Chatbots and Narratives

There has been a lot of discussion about ChatGPT and its ability to compose text, including essays, articles, and website content.

At Intuition, we recently invited ChatGPT to help us create content to promote 'Winter Wonders', one of our creative writing workshops. The result was playful and full of exclamation marks! It was ideal and suited our target market.

It then raised the question of whether, if ChatGPT could do this, it could write a piece of creative work, like a narrative. We then asked ChatGPT to compose a one-page horror story.

The result was reasonably coherent and included simple traits expected of a horror story. It featured a boy named Tom who takes refuge in an abandoned house and finds an old mirror. Looking into the mirror, he sees not just his reflection, but "eyes that were not his." In a mad panic, Tom tries to escape but fails and is never seen again.

Curious, we asked ChatGPT to write us another one-page horror story, and the result was very similar. This time a young woman named Clara finds herself caught in the rain, and seeking shelter runs towards a "mansion that had been left to rot." When she reaches the mansion, her curiosity gets the better of her, and she decides to explore. As she explores, she finds an old mirror and "steps back in horror" when she sees "shadowy figures" lurking behind her reflection. Clara then attempts to run but is dragged back and is never seen again.

What about the other chatbots out there, we wondered, what would they do? We then fed the same prompt to Meta AI with Llama 3.1, and the result was similar again: an old mansion and a curious character who goes exploring. This time, however, a room of antique dolls beckons instead of a mirror. But the result is the same. The dolls, like the mirrors, draw the main character in. When the main character tries to escape, they can't and are never seen again.

What did we conclude? Well, that chatbots can write reasonably coherent narratives that include simple traits expected of a horror story. How do these pieces rank compared to human-written pieces? If a marker were asked to assess these pieces and wasn't told a chatbot wrote them, they would assume the author was a reasonable student, perhaps ready to sit the NAPLAN.

Another conclusion is a warning to aspiring GATE students and writers suffering from writer’s block! Do not think you can use chatbots to trick markers or editors. The pieces produced by AI for this experiment were all too predictable and overused the simple idea of a main character being drawn in by something terrible, mirrors on two occasions and dolls on a third. If any aspiring GATE student memorised and reused such work, they would be quickly marked down. This is because professional markers would see the work for what it is, competent but ultimately a regurgitated piece of cliché writing.