarching concept that implies computers/machines/software have human brain-like functionality to perform complex tasks and learn from them to get better/smarter to be more autonomous (think asking ChatGPT to write something you ask it to). ML, while a subset of AI, is more algorithm focused that use data to allos us to perform complex tasks and make better decisions. Those are bit simplified, but you get the gist. ML clearly is part of AI since it’s learning, but ML on its own is its own use case. For the “Now and Then” demo, MAL was smart enough to know a piano from John Lennon’s voice and could extract the voice.
We’ve seen some interesting AI uses in the art and music world including fakes of “new” Beatles style tunes. One example is “Grow Old With Me” which can be found on YouTube. “Now and Then” is not that. The song was all done by humans with assistance from ML to isolate Lennon’s vocals.
There seems to be a lot of FUD and confusion around AI and ML. I hope this blog post helps you understand the difference.
Oh, and in 2023, The Beatles are as relevant today as they were back then.