There have been quite a few social media posts recently discussing artificial intelligence (AI) in the creative process, and I think the topic needs exploring in more detail. My first brush with AI writing was actually through a colleague at work. I lead a team writing proposals and pitches for R&D funding, including a number of projects involving AI. I have started writing a work-related blog for our website, and one of my team pointed out that there are software packages you can use to write blog articles on your behalf. We had a play around and, whilst the results are very impressive, I wanted my blog to have a particular feel, which the AI simply couldn’t provide. I then saw the topic of AI in children’s writing start to trend and saw the obvious parallels. My instant response was ‘there’s no way AI could write a really good children’s book.’ But, like any good scientist, I figured we needed to put this to the test.
Using chat.openai.com, a platform used by others to write children’s books, I started experimenting. I chose as my target text Oi Frog by Kes Gray. I figured this would be a really challenging target for the AI - it involves lots of rhymes and a clever humorous thread, and doesn’t follow a traditional children’s story narrative.
All of these AI platforms need an input - a set of instructions to guide their creative process. I experimented with lots of different variations, but the following string seemed to work just as well as any others:
“Write me a funny 400 word picture book, broken down into 12 spreads, about a frog that tells other animals that they can only sit on objects that rhyme with their names.”
And here’s what the chatbot managed to come up with:
Spread 1:
Introduce the main character, a frog named Fred, who is sitting on a lily pad.
Spread 2:
Fred is joined by a bird, who tries to sit on a branch.
Spread 3:
Fred tells the bird that it can only sit on objects that rhyme with its name. The bird looks confused.
Spread 4:
Fred suggests that the bird sits on a word instead, and the bird happily hops onto a word.
Spread 5:
Next, a rabbit hops over and tries to sit on a carrot.
Spread 6:
Fred tells the rabbit that it can only sit on objects that rhyme with its name. The rabbit looks disappointed.
Spread 7:
Fred suggests that the rabbit sits on a gab instead, and the rabbit hops onto a gab.
Spread 8: A duck waddles over and tries to sit on a pond. Spread 9: Fred tells the duck that it can only sit on objects that rhyme with its name. The duck looks confused. Spread 10: Fred suggests that the duck sits on a muck instead, and the duck happily waddles onto a muck. Spread 11: A bear walks over and tries to sit on a log. Spread 12: Fred tells the bear that it can only sit on objects that rhyme with its name. The bear looks disappointed. Fred suggests that the bear sits on a bog instead, and the bear happily hops onto a bog. The end. In some respects, this is a resounding success. With a simple input string, a computer has managed to create a completely original story, something that would have been unheard of 10 years ago. However, I think you’ll be hard-pressed to find anyone who would say this was a good story, even setting aside the obvious fact that bear doesn’t rhyme with bog.
So why is it that this, and other examples of AI-generated children’s stories, seem to fall so far short of good literature?
A big reason is that really good children’s stories, especially humorous ones, rely on creativity. AI, on the other hand, is bound by the limits of its training dataset, and will tend to produce results based on what is most commonly done by others. This is the very opposite of creativity. Thanks to our ability to use every experience as inspiration, humans can create genuinely radical ideas.
The other issue is that AI can’t judge if what is created is ‘good’; it can only say that it is created something that fits its rules. In theory, with a big enough dataset, AI could produce something truly original and good, but the problem is it would never know. Without a human checking each output and feeding back to the AI to guide the next result, it will continue blindly.
As such, I don’t think creative writers have anything to fear from AI of the moment, and probably not for a long time to come.
But…
I don’t think the same applies to computer-generated art. The topic deserves a whole post on its own, but I think AI-generated illustration pose a far greater risk to the creative industries than AI-generated children’s stories. As an example, the cover image for this page was created using Midjourney.
Then again…
This is Midjourney's attempt to create a cartoon frog talking to a bird. I think Jim Field can sleep easy for a little while longer.
What are your views? Is AI a menace or just a distraction? Send me your thoughts as a humorous rhyme, just so I know you aren’t an AI chatbot…