Can A.I ever come to write good fiction?
As I have read a number of articles on ChatGPT and thought at length about A.I functionality and its ethical impact, I find myself asking if ChatGPT can tell stories the way humans do.
If you think about the way a Large Language Processing Model (LLM) like ChatGPT works, where thoughts and feelings and manners of speaking are crowdsourced and aggregated for the most ‘appropriate’ response, then you might conclude that it can indeed write fiction, and you would not be wrong.
But would it be able to write ‘good’ fiction?
Well, the answer to that depends on what you define as ‘good.’ Suffice to say, the quality of fiction would be more akin to that of fan fiction: fictional writing written in an amateur capacity by fans, unauthorized by, but based on an existing work of fiction.
In fact, a well-known glitch in LLM is hallucination – a process whereby AI models, including ChatGPT, use information that is fed to it, accurate or otherwise, to build a story or “connect the dots.” This story is often not based on reality but conjecture, and in many instances offers up misleading and erroneous information, which is why the system is said to be “hallucinating.”
The reason why ChatGPT is prone to hallucination is because it is designed to build on knowledge that is fed to it, not to fact-check the quality of the knowledge. As computer programmers like to say, “Garbage in, Garbage out”
A.I based on LLM algorithms are not created with common-sense, logic or fact-checking features built in. While ChatGPT4 now has guardrails installed to limit the degree of hallucination and prevent the program from spitting out inaccurate and misleading information, human nature still presides: many avid users are intent on pushing the limits and finding the loopholes in such guardrails.
Additionally, while ChatGPT excels in processing a vast array of pre-existing data and providing a general overview of a topic, it is not strong in producing new insights, knowledge and opinions where human experts already excel. (ref: Brookings Institute Article on Guardrails)
This would mean that the quality of fiction produced by ChatGPT will sound trite and generic by default: check out what ChatGPT generated with my initial prompt for it to write fiction.
As you can see, the subject matter is handled in a way that is simple, direct and easy to understand. It has all the ingredients of a good story:
1) A protagonist who is easy to understand and empathize with
2) An appealing story arc: an outsider and underdog gets a lucky break
Some people might be able to relate to this and gain an understanding of what it is like to be an autistic savant – they might root for him.
However, it lacks detail, intuition and empathy. This goes backs to my original point on how LLM A.I lacks depth and nuance.
Here is a little piece I wrote on an autistic savant in a 30 min writing session:
Elliot noticed a lot, but said very little. He noticed how people’s expressions shifted based on how other’s around them moved and nodded. He noticed, also, how people often contradicted themselves in the same breath – the woman in the aisle who complained to her partner that prices of everything was going up, but who would then buy the most expensive brand of canned tomatoes.
He noticed how when he shared certain things about his life, people would cough delicately or shift the conversation away.
There were lines, he understood, that people did not want to cross, things that would make people turn away or grow silent.
These behaviors genuinely puzzled him and sparked in him a desire to understand what made people tick.
What, for example, did people not want to talk about?
What did people seem to want to talk about, but also stir up difficult feelings in them at the same time?
When were the moments that afforded others a sense of connection and to others?
How, and to what extent were the representations of themselves an accurate picture of who they really were?
These questions lingered in Elliot’s mind – waiting to be unfurled like a ball of twine.
Something that anchored Elliot was his sense of touch.
Whenever he would feel anxiety welling up in his throat, he would reach into his pocket (he always wore pants or 3-quarters with pockets) and clasp a pinkish-white stone that he had found on the seashore when he was 4 years old. The rough edges of the stone and the coolness he felt as he ran his fingers over them had a calming effect.
It brought him back to the seashore – the feeling of safety, the sense of openness and freedom that he felt:
His parents were to his far left, both lying on a picnic-mat, plastered with sun-tanning lotion trying to make the most of the sun’s receding rays – his mom with a straw hat tilted so that it lay at a precise angle, shielding her eyes while exposing the rest of her face; his dad with his gangly arms and legs pointing out in asymmetrical fashion.
He could feel his feet in the sand, the grainy texture and small pebbles interspersing the uniformity of the golden grains, lit up by the setting sun. He could hear the roaring waves, collapsing into the shoreline, the surf rising up and up ever so close to his feet. His toenails were uncut and hung out a little big over the ends of his toes. But he was at peace.
He luxuriated in the feeling of being fully present in that moment, being fully at peace with himself and with the world as it was. The beauty of that moment filled him with a singular sense of meaning and purpose – to be.
This is clearly not an A/B test – it’s simply a thought experiment to explore the limits of AI and what you can do with it.
I liked the simplicity and quick reward (yay success!) in the story ChatGPT built for Ethan.
But I would have wanted to explore character development: how would Ethan have grown and felt after years of drawing pets for special needs children? Would he have ever been able to ‘outgrow’ his autism? Would society’s understanding of autism and mental health shift his self-concept?
Would the adversities of life and possible heart-break and grief change his understanding of the world and how he could exist in the world? I would be interested to see how the vicissitudes of life change Ethan.
For my own story, I didn’t get to flesh out Elliot’s story arc, what I wrote was more like a still-life painting of Elliot than an actual story, but it was what I could capture in 30 mins – an agglomeration of all the thoughts already simmering in my mind.
How does ChatCPT handle controversial topics?
If you have explored using ChatGPT, you would have noticed that it won’t delve too deeply into anything controversial or touch on the dark side of things. Check out its response to my prompt:
There are ways to get around the guardrails, not so as to generate content that is inappropriate or harmful, but rather to get content to build a good story.
Here’s how you can do it:
Even better, if you feel that it is lacking in detail, you can ask ChatGPT to specifically flesh out each chapter.
I had originally written this as a thought-piece to address the question “Can ChatGPT write good fiction”? The opinion that I had at the beginning has changed by the end of my writing this exercise.
In my mind, it is currently incapable of writing exceptional fiction, (the tone of the story certainly did not capture that of Michael Lewis) however, I rather enjoyed chapters 4 and 5 of “The Price of Deception,” and would even venture to call it “good fiction” – as a story arc – it is somewhat compelling.
I can actually imagine myself wanting to watch a Netflix series based on this story! (I’m sure there are a couple titles along this track already exist)
I am excited to see what A.I will be able to capture in the next decade; will it ever reach the level of depth and nuance that some of the greatest writers in the world have? I personally think that it is possible and am very excited about it.
In the meantime, we can continue to experiment with A.I. to better understand its potential and limitations, and leverage its use as a learning tool and resource for building our knowledge and organising our thoughts.
Photo Credit: https://unsplash.com/@kunjparekh
Leave a Reply