The prospect of artificial intelligence (AI) gets us excited and fearful in equal measures, as we salivate over how much easier machines may be able to make our lives in the years to come, but do so with a knot of fear in our stomachs that one day, our jobs might be taken by computers.
We only have to look in the likes of banks and supermarkets to see how machines are beginning to do jobs traditionally taken by people, but how far could this go?
Perhaps the hardest nut for AI to crack is art and any kind of creative output, such as writing a story, or composing a song. A team from the University of Toronto has had a stab at it though, with its ‘Neural Karaoke’ project. To showcase its abilities, the team has shared a seasonal ditty in which the music and lyrics are entirely composed and ‘performed’ by AI, simply from analysing a picture of a Christmas tree, and sifting through hours of music and video game footage to get an idea of what ‘human’ songs are meant to sound like.
The results, depending on your attitude and how much you value human songwriting, are either reassuring or disappointing. You can have a listen to it below. It will only take up a minute of your time, though it’s a minute you’ll never get back.
The Guardian wrote a lengthy and generally encouraging piece on the song, though it did describe it as ‘vaguely unsettling’ and rubbished its prospects of ever being a Christmas No. 1. Entertainment website The A.V. Club was less kind, quipping:
“These are not lyrics, they are the moans of the damned, trapped between this world and something beyond it, just conscious enough to know they are not at rest.”
I’m afraid I’m rather more in the creeped out and somewhat amused camp than the awestruck one too. I don’t want to be too harsh on the team behind this, as it’s clear that a lot of work has gone into it and they’re trying new and clever things, but dear me, this shows that AI still has a long way to go to emulate even the weakest of human musicians.
Call me demanding, but I think we’re in an age now where this just isn’t very impressive. Is it that much to expect lyrics that make grammatical sense rather than “a hundred and a half hour ago”? Or that don’t try to mess with the linear nature of time like “I’ve always been there for the rest of our lives” does?
Musically, do the vocals need to be that tuneless? Can we have some drums and not just occasional thuds that sound like scratches on a record? Maybe some kind of chorus would help?
My favourite bit is when we hear ‘A fairy tale. A Christmas tree’ and you think it just might be veering towards something approaching lyrical relevance and melody, perhaps even a key change, but have you ever heard a weaker closing line of a song than ‘there are lots and lots and lots of flowers’?
For a start, there aren’t lots of flowers. I can’t see any flowers in the picture. That’s the second time in the space of a minute that Neural Karaoke has mistakenly told us there are flowers. Let’s hope this pursuit never develops into Neural Floristry, or there may be many irate customers as the AI quite earnestly sells them bouquets of nothing believing them to be irises and chrysanthemums.
As well as this though, you’d think after analysing all this music, Neural Karaoke would have learned that songs tend to come to some kind of ending, be it a crescendo or simply fading out. This one jolts to a stop abruptly, ending on the jarring word ‘flowers’ in a manner about as delicate as slamming the lid down on a piano while someone’s playing it.
It is what it is, I suppose, a song written with no human input. It seems like evidence though that Slade, The Pogues and Cliff Richard needn’t worry about AI tools taking a share of their Christmas royalties for a while yet!