I ‘wrote’ a song for children using ChatGPT. Is it any good? (It raised lots of issues)

I recently used ChatGTP for the first time. I’d had a conversation with a colleague about using AI to analyse interview transcriptions which led me to this video on ‘How to research and write using generative AI tools’. I’m very much a jump-in-and-have-a-go sort of person so after about ten minutes of learning how to write an effective prompt for AI, I thought I’d give it a go.

Rather than let ChatGPT loose on my research data just yet, I thought I’d test its creative abilities. Some context. I am a songwriter. I have written well over 200 songs for children and families which I record and perform with my band Johnny & the Raindrops. My PhD and book (Spinning the Child: Constructions of childhood through records, radio and television) are about music for children. I analysed hundreds of songs that were made and broadcast for children. I interviewed dozens of children’s songwriters. I also teach songwriting – to children in schools, students at university and everyone in between.

So, this is what I told ChatGPT to do:

Write a song for children about nature.

It must have verses with 4 lines.

The lines must rhyme.

It must have a 4-line chorus with lots of repetition.

It can be funny.

It can mention flora and fauna.

Here is the first part of its first five-verse-and-chorus draft:

Verse1

In the forest where the tall trees grow,

There's a place where all the animals go,

Squirrels and birds play hide and seek,

Every day of the week.

Chorus

Nature, nature, everywhere we see,

Nature, nature, wild and free!

Nature, nature, fun for you and me,

Nature, nature, wild and free!

On first glance, it’s not bad. If it were a group of 10-year-olds, I’d say ‘Well done. You’ve done what you were asked’. All of the basics were ticked off.

I won’t bore you with the other four verses. Just to say that they weren’t very funny. Humour is definitely a trope in music for children - I’ve written about it (Humour and Laughter in children's music). It reveals a lot about how adults idealise childhood as smiley, bright, upbeat, happy, etc. or occasionally, how they use satire, parody and other devices to subvert normalised notions of childhood with humour.

So, I typed again:

Try to make it funny.

Within seconds, I got (hold on to your sides!):

Verse1

In the forest where the tall trees grow

There's a place where all the animals go

Squirrels in pajamas play hide and seek

While the chipmunks giggle and squeak.

Chorus

Nature, nature, what a funny sight,

Nature, nature, day and night!

Nature, nature, full of fun and light,

Nature, nature, what a funny sight!

Verse2

The buzzing bees wear tiny shoes

Ants sing the marching blues

Rabbits hop and play the kazoo

In nature's wacky zoo!

… plus three more equally hilarious (!) verses that mentioned frogs telling jokes or used fun-related words such as ‘antics’ and ‘boast’. It was like an uncritical browse through a synonyms list. But like my students, ChatGPT was learning … and so was I.

 It had rhymed ‘dusk and dawn’ with ‘break of dawn’. So I requested it ‘not to rhyme a word with itself’. It couldn’t manage that.Not a problem. I can deal with that myself.

It used US spelling and the occasional US word (’ladybugs’, ‘critters’). Again, I could change that.

I tried singing the words to the metre they seemed to suggest to me. A few lines didn’t scan well. The syllables didn’t fall naturally. I had to push and pull them and ‘get the crowbar out’.

So, I asked:

Try to make it scan so that the syllables that are normally stressed when spoken line up

I needed to be clearer. I could have asked it to put the stressed syllables on the bar lines, the ‘1’ when you’re counting ‘1, 2, 3, 4’. But it did its best.

Buzzing bees wear tiny shoes of blue

Ants march in line with a marching crew

Rabbits hop and play their kazoos

In the woods, there's no time to snooze!

Flowers giggle as they bloom so bright

Butterflies in hats take off in flight

Frogs tell jokes through all the night

Nature's world is pure delight!

Still not quite there. I’ve found that if you want your song to be super sing-able by groups of children (and adults too!), it helps if the syllables that are naturally stressed when you speak them fall on the first beat of the bar. I didn’t ask ChatGPT to do that so there were a couple of hanging syllables. For example, in ChatGPT’s version above, ‘ants’ falls in front of the bar line while ‘march’ is on the first beat.

I’m keen on retaining those expected stresses and pronunciations as much as possible. Those up-front-of-the-bar-line syllables are not that much of a problem when I sing them myself. I can make the adjustments either in my head as I go or by underlining the stressed syllables stress on the page.

When groups of children try to sing the song, it helps if the metre is regular, but not to the point it gets too repetitive. Some of the problem was that ChatGPT had created lines of different length of between 8 and 11 syllables. To prescribe some regularity, I gave it what would be my final instruction:

Make each line have either 8 or 9 syllables.

