All AIs on Eurovision 2023: Don’t forget the lyrics
What is Generative AI capable of? In this series, We’re finding out by seeing if it can create a song to match the creative genius of Eurovision. First up, the lyrics…
Is generative AI about to revolutionise the inner workings of PR and comms? Will every campaign be AI-led in years to come? To separate the hype from reality, AxiCom set itself the challenge of using the current arsenal of free AI tools to tap into one of Europe’s annual cultural moments – Eurovision. The mission? Create a song worthy of the iconic competition (and have some fun along the way).
So, where do you start when you’re trying to compete with the likes of ABBA, Bucks Fizz, and Conchita Wurst? Like Bjorn and Benny, it begins with the lyrics.
Getting creative with ChatGPT
As Creative Content Manager at AxiCom UK, I jumped at the chance to take charge of the lyrics for our ’entry’. I wanted to experiment with ChatGPT in more detail. Maybe we could be the next Elton John and Bernie Taupin partnership.
The AI chatbot is famed for its ability to produce detailed and articulate answers to any prompt you can think of – from a new recipe to the intricacies of nuclear fusion. So, how hard could it be for it to come up with a song that would fit alongside the official entries at the Liverpool Arena on May 13?
(Narrator’s voice…it was harder than you think)
ChatGPT works by responding to specific, tailored prompts given to it by its human overlord/ a person inputting the information. In this case, that meant giving very specific instructions so the tool could get to the core of what makes the toe-tapping hits of Eurovision’s past so successful.
We were looking for lyrics that would form “the perfect Eurovision song” and embrace what the 2023 iteration of the competition is all about. It might produce a chart-topping song with lyrics to rival Bob Dylan or Taylor Swift. Or, if it didn’t hit the right Eurovision notes, it would be a swing and a miss.
I started by making sure ChatGPT was working from the correct reference point; asking it to name the 10 most well-known Eurovision songs of all-time and analyze them to find five characteristics they share. What it came up with was a strong foundation to build on.
But Eurovision is associated with much more than the song itself, and each year has its own specific theme. Taking these into account would be important for an authentic Eurovision song. So, I had ChatGPT summarize the consistent themes of the continent-wide competition (it identified things like the celebration of diversity, pushing boundaries, and creating memorable moments). I also provided it with the definition for 2023’s theme, ‘United by Music’.
Only then did I assign ChatGPT the role of a songwriter with the specific task to write the lyrics. This instruction also included information taken from a BBC article that identified the genre, structure, and tempo of winning Eurovision entries. The more specific you are with prompts, the better the results.
Within seconds, ChatGPT spat out a full set of lyrics, complete with verses, pre-chorus, chorus, bridge, and outro. Job done, right? It could have been, but I couldn’t resist tinkering after spotting a couple of obvious issues.
First, the title of the song – and the first line of the chorus – was “United by music”. ChatGPT was a little too on the nose with the 2023 theme.
The second issue was that ChatGPT seems to be obsessed with rhyming. Now, I know some of the greatest lyrics of all-time rhyme, but when every single pair of lyrics does so it can be a little much. But the real issue was that ChatGPT didn’t seem to understand what a rhyme was. Or, more accurately, it did. But it refused to factor that into adapted lyrics.
*Face palm*. This raises an interesting point about generative AI. We cannot take everything that is provided back to us as gospel. Just because it doesn’t make “human errors” doesn’t mean it is at all perfect.
Striking the right note?
In fact, before sharing the final lyrics, my biggest takeaway from this exercise is that ChatGPT supports and complements creativity, but it isn’t a substitute for it. The right prompts produce more valuable content, but that needs refining to produce something that is worth sharing and shouting about. We’re not in a position where we can click our fingers and AI will create the next ‘Waterloo’ (or Cannes Lion-winning campaign).
It’s also important to remember the legal aspect of using generative AI. With regulation still being formalised, it is necessary to take a cautionary approach. Anyone creating content with the support of generative AI like ChatGPT should consider what information they are putting in, whether you have the right to use what comes out, and continue to stay up-to-date on the latest developments.
After taking on board every instruction (some better than others), the results were in. But, before handing off the lyrics to the rest of the team, I wanted to make sure we were giving us the best chance of producing Eurovision magic.
I repeated the same prompt process to create a second set of lyrics. Then I provided both songs to ChatGPT, telling it its job was to “select a winning entry for Eurovision 2023 based on its knowledge of Eurovision and previous winners.”
Ladies and gentlemen, introducing the lyrics ChatGPT believes are set to take Eurovision 2023 by storm…Celebrate the night. It’s not quite “nil point”, but we’re not expecting the great songwriters of our time to be out of a job anytime soon.
Keep an eye out for the next stage of AxiCom’s Eurovision journey – the music production…