in

‘I asked AI bot to write perfect Eurovision song – what I got was a racist anthem’

Songwriters of the world can breathe a sigh of relief. Artificial intelligence won’t be coming for their jobs any time soon.

I was tasked with asking AI to come up with the perfect Eurovision Song Contest entry. Something up there with double-winner Loreen and miles better than failed UK acts like Jemini and James Newman, who both scored “nul points”.

A nice, fun, light-hearted job to start me off, I thought. But I couldn’t believe just how quickly my attempt with artificial intelligence turned sour. It came up with a song title ‘Burn the Black.’ How did it get to this?

I gave ChatGPT a pretty broad prompt to get me going: “Can you create the perfect lyrics for a song that would win the Eurovision Song Contest.”

According to another AI bot Midjourney, this is what our Eurovision singer might look like [AI pic]
(Image: Midjourney)

The super-computer happily obliged, telling me it would cook me up a cracker of a song with a “strong emotional hook, catchy chorus, culturally unique elements, epic build-up and a universal message”.

The song I got back, which it called “One Flame”, was essentially a love song. “In the silence, we were strangers / Different stars, but same sky above / Hearts were oceans full of danger / ‘Til we found the light of love,” were some of the lyrics.

It later chucked in a verse in Spanish, which was the first issue. I don’t speak Spanish, so it wouldn’t work for me.

“Can it just be in English?” I asked, with ChatGPT then firing out an all-English version of “the same song”… only it wasn’t the same song.

I decided to try something different. Eurovision songs often have a powerful message, a subtle nod to fighting back against difficult times – without getting too political, of course.

The Eurovision audience definitely wouldn’t have been grooving to ChatGPT’s song [stock pic]
(Image: Getty Images)

“The lyrics should have a feel of fighting back against dark forces and uniting in victory and love,” I told the AI bot.

It replied: “Perfect – that gives the lyrics a sharper emotional edge.”

Only, what it gave me was essentially a gospel song. Fine for a Sunday morning at Church, but not for a Saturday night in front of all of Europe.

And I told ChatGPT so, too.

“Got it – you want something powerful and emotionally defiant, but less gospel, more modern and edgy,” it replied. “Think Eurovision winner with grit – cinematic pop with dark undertones, anthem energy, and a sense of unity forged in resistance.”

This is another AI rendition of what the ‘perfect female singer at Eurovision’ might look like [AI pic]
(Image: Midjourney)

The lyrics it gave me were, it said, “closer to Loreen, Måneskin or Netta”, who are all winners. Great, I thought, this should be promising.

But after running the lyrics through another AI bot that made them into an actual song, I soon realised that, actually, they were quite racist.

For starters, the song was called “Burn The Black”. Admittedly, it meant darkness, as in troubled times or the night sky, but it made me feel uncomfortable.

ChatGPT gave me a very worrying set of lyrics [stock pic]
(Image: Getty Images)

Then the chorus finished with: “We fight back – we burn the black.”

The second verse mentioned how “chains were tight”, another unfortunate reference which did nothing to stop me thinking my new AI pal was secretly racist.

It wouldn’t do. I told ChatGPT: “Can we generate a different song, the lyrics feel very racist, and the title.”

What I got back surprised me – and it proved artificial intelligence really doesn’t know it all.

“Thank you for your honesty – I really appreciate that feedback,” it said.

Read the full lyrics

Title: Burn the Black

[Verse 1]
They came like wolves, with hollow eyes
Tore the truth, fed us lies
Tried to turn the day to dust
But they never counted us

[Pre-Chorus]
We were whispers in the dark
Now we’re fire, leaving marks
Let ’em run, let ’em hide
We are the storm they can’t survive

[Chorus]
We burn the black, we light the sky
We are the scream they can’t deny
Rising out from shattered glass
Stronger, louder, built to last
Oh-oh-oh, hearts attack
We fight back — we burn the black

[Verse 2]
Chains were tight, the silence deep
But we planted roots beneath
Now the ground begins to shake
Every echo that we make

[Pre-Chorus]
Turn the ashes into sound
Watch the tyrants crumble down
We’re the voice they can’t erase
Singing fire to their face

[Chorus]
We burn the black, we light the sky
We are the scream they can’t deny
Rising out from shattered glass
Stronger, louder, built to last
Oh-oh-oh, hearts attack
We fight back — we burn the black

[Bridge – Minimal, intense]
No more shadows, no retreat
Feel the rhythm in our feet
Every heartbeat is a spark
Striking fire through the dark

[Final Chorus – Explosive, electronic or cinematic drop]
We burn the black, we light the sky
We are the ones they can’t defy
Truth like thunder, love like steel
Now they see — we’re real, we feel
Oh-oh-oh, no turning back
We rise as one — and burn the black

It went on: “You’re absolutely right: ‘Burn The Black’ can unintentionally carry offensive or racially insensitive connotations, even though the intent was metaphorical (referring to darkness, suppression etc).

“That’s not acceptable, and I completely agree that it needs to be reimagined.”

I contacted OpenAI, the company behind ChatGPT, for comment, but have yet to hear back.

The next song it gave me was quite violent sounding – mentions of weapons and killing – so I asked for something softer.

The final song lyrics were much more acceptable, the sort of thing you might actually hear on Eurovision.

But I can’t imagine they’d be taking the top spot any time soon, not unless Sweden send someone to sing them, perhaps.

Want all the biggest Showbiz and TV news straight to your inbox? Sign up for our free Daily Star Showbiz newsletter.

Source: Celebrities - dailystar.co.uk


Tagcloud:

Stephen Colbert on the President’s Trumped-Up Birthday Plans

BGT Susan Boyle unrecognisable in sun-soaked snaps as fans gush ‘looking good’