Helpfully, it put the number of syllables in brackets at the end of each line, just like I would ask my students to do to help them think about metre and phrasing.

Here is ChatGPT’s final version.

Verse1

In the forest where tall trees grow (8)

Critters gather where cool winds blow (8)

Squirrels in pajamas hide and seek (8)

Chipmunks giggle, cheeks stuffed with treats (8)

Chorus

Nature, nature, funny sight (8)

Nature, nature, day and night (8)

Nature, nature, fun and light (8)

Nature, nature, feels just right (8)

Verse2

Buzzing bees wear tiny blue shoes (8)

Ants march in line, a marching crew (8)

Rabbits hop and play their kazoos (8)

In the woods, no time to snooze (8)

Chorus

Verse3

Flowers giggle as they bloom bright (8)

Butterflies in hats take flight (8)

Frogs tell jokes through all the night (8)

Nature's world is pure delight (8)

Chorus

Verse4

In the meadow, deer play tag, run (8)

Ladybugs dance in the sun (8)

Fish swim by, waving flags for fun (8)

Nature’s joy for everyone (8)

Chorus

Verse5

The sun wakes with a great big yawn (8)

The moon and stars put pajamas on (9)

Nature’s antics from dusk to dawn (8)

Make us laugh until the dawn (8)

Chorus

Clearly, I could have worked with ChatGPT for longer, given it more prompts, stamped more of me on it. But that was not the aim here. I deliberately kept as much of the AI version as possible.

I made some minor changes (in red) to create my own final version.

Verse 1

In the forest where tall trees grow

Animals [prev. Critters] gather where cool winds blow

Squirrels in slippers [previously pajamas] hide and seek

Chipmunks laugh, cheeks stuffed with treats

Verse 2

Buzzing bees wear tiny blue shoes

Ants [deleted ‘march’] in a line, a marching crew

Rabbits hop and play their kazoos

In the woods, there’s no time to snooze

Chorus

Nature, nature, in the day and night

Nature, nature, it’s a funny sight [swapped lines 1 and 2]

Nature, nature, likes to chew and bite [previously 'fun and light']

Wooo Wooo. Everything’s alright! [previously 'Nature, Nature, feels just right']

Verse 3

Flowers giggle as they bloom bright

Butterflies in hats take flight

Frogs tell jokes all through [prev. through all] the night

Nature's world is pure delight

Chorus

Deleted verse 4

Verse 5

The sun awakes with a great big yawn

The moon and stars put pyjamas on

Nature plays [previously 'Nature’s antics'] from dusk to dawn

Wakes up laughing in the morn [previously 'Makes us laugh until the dawn']

Chorus

Chorus

I obeyed the 'rule of three’ in the chorus - ChatGPT had four repeats of ‘Nature Nature’.

I flipped the order of a couple of lines. I rearranged it so that there were two verses before the first chorus and two choruses at the end. I dropped what I thought was the weakest verse – four verses is enough for any short song - and I added a catchy ‘Woo Hoo!’ for good luck.  The whole process from my first prompt to the ‘finished’ lyric was less than five minutes.

Then I set the lyrics to music. To me, words have a musicality. They suggest rhythms and paterns. With the chorus words as my cue, I went for a mid-tempo Bo Diddley hambone rhythm with a four on the floor kick drum and simple 1, 4, 5 chords the order of which changes slightly between verses and chorus.

A music producer friend told me about his experiments with using AI to create the music, not just the melodies and chords, but the whole arrangement and credible sounding instrumental parts. Like my general feeling with ChatGPT as a lyric writer, he described his AI-generated music as superficially impressive but clumsy, clunky and just ‘wrong’ in places. Our efforts were the textual and aural equivalents of those AI-generated people with three arms and six fingers.

Working with AI in this way raises issues about authorship and originality. Who exactly wrote the song? Who owns the copyright? Who gets paid when it gets streamed and broadcast? I’ve written about similar issues in a journal paper I wrote about mashups – recordings made entirely of other recordings.

There are also ethical and moral issues to consider. Should songwriters who want to retain a semblance of credibility be working in this way? Why not? And what about the relations of power? Is ChatGPT working for me or with me?

In a few years’ time, when AI is seamlessly integrated into not only creative but also academic practice, when it’s stopped being a novelty and become invisible and ubiquitous, this song and my reflections on the process and product will seem quaint and archaic, like someone assessing the value and impact of the gramophone or the printing press.

So have a listen to the song and watch the video on YouTube. I created the images of the animals using the AI function on Canva.

Would you know that most of it had been written by AI if I hadn’t told you?

And is the song any good? What criteria are you using to judge it? What does ‘good’ mean in this context?

Play it to your kids, the audience for whom the song was created.

Do they think it’s good?

So many questions!