Note to readers: This article contains content created by LLM ChatGPT. Specifically, the story portion is the product of my prompt and ChatGPT’s response. The rest of the article is my unadulterated creation, thunk up out of my own brain. Please, feel free to contact me with questions related to the content of this article or the series of articles.

In a rapidly changing technological landscape, artificial intelligence poses a threat for many, an opportunity for others. I propose an adventure of the mildly philosophical kind. How about we dive into a story generated by ChatGPT, give it a read, and explore the value, lack of value, dangers and potential reliability of this controversial technology? We might expose it as a threat. Or we might find it an innocuous tool that we can choose to use, like a monkey wrench, to bludgeon the brains of our intended victim.

In the upcoming weeks, I will share with you a short story written by ChatGPT. Each week will include a separate scene that I will analyze for its value as a fictional narrative. I do not intend for this critique to replace a sophisticated literary review of this “writer” of short stories. My intent is to evaluate how well ChatGPT works as a storyteller, raconteur, or teller of tales. I hope to make some useful comparisons, to learn a bit more about the craft and skill of writing fiction.

Are you up for the journey? Let’s get after it, then.

Do we need to fear the power of Large Language Model Artificial Intelligence to tell us a story? 

As a starting point, we must ask the question that this analysis hopes to shed light on: Do we need to fear?

Initially, yes, we should fear AI. But not necessarily because of its capacity to supplant the human creative mind. Our fear should be that AI will make us lazy as creatives, and superficial in our expectations of quality fiction. We can expand this concern to other human endeavors, as well. AI is akin to a television from the 1950s. Turn that baby on and watch the world go by.

This exercise is designed to entertain us, while also giving us the opportunity to prepare for a brave new world. (Uh oh. Did I just make a dystopian worldview reference?)

The prompt starts the narrative.

ChatGPT Presents “Shadows in the Smoke”

Scene One

The city was a labyrinth of secrets, the kind that thrived in the dimly lit alleys and the corners of smoke-filled bars. John Harland knew every twist and turn of it, every darkened door and whispered lie. He had to; it was his job.

As a private eye, Harland had seen it all—cheating spouses, stolen goods, missing persons who didn’t want to be found. But tonight, as he sat behind the wheel of his ‘46 Buick, staring at the worn photograph of his wife, Eleanor, he couldn’t shake the gnawing feeling that this case was different.

He flicked his cigarette into the street, watching the ember die out in the rain-soaked gutter. The client had been vague, just the way he liked them, but there had been something in her voice—a tremor, a hesitation—that made him uneasy. A man had to trust his gut in this business, and Harland’s gut was telling him that this was a job he shouldn’t take.

But the bills didn’t pay themselves, and Eleanor was waiting for him at home. She always waited, patient and loyal, a constant in his unpredictable life. Harland had never given her reason to doubt him, but the nature of his work weighed on him. He knew better than anyone that trust could be as fragile as the delicate tendrils of smoke that curled around his head.

This case, he decided, would be his last. One final job, one final payday, and he’d hang up his fedora for good. For Eleanor. For them.

He picked up the phone, dialing the number scrawled on the back of a matchbook. The line rang once, twice, before a woman’s voice answered. Soft, almost too soft to hear.

“Is this Mr. Harland?”

“It is,” he replied, his voice steady. “You still need that help?”

There was a pause, and for a moment, he thought she might hang up. But then she spoke again, her voice thick with something he couldn’t quite place.

“Yes,” she whispered. “But it’s more complicated than I thought.”

ChatGPT: What do you think? We can continue from here, adjust the direction, or develop the characters and plot further!

What did we learn about ChatGPT’s ability to tell us a tale?

This story opens with strong imagery. A “labyrinth of secrets,” “dimly lit alleys,” “corners of smoke-filled bars,” along with “every darkened door and whispered lie.” You get the picture. The story quickly establishes the character of John Harland as a jaded, 50s era private eye, albeit with a twist. By selectively using key phrases, the “author” effectively establishes the image of Harland. A commonly used technique comparable to picking low-hanging fruit in an easy read. Surprisingly, clichés don’t harm the narrative’s authenticity here. At least for me, during the first read. P.I.s specialize in cases like cheating spouses, stolen goods, and missing persons who don’t want to be found. of any time period. We expect the list and it works.

Two bits I really enjoyed were the details “behind the wheel of his ‘46 Buick,” and “staring at the worn photograph of his wife, Eleanor.” ChatGPT held fast to the prompt by including this last, uncommon element in our noir mystery.

The opening elements did their job, although I found them somewhat stale. The client’s vagueness, Harland’s gut feeling and his commitment to this “final job” all come across as flat and uninspiring. Even the idea that Harland will “hang up his fedora for good. For them” tastes like a flat soda. Maybe one stolen from the refrigerator of a poorly written romance?

The dialogue is also uninspired, echoing the dialogue of a thousand detective stories over a hundred years penned by dozens upon dozens of writers scrambling for the attention of a multitude of pulp fiction magazine editors around the globe. (You may now breathe.) This long-winded detail segues us to the last item of this post. Where does ChatGPT get this stuff? How can a LLM write a fictional narrative that sounds like Dashiell Hammett on his first day at the typewriter?

If you don’t know the answer already, LLM AI for written material gathers every bit and byte of (hopefully) pertinent data from the storage of millions of servers around the world and assigns value to the particular words, phrases and syntax based on their usage and context. It then attempts to reassemble a facsimile of the requested type of content using this data within the parameters of the end user prompt. Want to explore previous writings in a specific genre or any topic? Simply ask AI to create content for you. The construction will be an amalgamation of everything that’s gone before. Assuming the information is digitized and legally accessible.

Let’s pause before we get all negative and judgmental about this thought-stealing technology, and consider where our own creative ideas originate. We listen, we read, we assimilate massive amounts of data within our area of expertise, much of it the work of those who have gone before us. And then we regurgitate a version on the topic, with varied measures of creativity and equally varied levels of appeal. In some ways, we are a more sophisticated computer algorithm than the LLM AI we fear.

The difference lies within the filtering system used by computers versus that used by humans. Computers sort based on the relevance of discreet elements according to the programming. Things like words, phrases, sentences, paragraphs and pages. They must choose based on mathematical probabilities—how many times in this context was that particular sequence of words used. Given the massive quantity of written work available in the digital cloud, LLMs can make fairly precise recreations of human content. Their approach is so minute that it appears fresh and novel.

People work differently. When humans create out of the large quantity of assimilated knowledge that establishes our baseline. We use our mental and emotional filters to guide our creative efforts. This process may be analogous to that used by LLM AI. It depends in large degree on your worldview regarding the origins of humanity. I would argue that human cognitive and emotional processes are so complex the intricate workings of machine code can not mimic them.

In future posts, I hope to explore the components of storytelling as we evaluate ChatGPT’s construction. Since this article ran long and is about two days overdue, the rest will wait. Let me know your thoughts.